Compare commits

..
Author SHA1 Message Date
阿白叔 9eea9a4c20 fix: explain blocked export evidence and show batch results before refresh 2026-11-12 17:36:34 +08:00
阿白叔 466f400514 fix: scope release proofs and bind PCB opening consent 2026-11-12 16:02:53 +08:00
阿白叔 42d0cebc54 fix: address defects found in frozen native acceptance 2026-11-12 11:33:44 +08:00
阿白叔 b86e4e315a test: simulate all UI buttons and handle log folder errors 2026-11-11 16:22:14 +08:00
阿白叔 42e8719026 fix: keep residual release controls aligned with live queue state 2026-11-11 14:56:08 +08:00
阿白叔 fdd07224aa docs: record full review fixes and final 1775-test validation 2026-11-09 14:49:45 +08:00
阿白叔 a045683c8a fix: publish management receipt before orphan-exitable terminal state 2026-11-09 14:44:09 +08:00
阿白叔 74a45eb5cf Merge branch 'codex/review-residual-release' into codex/runtime-lifecycle-safety 2026-11-09 14:34:32 +08:00
阿白叔 a7df105616 fix: recreate batch resume action after release summary retirement 2026-11-09 14:34:28 +08:00
阿白叔 78ff118f82 fix: preserve accepted native calls and explain retained queue waits 2026-11-09 14:33:44 +08:00
阿白叔 d0a5ce5d17 Merge branch 'codex/review-residual-release' into codex/runtime-lifecycle-safety 2026-11-09 14:10:38 +08:00
阿白叔 363a279157 Merge branch 'codex/review-ui-integration' into codex/review-residual-release
# Conflicts:
#	src/cbb_refresh_verify/ui/main_window.py
2026-11-09 14:10:30 +08:00
阿白叔 306aa5c544 fix: centralize native dispatch and itemize retained session release 2026-11-09 14:03:07 +08:00
阿白叔 0fca6abb4e Merge branch 'codex/review-native-backend' into codex/review-ui-integration 2026-11-09 14:02:14 +08:00
阿白叔 342a7266c7 fix: retire result dialogs and clarify batch lifecycle 2026-11-09 13:51:32 +08:00
阿白叔 e8a1714d95 fix: clarify backend status freshness and scoped copy evidence 2026-11-09 13:50:04 +08:00
阿白叔 60101b7195 Merge branch 'codex/review-residual-release' into codex/runtime-lifecycle-safety
# Conflicts:
#	src/cbb_refresh_verify/ui/residual_release_dialog.py
2026-11-09 13:47:18 +08:00
阿白叔 14c9a74bbf fix: show confirmed scope beside actual residual release results 2026-11-09 13:45:53 +08:00
阿白叔 42898d33d2 chore: pin reviewed Python line endings and record stack validation 2026-11-09 10:48:20 +08:00
阿白叔 b1c36e3ca6 Merge branch 'codex/review-residual-release' into codex/runtime-lifecycle-safety
# Conflicts:
#	src/cbb_refresh_verify/residual_sessions.py
2026-11-09 10:41:24 +08:00
阿白叔 651b539d0e fix: isolate unreadable residual sessions and bound release evidence 2026-11-09 10:39:16 +08:00
阿白叔 35daa77118 Merge branch 'codex/review-ui-integration' into codex/review-residual-release 2026-11-09 10:26:20 +08:00
阿白叔 e73f9c6edc Merge branch 'codex/review-native-backend' into codex/review-ui-integration 2026-11-09 10:26:13 +08:00
阿白叔 b7baa24b5e fix: identify foreign CBB checkpoints in release evidence 2026-11-09 10:26:09 +08:00
阿白叔 ca39fb3517 fix: clear verified queue blockers and exit idle orphan workers 2026-11-09 10:11:36 +08:00
阿白叔 96e179ba65 docs: record corrected icon candidate verification 2026-11-06 17:29:18 +08:00
阿白叔 44bb734de8 fix: preserve the current application icon in lifecycle candidate 2026-11-06 17:24:58 +08:00
阿白叔 1494dc260c docs: link draft runtime lifecycle review 2026-11-06 17:07:20 +08:00
阿白叔 8e946c6e7a docs: record runtime lifecycle validation and candidate limits 2026-11-06 16:59:58 +08:00
阿白叔 5873b7146b chore: prepare runtime lifecycle candidate 0.2.4b2 2026-11-06 16:44:12 +08:00
阿白叔 051498ba57 fix: require a durable boundary before native add submission 2026-11-06 16:41:36 +08:00
阿白叔 d9a79b6535 fix: reconcile reused identities and protect every native dispatch 2026-11-06 16:21:08 +08:00
阿白叔 cb11624b68 fix: observe retained sessions without blocking independent libraries 2026-11-06 15:42:40 +08:00
阿白叔 6278091f21 fix: supervise management calls across cancellation and parent loss 2026-11-06 15:42:36 +08:00
阿白叔 8d3eab1164 fix: preserve native calls on stop and keep status reads observational 2026-11-06 15:42:33 +08:00
阿白叔 97a3de20c9 feat: persist and reconcile owned native sessions 2026-11-06 15:42:29 +08:00
阿白叔 7bd7cd3966 feat: distinguish save close and resource release facts 2026-11-06 14:33:10 +08:00
阿白叔 3e7ae2b76f docs: plan native lifecycle safety and read-only status boundaries 2026-11-06 14:25:09 +08:00
阿白叔 5eb4b2d7ed fix: persist release outcomes and guard active native backup sessions 2026-11-06 12:08:54 +08:00
阿白叔 d40192ee58 Merge branch 'codex/review-ui-integration' into codex/review-residual-release 2026-11-06 11:41:05 +08:00
阿白叔 3c1998f2ef Merge branch 'codex/review-native-backend' into codex/review-ui-integration 2026-11-06 11:41:03 +08:00
阿白叔 807aa26baf test: guard actual reads of unrelated CBB content 2026-11-06 11:41:00 +08:00
阿白叔 71029b19a2 Merge branch 'codex/review-ui-integration' into codex/review-residual-release 2026-11-06 11:36:21 +08:00
阿白叔 8c2c60fab1 fix: distinguish native outcomes from report and item recovery status 2026-11-06 11:36:19 +08:00
阿白叔 5b22787418 Merge branch 'codex/review-native-backend' into codex/review-ui-integration 2026-11-06 11:34:27 +08:00
阿白叔 679f33b8cc fix: keep prelaunch release-probe failures from stopping independent work 2026-11-06 11:34:25 +08:00
阿白叔 746e5b5145 Merge branch 'codex/review-native-backend' into codex/review-ui-integration 2026-11-06 11:21:21 +08:00
阿白叔 8eefecd9f0 fix: reconcile historical leases and record scoped native changes 2026-11-06 11:21:15 +08:00
阿白叔 d8156c57f0 Merge remote-tracking branch 'origin/main' into codex/review-native-backend 2026-11-06 11:14:58 +08:00
abaishu0730 959399b974 Merge pull request '审查拆分 1/4:只读占用探测与可取消任务队列' (#9) from codex/review-infrastructure into main 2026-11-06 10:25:10 +08:00
abaishu0730 46dae81190 Merge pull request 'docs: CBB Verify 全量扫描与阻塞分析(关联 #4)' (#5) from docs/scan-20261104-blockers into main 2026-11-06 10:20:23 +08:00
阿白叔 dc536da555 docs: record published review chain and completed verification 2026-11-05 23:33:46 +08:00
阿白叔 173dee1980 feat: add confirmed client release with durable consent and server protection 2026-11-05 23:31:36 +08:00
阿白叔 282be7be7f feat: integrate independent background checks and native management UI 2026-11-05 23:25:25 +08:00
阿白叔 da67de026f feat: integrate native diagnostics and coordinated CBB transactions 2026-11-05 23:18:00 +08:00
阿白叔 9e230e1a46 feat: add bounded read-only owner probes and cooperative native queue 2026-11-05 23:04:35 +08:00
abaishu0730 a56fd17b60 Merge pull request 'fix: release private desktop resources on launch failure' (#8) from codex/desktop-leak-main into main 2026-11-05 15:23:08 +08:00
阿白叔 766d67cb57 test: enforce desktop cleanup interfaces and guard reader setup 2026-11-05 14:59:19 +08:00
阿白叔 311d0dd9a4 fix: release private desktop on worker launch failure 2026-11-05 14:47:03 +08:00
abaishu0730 3dbfca0f44 docs: record full project scan and blocker analysis (#4) 2026-11-04 15:20:37 +08:00
193 changed files with 20681 additions and 3295 deletions

No files matched your search

+3
View File
@@ -1,3 +1,6 @@
# Keep the verified Windows source and binary assets byte-for-byte.
# Do not normalize the r4 baseline's line endings during add/checkout.
* -text
# Normalize only these two reviewed Python files at the chain tip.
src/cbb_refresh_verify/application.py text eol=lf
src/cbb_refresh_verify/ui/main_window.py text eol=lf
+24 -9
View File
@@ -1,15 +1,30 @@
# CBB 中心库上传后校验工具
当前临时分支的备份策略(2026-11-03):应用不再自动备份中心库或 CBB 工程。新增、替换、移除、校验、ICDB 维护和恢复在执行前提示手动备份;请自行保留可用副本。导入暂存、恢复用隔离验证库和轻量操作证据仍会生成,Mentor 原生 Repair 自身的备份行为保持不变。详见[自动备份检查与调整](docs/releases/2026-11-03-manual-backup.md)。此条优先于下方历史版本的备份说明。
当前补丁:**[v0.2.3-beta.1](docs/releases/v0.2.3-beta.1.md)(Beta 预发布,待现场验收)**,替代 0.2.2-beta.1。数据库所有权或关闭状态不确定时停止批次、保留现场;不自动移走地址文件、接管带锁设计或强杀 PCB 进程树。新增全库 ICDB 维护、同名 CBB 完整备份恢复弹窗及严格的原生复核流程。详见 [v0.2.3-beta.1 发布说明](docs/releases/v0.2.3-beta.1.md)。下方日期记录描述历史版本,以本补丁说明为准。
当前临时分支的备份策略(2026-11-03):应用不再自动备份中心库或 CBB 工程。新增、替换、移除、校验、ICDB 维护和恢复在执行前提示手动备份;请自行保留可用副本。导入暂存、恢复用隔离验证库和轻量操作证据仍会生成,Mentor 原生 Repair 自身的备份行为保持不变。详见[自动备份检查与调整](docs/releases/2026-11-03-manual-backup.md)。此条优先于下方历史版本的备份说明。
当前开发版本:**[0.2.4b2 生命周期候选](docs/releases/2026-11-06-runtime-lifecycle-candidate.md)(开发候选,待现场验收)**。ICDB 批量检查在后台逐项运行,普通校验、管理和恢复可排队;READY 工程在实际执行前再次排除。检查不自动 Repair、Restore 或选择历史版本,也不代表中心库 READY。
停止、超时或应用失联只停止后续原生动作,不结束正在 Save/Close 的调用进程。普通停止等待 5 秒,可能正在写入时最多等待 120 秒后交给后台观察;这不是强制结束时限。保存结果、关闭结果和资源释放分别记录。相关库等待释放,独立库可以排队继续;界面、筛选和日志保持可用。重启只核对未完成会话,不重放管理写操作。legacy 状态读取只观察已打开且匹配库的 LM,不启动 LM 或回答弹窗;无法读取保持 UNKNOWN。
历史释放记录会触发实际占用复核,确认释放后归档继续;会话释放不等于历史数据恢复。释放结果逐项保存;记录保存失败会单独显示,不把已确认释放的会话重新标成占用。可查看后台窗口的原生提示;仅对当前可识别的原生取消按钮提供确认操作。
自动流程不强杀原生会话;释放入口默认正常关闭。另有单独确认的“强制结束客户端”操作,仅限属于当前 Windows 用户和会话、归属及进程身份已确认的 PCB、Library Manager、DxDesigner/viewdraw、iCDB Project Backup 客户端。**ICDB 数据库服务器、后台导入器和归属未知的进程不允许强制结束**,即使命令指向当前库也不例外。客户端中其他已打开工程可能一并退出、未保存更改可能丢失;必须先记录确认名单,结束后重新证明库已释放,不能宣称数据库已修复。详见 [操作规程](SOP.md) 与 [版本说明](docs/releases/v0.2.4-beta.1.md)。下方日期记录为历史行为,以本说明为准。
本目录是独立的 CBB 中心库上传后校验工具,不属于 Part Replacement,也不负责 CBB 工程复制或中心库上传。
针对 `The last server session ... lost the connection to the database`,修复分支增加
`ICDB_REPAIR_REQUIRED` 专项识别、共享符号库路径诊断及停止批次处理。现场原生修复步骤及实现边界见
[iCDB 旧服务器会话失联分析](docs/diagnostics/2026-10-27-icdb-lost-server-session.md)。
当前分支会在目标明确、所有权释放和独占访问证据齐全时尝试一次同版本原生 Repair;操作前请手动备份,共享符号库没有 PRJ 入口时停止并要求人工处理。
当前应用入口仅检查工程打开情况,不自动 Repair。需要修复或回退时由用户在原生版本与备份界面选择,操作前请手动备份;共享符号库故障不得通过修复无关的 CBB 工程替代。
## 当前验证范围与已知限制
Windows 11 / Mentor EE7.9.5 隔离库已完成原生更新、备份恢复、校验到 READY 和释放验证,记录见 [2026-11-12 验证](docs/releases/2026-11-12-scoped-release-validation.md)。历史候选包仅证明对应提交,后续代码合并不等同于现场验收或正式发布。
- A/B 输入的原生 EDIF 导出曾返回 `0xC0000005`,根因尚未定位,相关输入继续按失败处理。
- Windows 10、跨机器共享库和现场服务器会话失联的一致故障副本尚未验收。跨机器释放依赖 SMB 共享模式语义,目前没有跨机器实测证明。
- 自动 ICDB Repair 保持关闭。共享库原生服务器若无法删除自身地址文件,或早停后仍有原生弹窗,需要继续核对;不以进程列表为空或端口不可达代替释放证明。
- 验收文档中的本机路径、原始日志、截图、测试 XML 和冻结包不随 Git 分发;仓库保留说明及摘要,原始材料须在验收机器查看或另行归档。
## 功能边界
@@ -143,7 +158,7 @@ python -m cbb_refresh_verify
- **不注入鼠标或键盘输入,不切换 Windows 桌面,不改写中心库源文件。** 生产路径在本次运行创建的私有桌面内启动中心库目录绑定的 ReusableCircuit PCB,随后用 COM 完成刷新、保存和关闭;只读状态桥在结束后确认 LMC 的 READY。Library Manager 菜单派发仅作为显式诊断路径保留。`tests/test_safety.py` 会在离线测试中阻止真实输入注入 API 重新进入生产代码。
- 所有自动化都跑在本进程创建的**私有 Windows 桌面**上:用户看不到窗口,光标不会被移动。
- 弹窗只认领本次运行的私有桌面/进程,依据已知规则投递 `BM_CLICK`。数据库恢复、会话失联和设计锁接管提示会阻断并保留现场;不会自动选择恢复或接管。未知弹窗也会阻断,Project Integration 进度窗保持观察。动作/错误弹窗和实际控件 ID 写入 `dialogs.jsonl`。
- 超时或失败时停止自动化客户端,保留可能仍在写库的原生会话并停止批次。内容检查超时也不再强杀整个进程树;监督器核对进程是否退出,记录 `process_exit_confirmed` 或 `process_retained`;无法确认释放时输出 `recovery-required.json`,不以强杀代替数据库安全关闭。已确认身份且原生调用返回的失败文档会通过 32 位助手执行 `Close(False)`,再确认进程和 iCDB 地址释放。
- 超时或失败时请求合作停止,保留可能正在原生调用的 worker 和保存/关闭助手,并交后台观察。内容检查超时同样不强杀进程树。保存结果未知时不再自动丢弃关闭;无保存助手且身份、空闲状态已确认的启动/集成失败仍可正常收尾。无法确认释放时保留 `recovery-required.json` 和相关库租约。
## 环境要求
@@ -170,7 +185,7 @@ Windows 版本判断以 Build 为准:`22000` 及以上归为 Win11,`10240–
## 批量策略
每个 CBB 独立执行并写入自己的 evidence 和结果行。可确认安全的单项失败允许继续;遇到 `ICDB_REPAIR_REQUIRED`、超时、原生会话释放未确认等情况,会停止整批并把后续项目标为 `SKIPPED`。点击“当前项完成后停止”或执行器返回 `QUEUE_STOP` 也会停止队列。
每个 CBB 独立执行并写入自己的 evidence 和结果行。资源已释放的单项失败允许继续;未释放时跳过后续同库写入,独立库仍可运行。ICDB 错误本身不等于占用。用户取消或执行器明确返回 `QUEUE_STOP` 才停止本批次后续项,不自动恢复或重放该批次。
GUI 中状态为 `绿色 / Ready` 的条目不能勾选、也不会进入批量队列。
@@ -203,11 +218,11 @@ Mentor 桥耗时需在目标机器另行测量。
- `save.log`:`LICENSE_OK / SAVE_OK / IS_SAVED / POST_SAVE_SYNCHRONIZED / CLOSE_OK / QUIT_OK`;
- `result.json`:状态、中心库标志读取结果、阻断原因;
- `image-status-map.json`:按中心库路径隔离的本机动态 `image_id` 校准快照(当前格式 `schema_version=2`);旧版本或无版本缓存会自动忽略并重新校准,日志会记录当前样本、缓存样本、Ready 主 ID、未映射 ID 和逐项决策原因;
- `backup/`:历史版本的辅助文件副本;当前分支不再生成,所有路径记录 `backup_skipped / manual_backup_policy`。已有历史副本不由本轮改动清理。
- `backup/`:历史版本的辅助文件副本;当前分支不再生成,所有路径记录 `backup_skipped / manual_backup_policy`。已有历史副本不由本轮改动清理。
超时采用无进展计时:界面和批量请求默认 600 秒无进展超时;它不是整个 CBB 的固定总时限,而是连续 600 秒没有可观察进展时才终止。PCB/COM 阶段切换、Project Integration 状态变化、保存标记变化和弹窗处理都会刷新计时器;重复的状态轮询不算进展。需要更快失败时仍可在界面将单项预算调低,但 30 秒只适合已确认 PCB 启动稳定的环境。
超时采用无进展计时:界面和批量请求默认 600 秒没有可观察进展时进入停止与观察流程,不强杀正在进行的调用。PCB/COM 阶段切换、Project Integration 状态变化、保存标记变化和弹窗处理都会刷新计时器;重复的状态轮询不算进展。需要更快失败时仍可调低单项预算,但 30 秒只适合已确认 PCB 启动稳定的环境。
结果不会把“内容检查 PASS”冒充为中心库绿色状态:`library_flag` 只来自对条目状态的只读读取,读不到就是 `UNKNOWN`。Mentor 的树节点 `iImage` 是本机资源列表索引,不能跨安装固定写死为绿色;工具会在日志中为每个 CBB 记录 `image_id`,并从独立的 `Cert/Work` 持久化证据动态校准本机 ID。树 ID 可读时用于交叉核对,冲突不允许标记 Ready;LM 忙导致树 ID 暂时读不到时,完整的持久化证据仍可保留 Ready,缺少持久化证据的条目仍进入待校验。右键菜单文字(包括 `Unverify`)不参与状态判定。
结果不会把“内容检查 PASS”冒充为中心库绿色状态。默认权威桥读取 READY/raw=1;桥失败保留未知/不可用。显式 legacy 模式只观察已有匹配 LM 的树图标,图标不可读时保持 UNKNOWN,不以 Cert/Work 文件直接推断 READY 或 PENDING,也不自动回答原生提示。右键菜单文字(包括 `Unverify`)不参与状态判定。
### 状态接口审计
+47 -1
View File
@@ -1,4 +1,50 @@
# CBB 中心库上传后校验操作规程
# CBB 检查、校验、管理与残留释放操作规程
适用开发版本:0.2.4b2。未完成现场验收,不自动发布或关闭工单。以下为当前规程,下方旧上传校验检查表保留为历史参考。
## 当前操作与队列
选择中心库 `.lmc`,核对 CBB 名称、分区及完整 PRJ/PCB 工程。写操作和人工修复前由操作员保存可用一致备份;应用不自动备份整个中心库,隔离验证/暂存副本不代替人工备份。
普通校验完成原生打开、身份确认、Project Integration、保存和关闭后,必须由权威状态 READY/raw=1 确认成功。ICDB 检查仅尝试打开并检查本次错误,不自动 Repair、Restore 或选择版本,CHECKED 不等于 READY。排队期间变成 READY 的工程在执行前跳过。
原生任务串行排队,检查每项结束后让出机会;缓存筛选保持可用。失败项确认释放后继续独立项目;未释放会话转后台观察,只保留相关库的写入限制,独立库可继续。“停止检查”只取消检查,“取消当前任务并停止队列”取消当前批次后续动作;正在保存/关闭的调用只等待返回。普通等待 5 秒,可能正在写入时最多再等到 120 秒后交后台,这不是结束进程的期限。
保存已返回成功、关闭失败、资源仍占用可以同时出现。查看本轮保存/关闭标记和独立释放结果,不能因进程退出就认定工程已保存。管理更新若已经移除、尚未重新导入就取消,按“需恢复”核对,不自动补做下一次导入。重启发现未完成任务时只核对身份和资源,不自动重放。
权威状态桥不可用时保留未知/不可用;legacy 读取只观察已打开且库身份匹配的 Library Manager。没有匹配窗口、存在模态提示或树不可读时返回 UNKNOWN,不自动启动或应答窗口。
新增拒绝同名冲突;更新先原生移除再导入,失败可能需要人工备份恢复;移除后重新读取目录确认。版本回退由操作员在原生界面选择,核对工程内容后单独校验。
## 历史记录、操作结果与报告
校验和管理遇到历史释放记录时,先复核实际进程、地址、独占文件及导出副本。确认释放后归档记录再继续,不以历史 JSON 本身判定真实占用;同目标未完成写入仍需核对,不自动重放。释放证明在管理首次原生调用前写入检查点,并在结果中保留。
管理操作生成目标工程、库配置和共享依赖的写前/写后主要文件清单及新增、删除、修改差分;不增加自动全库备份,也不声称检查了其他 CBB 内容。报告不可用与原生操作失败分别记录;本任务已释放时可继续独立项。
## 默认释放与单独确认的强制客户端操作
1. 停止新任务,预览当前库的工程、程序、PID 和历史记录。默认“确认释放”仅正常关闭作用域及当前身份已确认的客户端。未保存更改由操作员选择保存、放弃或保留;未知提示保持现场。
2. 数据库服务器可能服务同库其他用户。本工具不允许强制结束服务器,命令行指向本库也不构成独占授权;由库管理员核对用户和服务器状态。
3. “强制结束客户端”是另行确认的异常处理,只允许属于当前 Windows 用户和会话、已确认作用域和精确进程身份的 Expedition PCB、Library Manager、viewdraw、iCDB Project Backup 客户端。后台导入器、脚本宿主和归属未知的进程同样不在强制名单。
4. 警告明确告知客户端中其他已打开工程可能一并退出、未保存更改会丢失、写入中的工程可能需要恢复。执行前持久保存库外 `consent.json`,包含目标库、名单、创建身份和排除项;写入失败不执行。派发处核对快照身份和原预览范围,最终句柄处核对创建时间、完整映像、用户 SID、Windows 会话和原生活动;变化时拒绝。
5. 预览和执行前检查进度与本轮 Output。本工具登记的 Save/Close/原生调用尚未确认返回,或 Backup/Restore/Repair 进行中、状态无法确认时,不放行普通“强制结束客户端”。可查看后台窗口的标题、输出和所在桌面;仅当当前窗口唯一存在已识别的 Cancel/取消按钮时,提供单独确认的原生取消。核对身份、用户会话和按钮快照后发送请求,再重新检查实际结果;没有识别到按钮时不猜菜单参数,不抢占前台。
6. 每项动作前后持久记录实际结果,异常保留已完成/未执行项。若最终记录保存失败但独立库释放证明已确认,界面明确提示记录失败并接受释放事实,不伪造新的占用。
7. 客户端退出仅证明进程结束,不表示修复成功或 READY。服务器仍活跃或库状态未知时保留 PARTIAL、检查点和映射。只有完整的独立释放证明才解除匹配的历史记录;空进程列表本身不是释放证明。
## 失败、证据和回退
保留原生错误、操作员、Mentor 版本、库/工程身份和证据目录;原生日志上传前脱敏。不得删除原生地址/锁文件或修改 LMC 伪造状态。强制客户端退出不可撤销;工程恢复依靠人工一致备份或原生历史版本,不自动回退共享库或重放写操作。
程序回退须先关闭当前客户端,再运行此前候选包;它不恢复丢失更改或被中断的原生事务。工程恢复另行在隔离副本验证,保留前后证据并重新校验。Windows 10、共享库跨机器并发和现场服务器失联恢复仍待验收,fake 测试和文件头损坏试验不代替现场验收。
## 已知限制与证据位置
A/B 输入的原生 EDIF 导出曾返回 `0xC0000005`,尚未定位,不能将该输入记成导入通过。Windows 10、跨机器共享库和现场失联故障副本仍待验收;跨机释放依赖 SMB 共享模式语义,目前没有跨机实测。自动 ICDB Repair 保持关闭,共享数据库地址残留或晚到弹窗按现场状态继续核对。
代码合并不表示现场验收完成。文档中的本机路径、日志、截图、测试 XML 和候选包未入库;仓库仅保留验证摘要,原始材料需在验收机器查看或另行归档。
## 历史上传校验检查表
## 1. 输入与证据
+12
View File
@@ -8,6 +8,18 @@
操作前正常关闭使用相同库的 PCB、Library Manager、原理图/符号编辑客户端,核对原服务器会话已结束,保留包含数据库、原备份和日志的完整一致副本。以下写入先在副本上进行。
## 残留客户端与共享服务器
0.2.4b2 的默认检查、校验、取消和释放流程保留正在原生调用中的 worker 和保存/关闭助手。停止只阻止后续动作,必要时转后台观察;保存、关闭、进程退出和资源释放分别记录。独立确认的“强制结束客户端”也不放行本工具尚未返回的 Save/Close。符合其他释放条件的受支持客户端仍可另行确认结束;这会影响其其他工程,属于异常退出处理,不能替代 Repair。
ICDB 数据库服务器即使命令行指向当前中心库,也可能服务其他用户,**本工具不允许强制结束服务器**。归属未知的进程、后台导入器和脚本宿主同样不允许强制结束;由库管理员核对用户和服务器状态后处理。
强制客户端动作必须先在库外证据目录持久写入确认名单、库和进程创建身份,再核对当前进程句柄、用户 SID 和 Windows 会话;记录失败或身份变化时不执行。客户端结束后仍需独立证明库已释放,重新打开工程并核对版本,必要时从人工一致备份或原生历史版本恢复。不能把“进程已结束”解释为“ICDB 已修复”。具体步骤见 [操作规程](../SOP.md)。
原生 Backup/Restore/Repair 进行中时不强制结束。释放预览可查看后台桌面上的原生输出;只有当前窗口唯一可识别的取消按钮才提供确认请求,发送后仍需重新检查是否停止。输出或取消按钮无法确认时保留现场,不猜测原生命令。
## 原生工程数据库 Repair
1. 启动本机的 iCDB Project Backup,并打开实际报错工程对应的 `.prj`。本机可用 PowerShell 命令形式如下,工程路径需替换为自己的隔离副本:
@@ -0,0 +1,366 @@
# CBB 中心库上传后校验工具(cbb-refresh-verify-795)全面扫描与阻塞分析
- **扫描对象**:`https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795`
- **默认分支**:`main` @ `39985f7`(`chore: release v0.2.3-beta.1`,2026-10-29)
- **活跃分支**:`codex/temp-issue-2-icdb-recovery` @ `29959e2`(2026-11-04 14:16,领先 `main` **49 个提交 / +13,173 行**,未合并、未发布)
- **标签**:`r4-stability-20260929`、`v0.2.2-beta.1`、`v0.2.3-beta.1`
- **扫描时间**:2026-11-04
- **扫描方式**:全量克隆 + 逐模块源码阅读(38 个源文件 / 8,180 行)+ 两个分支完整 pytest + 32 位桥 PE 导入表解析与实机启动 + 关键判定函数行为探针 + 与 `docs/` 现场记录交叉核对
---
> 关联工单:<https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/issues/4>
>
> 扫描基准:`main@39985f7`(v0.2.3-beta.1)、`codex/temp-issue-2-icdb-recovery@29959e2`。
> 本文是扫描时的静态与实测记录;修复进展以工单和后续 PR 为准。
## 0. 一页结论
1. **单机、Win11、有 32 位状态桥、CBB 无异常** 的条件下,`v0.2.3-beta.1` 的"ReusableCircuit PCB → Project Integration → Save/Close → `IsVerifiedReuseBlock` READY"链路是成立的(发布说明有隔离实机 PASS 记录),工具没有伪造绿色状态,失败即阻塞的设计方向正确。
2. **批量场景是这个工具真正的短板**:状态判定依赖外部 32 位桥,但**桥的 x86 VC++/MFC 运行时不随包提供**(本机实测 `0xC0000135` 无法启动);没有桥时工具仍会对每个 CBB 做完整刷新再判 FAIL;叠加"未知弹窗即停批""残留 `sAddress.adr` 即阻断""本机任何 PCB 打开即拒跑",批量任务在现场高概率中途停止。
3. **Win10 是明确的未验收边界**:LM 原生菜单在私有桌面上不派发 MFC 命令(`docs/API-FINDINGS.md`),直接 PCB 路线在 Win10 上是否会提交 Verify 状态没有现场证据。
4. **共享中心库并发没有真正的互斥**:`main` 只有进程内 `Local\` 命名互斥 + 本机进程探测;另一台机器/另一个工程师同时操作同名 CBB 时无任何防护。活跃分支新增了 sidecar 库锁,但它是"合作式"文件锁,Mentor 本身不参与。
5. **ICDB 失联修复尚未闭环**:分支已经做到"原生 Repair → 重开 → 历史版本选择 → 校验 → READY",但文档明确"服务器会话失联经实际按钮触发的完整链路仍需通过",并且出现过"Repair 后 BackAnnotate 被原生拒绝"的失败结论。
6. 代码质量整体高于同类现场工具(分层清晰、证据链完整、失败关闭),但存在若干**可复现的功能缺陷**(内容检查文本误判、`.prj` 编码假设、超时后孤儿 worker、save helper 无超时等),详见第 3 节。
**一句话**:这个工具现在能"证明一个 CBB 是否真的绿",但还不能稳定地"把一批 CBB 刷成绿";阻塞点不在算法,而在**判定链的部署完整性、现场弹窗/占用环境、以及批量的工程化**。
---
## 1. 系统定位与运行链路
### 1.1 定位
| 项 | 内容 |
| --- | --- |
| 处理对象 | 已上传到 Mentor 中心库(`.lmc` + `ReusableBlocksLP`)的 CBB 条目 |
| 核心动作 | 打开该 CBB 的 ReusableCircuit PCB → Project Integration 四项同步 → 保存 Reusable Block cell → 关闭 → 只读读取权威 Verify 状态 |
| 不做什么 | 不做 Part Number 替换、不做上传、不直接改写中心库数据来伪造绿色 |
| 成功判据 | `worker` 完成 + 32 位桥 `IsVerifiedReuseBlock` 返回 `READY / raw=1`,二者缺一即 FAIL |
### 1.2 主链路(`main`)
```
GUI (ui/main_window.py)
└─ run_batch (verify_runner.py) # 串行,max_workers=1
└─ run_verify # 会话级互斥 Local\CbbRefreshVerify795
├─ preflight # Work 下唯一 PRJ / PCBDesignPath / 链接越界
├─ unreleased_addresses # sAddress.adr 残留即 BLOCKED
├─ ReuseBlockImporter -chkContents # 可选:LibSpec + DMS 登录配置
├─ icdb_ports_bindable # 10000-10100 端口预检
├─ DesktopWorker (私有桌面) # 32 位桥/弹窗代理同桌面
│ └─ execute_job (mentor_adapter.py)
│ ├─ 备份(仅本地盘) / CentralLibrary 路径修正
│ ├─ ExpeditionPCB.exe <pcb> -ReusableCircuit
│ ├─ wait_for_target_document (COM 身份核对)
│ ├─ refresh_integration (四项 InSynch)
│ └─ cscript save_verified_block.vbs (Validate/Save/IsSaved/Close/Quit)
├─ cleanup_verify_processes # 只观察退出,不强杀
└─ IsVerifiedReuseBlock (status_bridge.py) → READY / PENDING / UNAVAILABLE
```
对话框监督是一条独立链路:`DesktopWorker` 在私有桌面启动 `--dialog-agent` 伴生进程,agent 用 Win32 + UIA 抓窗口快照,`dialog_supervisor` 按 `assets/dialog_rules.v1.json` 的**阶段化白名单规则**决策,通过 JSONL loopback IPC 回传。
### 1.3 每次校验的产物
`evidence/verify-*/`:`result.json`、`events.jsonl`、`dialogs.jsonl`、`save.log`、`performance.json`、`bridge-status.json`、`diagnostics/`(preflight、native-logs、icdb-recovery)、失败时的 `recovery-required.json`。
### 1.4 现场自述的验证状态(文档)
- `v0.2.3-beta.1`:`pytest` 471 passed;EE7.9.5 隔离实测"破坏 ICDB 文件头 → 备份恢复 → 重新打开 → PASS/GREEN/READY"。
- 明确未验收:Win10、真实网络库大批量、现场 ICDB 失联的自动 Repair 全链路。
- 旧批次 68 项中 51 PASS / 17 FAIL,失败类型已在 `docs/2026-09-29-stability-r4.md` 分类。
---
## 2. 规模与结构
| 指标 | main | temp-issue-2 分支 |
| --- | --- | --- |
| 提交数(相对 main) | — | +49 |
| 源文件 / 行数 | 38 / 8,180 | 47 / 约 13,000+ |
| 测试文件 / 行数 | 35 / 5,752 | 约 50 / 9,000+ |
| 测试用例 | 471 | 1,196 |
| 本机 pytest 结果 | **463 passed / 8 failed** | **1,178 passed / 18 failed** |
| Python / 依赖 | 3.12 + PySide6 6.11 + pywin32 + pywinauto + Pillow | 同 main |
| 打包 | PyInstaller onedir(`CbbRefreshVerify-Background`) | 同 main |
> 本机失败均与环境有关(未完成 `pip install -e .`、无 Mentor 安装、缺 x86 VC 运行时、长路径/编码),不是断言暴露的产品逻辑失败。具体归类见第 7 节。
---
## 3. 功能性问题清单
严重度:**P0** 阻塞交付 / 数据安全风险;**P1** 现场高概率失败或明显功能缺陷;**P2** 可用性、可维护性、规模化问题。
### 3.1 判定链与证据
| 编号 | 严重度 | 问题 | 代码证据 | 影响 |
| --- | --- | --- | --- | --- |
| **F-01** | P0 | **状态桥不可用时仍执行完整刷新,最后才判 FAIL**。批量入口不检查桥是否可用;桥只在 worker 成功后才查询 | `verify_runner.py:981-1000`;UI 读库 `ui/main_window.py:63-68`;`batch_status.authoritative_status_map` 失败只返回 `UNAVAILABLE` | 每个 CBB 白做 2-10 分钟原生刷新 + 一次 Save/Close,然后 `BRIDGE_UNAVAILABLE`;3000 项批量等于 3000 次无效写库 |
| **F-02** | P0 | **32 位状态桥依赖 x86 `mfc140/MSVCP140/VCRUNTIME140`,打包清单只带 exe、未声明或检测该运行时**。本机 SysWOW64 无 x86 运行时,桥启动即 `0xC0000135`(发布 ZIP 不在仓库中,需在发布包上复核是否另行分发) | `packaging/CbbRefreshVerify.spec:4-14` 仅 `binaries=[CbbStatusBridge32.exe]`;PE 导入表实测:`KERNEL32, mfc140, MSVCP140, VCRUNTIME140, api-ms-win-crt-*` | 未额外安装 VC++ 2015-2022 x86 运行时的机器,桥永远不可用 → 所有项最终 FAIL |
| **F-03** | P1 | 内容检查结果用正则 `failed\|error\|cannot connect\|bad data` 判失败,**"0 errors"、路径含 `error` 也会判 FAIL** | `verify_runner.py:30-31, 128-141`;本机实测 `classify_output(0, "...successfully! 0 errors...") == 'FAIL'` | 内容检查偶发误报阻断,且现场难以看出是文本匹配导致 |
| **F-04** | P1 | `preflight` 用 `read_text(encoding='utf-8', errors='replace')` 解析 `.prj`;UTF-16/ANSI 工程会读不到 `PCBDesignPath` → 误判"工程没有 PCBDesignPath" | `verify_runner.py:157-161`;对照 `backup_recovery._project_text`(`backup_recovery.py:129-136`)已正确区分 UTF-16/UTF-8/mbcs | 老工程/修复后工程被预检直接判 BLOCKED,且看起来像"工程数据坏了" |
| **F-05** | P1 | 内容检查报告写入**中心库目录** `LogFiles/`;写失败只拼进局部 `stderr`,PASS 路径下该信息被丢弃 | `verify_runner.py:688-693` | 只读共享/权限不足时无任何提示;中心库额外产生日志文件 |
| **F-06** | P2 | 备份只在 PASS 且 `safe_to_continue` 时清理;失败项 `backup/`(可能数 GB)永久保留,无保留策略 | `verify_runner.py:641-651`;`mentor_adapter.cleanup_evidence_backup` | 长时间批量后磁盘被失败证据占满 |
### 3.2 现场自动化(弹窗 / 桌面 / 进程)
| 编号 | 严重度 | 问题 | 代码证据 | 影响 |
| --- | --- | --- | --- | --- |
| **F-07** | P0 | **任何不在白名单里的弹窗 = `UNKNOWN_DIALOG` = 当前项 FAIL + 整批停止**。规则要求精确英文文本、`GXWND`+`Button` 类名 | `dialog_rules.py:216-236`、`188-236`;`dialog_supervisor.py:458-471` | Mentor 版本/语言/主题差异、任何新提示("Reusable block already exists""license checkout failed"…)都会中断整批;现场 6 项 IME 语言栏误判即此类 |
| **F-08** | P1 | 弹窗代理客户端**每请求 2 秒超时**,而 agent 端 `capture()` 最坏为"N 个窗口 × 0.75s UIA + 子控件枚举";UIA 探测超时后线程泄漏继续跑 | `dialog_agent.py:383-388`(默认 2.0s)、`dialog_probe.py:37-44, 242` | 机器繁忙/窗口多时 `DIALOG_PROTOCOL_TIMEOUT` 被当成致命错误;泄漏的 STA 线程与主探测并发调用 UIA,进一步拖慢 |
| **F-09** | P1 | **Win10 私有桌面上 UIA `InvokePattern` 只关闭原生菜单、不派发 MFC 命令**;`native_lm_verify` 默认关闭,只有显式勾选才走菜单路线 | `docs/API-FINDINGS.md` §7、README「Win10/Win11 兼容性边界」;`verify_runner.py:466-470` | Win10 现场两条路线都不保证成功;目前只能以"Win11 已验证"交付 |
| **F-10** | P1 | **本机只要有任何 `ExpeditionPCB.exe` 在跑就拒绝校验**(即使与本 CBB 无关) | `mentor_adapter.py:619-621` | 工程师日常开着 PCB 时无法点击校验;现场"先关掉所有 PCB"成为隐性前置条件 |
| **F-11** | P1 | `sAddress.adr` 任一残留即 `ICDB_SESSION_UNCONFIRMED` 阻断;正常/异常退出后残留都要人工处理 | `icdb_ownership.py:13-16`;`mentor_adapter.py:644-650`;`verify_runner.py:661-670` | 一次现场崩溃会让后续每次校验都失败,且错误信息指向"未释放",与真实原因(陈旧文件)难区分 |
| **F-12** | P1 | save helper 轮询循环**无超时、无取消检查**;helper 卡在 `doc.Save()` 时 worker 只能等监督器 600s 空闲超时 | `mentor_adapter.py:794-816` | 网络库/许可异常时单次长时间挂起;超时后 helper 仍可能继续写 |
| **F-13** | P0 | **超时/取消只"请求停止"(touch 取消文件),不终止 worker**;worker 只在启动前、PCB 启动循环、保存前 3 处检查取消 | `desktop_worker.py:149-157`;`mentor_adapter.py:684-689, 744-746, 787-789`;`verify_runner.py:892-903` | 监督器已返回 TIMEOUT/FAIL 并释放互斥;worker 仍可能在随后的几秒内完成 Save/Close。若清理时恰好观察到 PCB 已退出,会判定 `safe_to_continue=True` 并启动下一项,而上一项仍可能有迟到的原生写入 |
| **F-14** | P1 | supervisor 崩溃看门狗在 `cancel_path` 存在时(**生产总是存在**)只写取消文件然后返回,**不退出 worker**;只有 `cancel_path=None` 才 `os._exit(75)` | `desktop_worker.py:34-52`;`verify_runner.py:724` 恒写入 `cancel_signal` | 与 F-13 叠加:父进程崩溃后可能留下无监督的自动化进程继续写库 |
### 3.3 数据、权限与边界
| 编号 | 严重度 | 问题 | 代码证据 | 影响 |
| --- | --- | --- | --- | --- |
| **F-15** | P1 | 工具会**改写中心库内 `PCB/pcb.prj` 的 `CentralLibrary`**;修改前只在本次证据目录留单文件副本,网络库不做整工程备份 | `mentor_adapter.py:93-162, 707-712`;README 第 22 行亦承认 | 现场(网络中心库)没有可回退的工程副本;与 README「不允许直接改写中心库 .prj」的表述冲突 |
| **F-16** | P2 | `Work` 下 `.prj` 不唯一即预检跳过;修复/导入后残留的备份 PRJ 会被当作"无法唯一确定" | `verify_runner.py:152-155`;`preflight_skip_reason` | 条目被静默跳过(黄行),用户以为已校验 |
| **F-17** | P2 | Cert 完全不参与判定(设计取舍),但也没有"保存成功但未提交"的独立证据维度 | `verify_runner.py:313-330`;`collect_cert_evidence` 仅记路径 | 只能靠桥区分 PASS/PENDING;桥抖动时无法自证 |
| **F-18** | P2 | 备份策略:网络盘直接跳过递归备份(`backup_skipped`),本地盘 3-30s 预算 | `mentor_adapter.py:632-640, 399-453` | 现场中心库基本都在网络盘 → 实际上没有前置备份 |
### 3.4 性能与规模化
| 编号 | 严重度 | 问题 | 代码证据 | 影响 |
| --- | --- | --- | --- | --- |
| **F-19** | P1 | 每次"读取 CBB"都会:整文件 SHA-256 `.lmc` + 顶层目录指纹 + 逐 CBB 桥查询 + **逐 CBB `preflight_skip_reason`(网络 stat/glob/读 PRJ)** | `batch_status.py:95-122, 194-197`;`ui/main_window.py:257-263` | 3000 项网络库首屏可能数分钟到数十分钟;期间只有一个"正在读取中心库…"标签,无进度、不可取消 |
| **F-20** | P1 | 批量无进度条(仅全库 ICDB 模式有)、**不能取消当前项**(只能"当前项完成后停止")、崩溃后队列不持久化、不续跑 | `ui/main_window.py:417-454, 547-548`;`run_batch` 只在项间检查 stop | 长批次一旦中断只能从头再来;现场无法预估剩余时间 |
| **F-21** | P2 | 每次桥查询(单条路径)都新起一个 32 位进程;批量依赖常驻 `--server`,桥崩溃后整段剩余项直接标 UNAVAILABLE | `status_bridge.py:141-203, 206-332`;`batch_status.py:258-270` | 单条校验慢;桥异常时"剩余全部不可用",无法只重试当前项 |
| **F-22** | P2 | 多处 0.2s 轮询叠加 WMI 查询(`Win32_Process` 每次约 50-200ms)、integration 每次 `native_error()` 读 3 个原生日志 | `mentor_adapter.py:744-761, 800-809`;`integration.py:33-39`;`post_verify_diagnostics.py:32-39` | CPU/网络 I/O 放大;网络日志读取可能反过来拖慢原生操作 |
### 3.5 工程性 / 可维护性
| 编号 | 严重度 | 问题 | 证据 |
| --- | --- | --- | --- |
| **F-23** | P2 | 38 个源文件里 **102 处** `except Exception`/裸 `except`,大量 `except: pass`;故障常被降级成 `UNKNOWN/FAIL`,根因只留在日志 | 静态统计 |
| **F-24** | P2 | GUI/CLI 顶层异常只 `return 1`,不打对话框、不打印到控制台;批处理用户只看到退出码 | `application.py:134` |
| **F-25** | P2 | 无 CI 配置、无 lint/类型检查、无 LICENSE/CHANGELOG;300KB 二进制入库;`.gitattributes` 全量 `-text` | 仓库根目录 |
| **F-26** | P2 | 测试依赖"本机已 `pip install -e .` + 已装 Mentor + 有 x86 运行时",新克隆机器直接 pytest 会有 8 项环境失败 | 第 7 节 |
| **F-27** | P2 | 分支 49 个提交、+13k 行,`pyproject.toml` 版本仍是 `0.2.3b1`,候选包与正式包同版本号 | `pyproject.toml:6`(两分支相同) |
---
## 4. 阻塞点划分(任务视角)
把"阻塞"分成三类,避免把平台限制、工程缺陷、规模问题混在一起:
- **K1 平台/产品硬约束**:Mentor/Windows 决定,改代码无法消除,只能绕行、限定验收范围或推动替代方案。
- **K2 交付阻塞**:团队可修,但当前版本不满足"可交付/可批量"的最低条件。
- **K3 规模化阻塞**:单机少量可用,上批量/上团队前必须解决。
### 4.1 阻塞点总表
| 编号 | 阻塞点 | 类型 | 证据 | 受影响任务 | 解除条件 | 优先级 |
| --- | --- | --- | --- | --- | --- | --- |
| **K1-1** | Library Manager 没有公开的 CBB `Verify()` COM 方法,只有内部 32 位导出 | K1 | `docs/API-FINDINGS.md` §3;`LibraryManager.tlb` 审计 | 所有"编程触发 Verify"的路线 | 接受"只读判定 + ReusableCircuit 刷新"路线,或继续 ABI 研究 | — |
| **K1-2** | 状态判定只能靠 32 位 MFC ABI 桥,64 位 Python 不能直连 | K1 | `docs/API-FINDINGS.md` §3、§9 | 状态读取 | 保持独立 x86 桥(现状) | — |
| **K1-3** | 私有桌面上 LM 原生菜单不派发(Win10 UIA),`SwitchDesktop` 在本机不可用 | K1 | `docs/API-FINDINGS.md` §7 | Win10 现场 | 现场 Windows 10 真机验收;或限定仅 Win11 | — |
| **K1-4** | iCDB 服务器端口必须落在 10000–10100,可能被 Hyper-V/WSL 保留 | K1 | `mentor_adapter.py:978-1082`;文档 §8 | 所有 PCB 启动 | 管理员 `net stop winnat && net start winnat`(工具已给提示) | — |
| **K1-5** | Mentor 对"已验证工程"禁止 BackAnnotate;共享符号库没有 PRJ 入口 | K1 | 工单 #2 文档、`integration.py:93-100` | 已绿工程复检、共享库修复 | 按原生限制记录为"已同步/不支持",或人工处理 | — |
| **K2-1** | **桥不可用(缺 x86 运行时 / 未打包 / 路径找不到)仍可启动批量** | K2 | F-01、F-02 | 全部批量 | 打包 `VC_redist.x86` 或 `/MT` 静态链接;启动自检;无桥禁止批量 | **P0** |
| **K2-2** | **未知弹窗即停批**,白名单覆盖不了现场提示 | K2 | F-07 | 批量、Win10 | 规则分级(致命/可确认/仅记录)、未知弹窗可配置停机或继续并留证 | **P0** |
| **K2-3** | **超时/取消不终止 worker,孤儿进程可能继续写库** | K2 | F-13、F-14 | 数据安全、连续批量 | 明确"取消=停自动化客户端、保留原生会话"的边界;对 worker 进程做可验证的终止/对账;看门狗在崩溃时真正退出 | **P0** |
| **K2-4** | 共享中心库无跨机互斥;`main` 只有进程内互斥 + 本机 PCB 探测 | K2 | `verify_runner.py:504-547, 742-746`;分支 `library_lock.py` | 多人/多机使用 | 合并分支库锁并让上传工具/Mentor 侧协同;或规定"单人独占窗口" | **P1** |
| **K2-5** | 残留 `sAddress.adr` 永久阻断;本机任意 PCB 打开即拒跑 | K2 | F-10、F-11 | 日常使用 | 区分"陈旧地址文件"与"活会话"(探测可达性已具备,但策略过严);PCB 占用按 CBB/工程匹配而非全局 | **P1** |
| **K2-6** | ICDB 失联的自动 Repair 全链路未验收;Repair 后 BackAnnotate 被拒 | K2(分支) | `docs/releases/2026-11-03-field-feedback.md`、`2026-11-04-icdb-review-fixes.md` | 库损坏恢复、全库维护 | 现场真实失联库端到端验收;修复后重开/保存/READY 全绿 | **P0(分支)** |
| **K2-7** | Win10 未验证,且没有可用的菜单触发兜底 | K2 | F-09 | Win10 现场 | Win10 真机单 CBB 验收,或书面限定 Win11 | **P1** |
| **K2-8** | 内容检查文本误判 + `.prj` 编码假设 | K2 | F-03、F-04 | 启用内容检查的批次 | 修正解析/判定,补回归 | **P1** |
| **K3-1** | 首屏扫描 O(全库) + 逐 CBB 预检,无进度/不可取消 | K3 | F-19 | 3000 项批量 | 增量扫描、缓存指纹、预检并行/惰性、进度条 | **P1** |
| **K3-2** | 队列不持久化、不能取消当前项、无断点续跑 | K3 | F-20 | 长批次 | 队列清单落盘 + 逐项检查点 + 取消当前项 | **P1** |
| **K3-3** | 失败证据/备份无上限 | K3 | F-06 | 长期运行 | 证据保留策略(按天/按大小) | **P2** |
| **K3-4** | 质量门禁缺失(无 CI、版本号不变、环境耦合) | K3 | F-25、F-26、F-27 | 发布/回归 | CI(无 Mentor 的 mock 层 + 一台 Mentor 自托管 runner)、版本号规范 | **P2** |
### 4.2 任务可达性矩阵
| 任务场景 | 结论 | 关键阻塞 |
| --- | --- | --- |
| 单 CBB,Win11,本地/网络库,桥可用,无人占用 | ✅ **可用**(有隔离实机 PASS 记录) | 无 |
| 单 CBB,桥不可用/未装 x86 运行时 | ⛔ **必然 FAIL** | K2-1 |
| 小批量(10–50),工作时段 | ⚠️ **可用但易中断** | K2-2、K2-5、F-12 |
| 大批量(数百–3000) | ⛔ **时间与稳定性不可接受** | K3-1、K3-2、K2-2 |
| Win10 现场 | ⛔ **未验收** | K2-7、K1-3 |
| 共享中心库多人同时操作 | ⛔ `main` 无锁;分支为合作式锁 | K2-4 |
| 库损坏/ICDB 失联恢复 | ⚠️ 诊断与手动路径可用;自动 Repair 未闭环 | K2-6 |
| 无 Mentor 的开发机 | ✅ 仅单元测试(需先装包) | F-26 |
### 4.3 建议的解除顺序
1. **先让"批量的失败是可信且廉价的"**:桥预检 + 打包运行时(K2-1)→ 不做无用刷新。
2. **再让"批量的中断可控"**:弹窗分级(K2-2)、取消/超时语义(K2-3)、占用与残留地址策略(K2-5)。
3. **然后才谈规模化**:增量扫描、进度、续跑(K3-1、K3-2)。
4. **平台边界单独验收**:Win10(K2-7)、共享库并发(K2-4)、ICDB 修复闭环(K2-6)。
---
## 5. 现场失败模式 → 代码根因映射
| 现场现象(文档) | 代码根因 | 对应问题 |
| --- | --- | --- |
| 6 项 `UNKNOWN_DIALOG: TF_FloatingLangBar_WndTitle` | IME 语言栏过滤条件过窄(类名+标题+无控件同时满足) | F-07 |
| `TUNER_R836` Save/Close 后桥返回 PENDING | 桥判定 + Verify 未提交;或原生 BackAnnotate 被拒 | F-17、K1-5 |
| `TUNER_R836_A` 数据库打开失败被记 UNKNOWN_DIALOG | 早期靠文本分类;现走规则白名单 | F-07 |
| `1GPHY_8211` 打开时失联 → 后续 Repair `REPAIR_DATABASE_IN_USE` | 失败打开的 PCB 未被关闭(分支已修:启动失败关闭路径) | K2-6 |
| 5 项 Project Integration 符号错误 | 工程/符号数据问题,工具只负责分类与留证 | — |
| 3 项 JobPrefs / 本地 Cell Library 失败 | 环境/工程状态;工具仅诊断不修复 | — |
| 端口 10000-10100 被保留 | `WinError 10013`,Hyper-V/WSL 动态预留 | K1-4 |
| 桥 UNAVAILABLE 但报告"四绿"风险 | 已被 fail-closed 设计排除(不会伪造 READY) | — |
---
## 6. 活跃分支现状(`codex/temp-issue-2-icdb-recovery`)
### 6.1 新增能力(相对 main)
- 库锁与占用探测:`library_lock.py`、`owner_probe.py`、`native_occupancy.py`(与上传工具兼容的 `.cbb-upload.lock` / `.blocked.json` 协议)。
- CBB 管理:`management.py`、`native_management.py`、`management_transaction.py`、`native_engineering_identity.py`、`native_export_guard.py`(新增/同名替换/移除、身份与导出副本保护)。
- 恢复:`reimport_recovery.py`、`scoped_manifest.py`、`post_repair_consent.py`、`pcb_restore_prompt.py`、`maintenance_entry.py`(库外完整工程副本恢复、历史 PCB 版本选择)。
- 策略调整:取消应用自动备份,改为**弹窗提醒手动备份**(`docs/releases/2026-11-03-manual-backup.md`)。
- 启动失败关闭:`failed_session.py` 扩展为"预检失败/数据库打开失败也协作关闭 PCB"。
### 6.2 本机验证结果
- 1,178 passed / 18 failed;其中 8 项与 main 同因(环境),新增失败集中在:
- **长路径**(`native_add_failures` 两个用例,150 字符目录名 → Windows `MAX_PATH`);
- **owner probe 依赖安装**(`native_export_guard` 6 个用例:子进程 `-m cbb_refresh_verify` 不可用);
- **编码**(`native_repair` 一个用例按 cp1252 解码失败);
- **无 Mentor 安装**(`native_repair` 取消用例)。
### 6.3 分支仍未满足的门槛(文档自述)
- 服务器会话失联 → 实际按钮触发 → 自动 Repair → 重开 → 保存关闭 → READY 的**完整链路未验收**。
- `1GPHY_8212` 原生 Repair 后 **BackAnnotate 仍被拒**,结论未变。
- Windows 10 未验证(当前实机 Win11 / EE7.9.5)。
- 历史 PCB 版本恢复记为 `RESTORED_REVIEW_REQUIRED`,**不等于无损修复**。
- 目标缺 `PartsDB` 分区时无法完成完整校验。
- 共享符号库不支持自动修复。
- 版本号未递增、分支未合并。
> 结论:分支在"管理 + 恢复"上补齐了很多能力,但**没有解除 K2-6(ICDB 自动修复闭环)和 K2-7(Win10)**;同时引入了新的环境敏感性(长路径、owner probe 子进程、编码)。
---
## 7. 测试与质量评估
### 7.1 本机结果
```
main: 463 passed, 8 failed, 1 warning, 37.09s
branch: 1,178 passed, 18 failed, 1 warning, 92.29s
```
main 的 8 项失败归因:
| 失败 | 原因 |
| --- | --- |
| `test_application`(2 项)、`test_abnormal_shutdown`、`test_extended_safety`、`test_failure_lifecycle`(共 5 项) | 子进程 `python -m cbb_refresh_verify` 依赖 `pip install -e .`,本机安装未完成 |
| `test_native_repair::test_repair_cancellation...`、`test_verify_runner::test_pcb_launch...` | 本机无 `D:\Mentor\7.9.5EE` 安装 |
| `test_packaging_status_bridge::test_compiled_bridge...` | 缺 x86 VC/MFC 运行时,`0xC0000135` |
### 7.2 覆盖情况
**做得好的**:状态判定(PASS 只能来自桥 READY)、对话框规则/后置条件、ICDB 日志新鲜度、备份/恢复、进程释放、UI 交互都有大量单测,且测试直接编码了"不得伪造绿色"的安全契约。
**缺口**:
- 没有真实 Mentor 的集成测试(全部依赖 fake COM/进程边界);
- 没有 Win10 测试矩阵;
- 没有网络库并发/多人场景;
- 没有长时间稳定性与性能回归(600s 空闲超时、3000 项批量);
- 测试与"本机安装 + 本机 Mentor 路径"耦合,换机即红;
- 无 CI,测试结果靠人工粘贴进 release note。
### 7.3 代码静态指标
- 38 个源文件 / 8,180 行;**102 处**宽泛异常捕获;15 处 `time.sleep`;15 处 subprocess 调用。
- 硬编码 `D:\Mentor\7.9.5EE\SDD_HOME` 兜底 5 处(可作为默认,但需在 UI/日志里明确"发现失败")。新机器上会静默落到不存在的路径,报错信息偏底层。
---
## 8. 修复路线图(建议)
### P0(1–3 天,先让批量"失败得正确")
1. **桥部署闭环**
- 打包 `VC_redist.x86.exe`(或桥改 `/MT` 静态链接),安装脚本检测/安装;
- 启动时执行一次桥自检(`--codec-test` 或空查询),失败则禁用"批量校验"并给出可操作提示;
- `read_library_fast` 返回 `UNAVAILABLE` 时,UI 明确"无法判定绿色,批量将全部失败",不静默入队。
- 验收:干净 Win11 + 无 VC 运行时 → 工具提示安装而不是跑 3000 次无效刷新。
2. **未知弹窗分级**
- 规则增加 `on_unknown: stop | record_continue | block_item`;
- 默认对"无按钮/纯文本/IME/浏览器宿主"仅记录,对"数据库恢复/许可/保存失败"阻断;
- 每次未知弹窗都落 `dialogs.jsonl` + 截图证据,批量结束汇总"待人工确认弹窗"。
- 验收:新增一个未登记弹窗的模拟场景,批量不中断且证据可查。
3. **取消/超时语义**
- 明确"取消 = 停止自动化客户端,保留原生会话";对 worker 增加强制终止开关(仅在未进入 Save 阶段可用);
- 看门狗在 supervisor 退出时按"是否处于 Save/Close"决定终止或保留,并落 `watchdog-decision.json`;
- `run_verify` 在 worker 未确认退出时不得释放互斥、不得继续下一项(当前已部分满足,需把"进程仍活"纳入 `safe_to_continue`)。
- 验收:模拟 Save 卡死 → 工具超时后 worker 确实退出或明确标记"仍活",下一项不启动。
### P1(1–2 周,现场可用性)
4. 弹窗代理超时预算与 `capture()` 并发控制(F-08);UIA 探测改为按进程白名单 + 单窗口超时,禁止泄漏线程。
5. 占用/残留策略(F-10、F-11):PCB 占用按 CBB 路径匹配;`sAddress.adr` 增加"不可达 + 非本机进程 + 超时"三条件判定为陈旧并可清理(需用户确认)。
6. 内容检查判定(F-03)与 `.prj` 编码(F-04)。
7. 中心库写入保护(F-15):改写 `pcb.prj` 前做整文件副本 + 变更清单,网络库至少保留原文件并可一键回滚。
8. 合并分支库锁(K2-4),并与上传工具约定锁协议。
### P2(2–4 周,规模化)
9. 增量扫描/指纹缓存、预检惰性化、进度条、取消当前项、队列持久化与续跑(F-19、F-20)。
10. 证据保留策略与磁盘配额(F-06)。
11. CI:mock 层全量测试 + 一台 Mentor 自托管 runner 做冒烟;版本号规范(`0.2.3b1+candidate.49`)。
12. 清理宽泛异常,至少在关键路径保留 `error_code` 与原始异常链(F-23)。
---
## 9. 附:关键证据索引
| 主题 | 文件:行 |
| --- | --- |
| 成功判据(桥 READY) | `src/cbb_refresh_verify/verify_runner.py:313-330, 981-1000` |
| 内容检查判定 | `src/cbb_refresh_verify/verify_runner.py:30-31, 128-141` |
| 预检(PRJ/PCB/链接) | `src/cbb_refresh_verify/verify_runner.py:144-174` |
| 端口预检 | `src/cbb_refresh_verify/mentor_adapter.py:1061-1082` |
| 数据库地址阻断 | `src/cbb_refresh_verify/icdb_ownership.py:13-16` |
| PCB 全局占用检查 | `src/cbb_refresh_verify/mentor_adapter.py:619-621` |
| CentralLibrary 改写 | `src/cbb_refresh_verify/mentor_adapter.py:93-162, 707-712` |
| save helper 循环 | `src/cbb_refresh_verify/mentor_adapter.py:790-816` |
| 取消信号 | `src/cbb_refresh_verify/desktop_worker.py:149-157`;`mentor_adapter.py:684-689, 744-746, 787-789` |
| 看门狗 | `src/cbb_refresh_verify/desktop_worker.py:25-56` |
| 弹窗规则 | `src/cbb_refresh_verify/dialog_rules.py:188-236`;`assets/dialog_rules.v1.json` |
| 弹窗代理超时 | `src/cbb_refresh_verify/dialog_agent.py:383-388`;`dialog_probe.py:37-44, 242` |
| 状态桥 | `src/cbb_refresh_verify/status_bridge.py:115-203, 206-332` |
| 批量状态扫描 | `src/cbb_refresh_verify/batch_status.py:95-122, 184-277` |
| UI 读库/队列 | `src/cbb_refresh_verify/ui/main_window.py:225-335, 417-548` |
| 打包清单 | `packaging/CbbRefreshVerify.spec:4-25` |
| 分支库锁 | `src/cbb_refresh_verify/library_lock.py:80-200`(分支) |
### 复现命令
```powershell
git clone https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795.git
cd cbb-refresh-verify-795
python -m pip install -e .
python -m pip install pytest pytest-qt
python -m pytest -q # main: 463 passed / 8 failed(本机环境)
git worktree add ..\cbb-branch origin/codex/temp-issue-2-icdb-recovery
cd ..\cbb-branch; python -m pytest -q # branch: 1178 passed / 18 failed(本机环境)
```
桥依赖验证:
```powershell
# 1) 导入表:KERNEL32 / mfc140 / MSVCP140 / VCRUNTIME140 / api-ms-win-crt-*
dumpbin /dependents .\tools\status_bridge\bin\CbbStatusBridge32.exe
# 2) 实机启动:缺 x86 运行时时的退出码 -1073741515 (0xC0000135)
.\tools\status_bridge\bin\CbbStatusBridge32.exe --codec-test
```
@@ -0,0 +1,37 @@
# 后台 ICDB 检查与原生任务排队
## 用户约定
ICDB 检查保持界面可用;校验、库管理、恢复、原生版本窗口等操作提交后串行排队,检查每完成一个 CBB 让出原生执行机会。检查与校验相互独立;检查不自动 Repair、Restore,不改变权威 READY 状态。
## 已确认的原因与修改
- 旧检查占用 MainWindow 的整批 busy 标志,缓存筛选及其他入口随整批禁用。改为独立 IcdbCheckTask、独立取消信号和捕获的工程/库快照;NativeQueue 每项准入,可取消等待,暂停检查不挡后续任务。
- 全局 PCB COM/ROT 可能等待其他工程的模态窗口。检查改为本任务完整 PCB 参数/PID、当前私有桌面、目标 MDI 和本次原生日志联合取证;正式校验继续使用原 COM 集成接口。
- PCB 参数匹配原为子串匹配,会把 `.pcb.backup` 或 `-ReusableCircuitOther` 误认。现按完整命令行参数匹配。
- DialogAgent 在读取子控件/UIA 之前按本任务 PID 范围过滤,执行前重新确认 HWND/PID/class;不读取或点击同桌面的无关弹窗。
- PCB 的临时启动窗口、普通/Reused Block 标题转换及 WM_CLOSE 后 IE 辅助窗口的短暂变化,不能解释为 ICDB 损坏。关闭前必须有目标加载证据;关闭后等待本任务退出,未知窗不点击、不重新发 WM_CLOSE。持续未知窗至期限结束才保留现场并报告。
- 诊断跳过无关集成历史日志,日志读取限定内容和两秒读取期限;超时结束该项,不反复堆积读线程。历史目录较多、历史日志较大不能直接判工程异常。
- ProcessWatch 用已打开的进程句柄观察退出,避免 COM 等待阶段每次轮询都启动 WMI 辅助进程。
- 原生退出后才产生完整 PCB Job Name 的缓冲日志必须在关闭后再核对。PASS 同时要求新 PCB 成功日志、新数据库路径和 PCB PID 装载日志,以及完整 PCB Job Name。
- 失败性质和实际释放事实分别记录。错误包含 ICDB 才标明 ICDB 错误;读写确认、版本选择和未确认原因单独显示。单项失败且释放已确认时继续下一项。
- 后台切库时只更新匹配的库清单;旧工程检查建议在后续管理操作后失效。旧库结果的释放入口不能释放当前另一个库。
- 网络映射由全部活跃 lease 登记保护,包括读取线程尚未交付给界面的 lease;拥有者结束时向仍在使用者转移映射。网络收尾需离开 Qt 线程,避免慢网络同步冻结界面。
## 原生验证范围
Windows 11、Mentor EE7.9.5。实验仅使用 `C:/Temp/CbbBackground-*` 中的完整隔离库;原始六个主要文件记录 SHA-256,来源保持不变。故意修改副本 `icdb.dat` 文件头得到 `SingleFile: ERROR ... in old, unsupported format`,用于验证当前错误分类和失败后的安全退出,不能代替现场失联故障验收。
`diagnostics/icdb-authority-20261105/validate_background_native.py` 运行应用的 `run_verify(maintenance=True)` 完整检查入口,在每次工作桌面放置另一个 PID 的真实模态窗口。要求损坏项报告明确 ICDB 错误并安全退出,下一正常项完成打开检查;无关弹窗在本任务结束前保持未操作;不得执行自动修复。
原始日志及结果在 `diagnostics/icdb-authority-20261105/background-native-result.json` 和对应 C:/Temp 证据目录。冻结包的 BUILD.json、VALIDATION.json 分别记录源码哈希与本包验收证据。
源代码完整链路于 19:16—19:17 通过:损坏项 FAIL / ICDB_REPAIR_REQUIRED / safe_to_continue=true,耗时 62.062 秒;下一正常项 CHECKED / safe_to_continue=true,耗时 26.704 秒。两个其他 PID 的模态窗口均保持未操作,来源六文件哈希不变。此记录另存 `background-native-source-result.json`;最终候选包必须再次通过自身冻结执行器的原生验证。
`validate_background_gui.py` 额外通过真实检查按钮提交检查,再提交权威状态读取并测试排队期间缓存筛选。其单行目录 PENDING 状态为测试注入,原生打开、退出和后续状态读取均实际执行;这不构成现场真实库状态的验收。
r9 冻结执行器实测通过同一批次:故障项 68.531 秒,安全释放;正常项 33.313 秒,CHECKED 且安全释放,其中启动 17.093 秒、关闭 8.703 秒;两个无关弹窗未操作。界面实测 34.703 秒,检查和状态读取顺序为 check-start → check-end → status-start → status-end,排队期间缓存筛选可用,50ms 事件循环计时器最大间隔 79ms,后续原生状态为 READY。该时间只是本机当前样本结果,不作跨样本速度提升比例。
## 保留限制
检查仍需实际启动 Mentor。去掉 COM/WMI 和无关日志开销,不等于消除原生程序启动和数据库加载时间。日志或文档身份不足时报告未确认,不推断 ICDB 损坏。实际未释放的当前任务会停止后续原生写入并提供释放入口,界面缓存操作继续可用。ICDB 自动修复和现场服务器会话失联恢复未验收;本候选包不关闭工单、不作为正式发布。
@@ -1,5 +1,7 @@
# 阻塞范围收紧记录
> 后续审查已修正本记录中“切库清除旧标志”和“只检查 Work”的不足。当前行为与验证请以 [全链路审计](2026-11-05-blocking-lock-audit.md) 为准:状态按库保存,释放包含共享数据库并绑定原任务范围。
本轮处理的是“一个 CBB 出问题后把整批和其他库一起停掉”的误阻塞。
## 已收紧
@@ -0,0 +1,132 @@
# 全链路阻塞与锁审计(2026-11-05)
本次目标:枚举从按钮到批次结束的阻塞条件;消除把单项失败、历史记录或无关进程当成本任务仍占用的情况;用实际运行核对退出和下一项继续。
这是一份开发与验证记录,不把维护说明加入产品 UI。ICDB 批量检查不执行自动 Repair,不把普通打开失败推定为 ICDB 损坏。真正仍在运行的同库写会话继续串行;不能通过删除 Mentor 地址/锁文件、强杀数据库进程来制造“已释放”。
## 已复现的现场链路
旧候选包对 IP101GR 打开检查时,原生 PCB 显示完整的 `Continue to open the design with read/write access permission?` 问题。程序报 `DATABASE_RECOVERY_REQUIRED`,启动收尾函数却只接受 ICDB/数据库加载失败类别,返回 `STARTUP_FAILURE_UNCONFIRMED`,未取消打开。PCB 60544 和启动器 63824 留在本任务私有桌面,导致后续会话不释放。
2026-11-05 使用修改后的真实退出函数,核对本次进程、工程参数、私有桌面和精确原生全文后点击 **No (7)**;两个进程均退出。6 个主要文件(LMC、PRJ、PCB、工程数据库、Layout、PCB工程配置)前后 SHA-256 一致。原有两个 Project Backup 会话没有被终止,没有执行 Repair/Restore。
证据:`diagnostics/icdb-authority-20261105/field-readwrite-cleanup/result.json`。旧候选包原始 evidence 保留,不覆盖。
## 逐阶段清单
“保留”表示正在执行或归属无法确认时仍需要该条件;“修正”表示本次已定位误阻塞并实施回归修复。最终运行结果在文末登记,不能仅凭表格认为现场验收成功。
|序号|阶段 / 代码入口|触发条件及原有影响|处理|
|---|---|---|---|
|1|UI `busy` / ServiceWorker|真实任务执行中禁用启动、编辑路径、关闭;任务异常后必须恢复按钮|保留串行;检查 finished/failure 收尾|
|2|UI 已加载路径|输入路径改变但仍使用旧清单|保留身份检查,不启动旧目标|
|3|UI 选择 / 筛选|隐藏项、READY、缺工程进入检查队列|保留:隐藏项取消勾选;READY 即使选中也排除检查;管理选择独立|
|4|UI 状态刷新|检查结束后条目仍按旧状态禁用或勾选|刷新受影响条目,保留搜索;READY 清旧勾选|
|5|UI management unsafe|切换库清掉唯一的未释放标志,返回原库失去状态|修正:按库保存;A 不影响 B,回到 A 仍保留真实未释放事实|
|6|UI 原生版本会话|正在运行的同库版本窗口与写任务冲突|保留;其它库和清单筛选独立|
|7|版本窗口返回|启动器退出但结果缺失/身份错/子会话未确认|保留当前会话,不因窗口消失虚报恢复成功;释放入口可独立重查|
|8|全局原生 mutex|真实另一个校验/管理/恢复正在使用共享原生通道|保留实际串行;不把互斥失败持久化成本任务数据库 owner|
|9|库协调 OS handle|上传工具/本程序对同库持有独占句柄|保留真实冲突;不能用删文件绕过|
|10|协调文件创建|权限/路径/网络错误也被统一报“有锁”|修正:区分 sharing/lock violation 和访问错误;启动前失败可重试|
|11|历史 checkpoint|同库旧记录、pending、损坏 JSON 一律阻止管理|提供有证据的释放/归档;归属完全丢失时不能证明未知分析副本,列明未确认原因,不擅删|
|12|retained 管理 lease|失败时保留的句柄没有可再次检查的 release probe|修正可重查的精确 operation / scope;未退出仍保留|
|13|显式释放 scope|全机 retained exporter/worker 阻止另一库释放|修正按 operation / library 范围,不采用同名即归属|
|14|释放范围|只检查目标 Work,遗漏共享符号库;或遍历全部库和快照|修正目标库 / 目标 CBB / 实际共享路径,保留共享库互斥|
|15|owner 匹配|原始字符串包含 Work 会误匹配 WorkOld,job.json worker 又漏匹配|精确参数/目录边界及记录的 operation 归属;无法读取归属单列未确认|
|16|PID 重用|旧 PID 恰巧分给另一进程,形成永久旧 owner|新记录绑定创建时间/句柄,旧格式需独立释放证据|
|17|工程 preflight|缺 PRJ/PCB、无工具、非法参数|保留单项失败;未启动会话不保留执行锁、不停其它项|
|18|post-repair proof|非法凭证在任何 worker 创建前被标为 unsafe|修正为启动前拒绝,下一项可继续|
|19|content-check timeout|只读辅助进程超时与真实原生会话混淆|必须区分子进程是否退出,不以 timeout 字样推断 owner|
|20|DesktopWorker 构造|CreateDesktop/CreateJob/CreateProcess 失败,未创建 worker 仍被标 unsafe|修正句柄回滚并返回显式 no-process 证据;已部分启动仍未确认|
|21|原生读写确认|要求确认无其他用户,原程序不 Yes 也不 No,留下 PCB|修正:仅精确全文、PID、工程、桌面一致时 No;等待退出,绝不自动 Yes|
|22|PCB 历史版本选择|有两个版本,需要用户选择;不能代选|记录候选版本;扫描不执行历史回退,取消已识别的本次打开后收尾|
|23|其它已知 fatal|确认错误后仍存在空 PCB|正常关闭本任务空壳;不能要求先判为 ICDB 才允许收尾|
|24|未知弹窗 / 文档|不能证明是空壳、存在文档、非本进程、不同桌面|保留,不盲点按钮;显式给出未释放窗口证据|
|25|PCB 先退出|common 启动器延迟退出|等待 PCB 与启动器,不在任一仍活着时说释放|
|26|No 后旧地址|尝试打开被明确拒绝,但旧地址让本任务空壳不能退出|修正:取消本次打开不要求先取得外部数据库;保留地址,不声称数据库无人占用|
|27|COM 无文档 / ROT|COM 指向他人会话或 ActiveDocument 为空|只依据本任务进程/路径/桌面关闭空壳,不借全局 ROT 关闭他人工程|
|28|保存 / Close 超时|COM 仍可能在写;第二个 close helper 会重复操作|保留已有 helper,观察其释放,不强杀 PCB/数据库或重复保存|
|29|取消 / 主窗口关闭|取消发生在启动、校验、保存任意阶段|合作停止本任务,禁止新项;收尾确认后可再次启动|
|30|会话释放证据|worker 已退出,但 native/automation client 仍在|保留真实未释放;只有独立退出证明才能提升安全状态|
|31|可选报告/性能记录|释放成功后写报告失败又变成 unsafe|修正:报告失败保留告警,不覆盖执行结果和退出事实|
|32|progress / 批次报告|回调/磁盘写入异常终止循环,UI generic failure 挂锁|修正隔离展示错误,保留已完成结果与释放事实|
|33|批次单项 FAIL/BLOCKED/TIMEOUT|某项失败但已正常退出,被错误扩大成全批失败|依据实际 release 继续;用户取消/真正未知会话仍停止同通道任务|
|34|管理预检查|desktop/mapping/capability/source 失败,尚未原生激活即 unsafe|修正显式 activation-start 状态;未启动可重试|
|35|管理 catalog 失败|返回 success=false/safe=true 被异常丢掉 release|修正操作成功与会话释放分别传播|
|36|管理检查点恢复|catalog_before 中断就阻止所有 CBB,和实际 remove 中断混同|区分 mutation 前失败和已发生修改;后者保留数据恢复要求|
|37|无关 LM/PCB/viewdraw|全机任意同名进程阻止目标库检查/恢复/导出|绑定目标工程、库和共享数据库;COM 激活若无法证明隔离仍单独说明|
|38|工程内容导出|收养所有同名 iCDB server / exporter,等待无关进程永不退出|修正仅本次后代和精确分析副本 owner;保留实际后代|
|39|导出副本回收|一个库的 retained export 挡住另一个库|修正 registry 库/副本身份,配合数据库独占读取|
|40|sAddress.adr|存在即被称为“有人占用”|存在只表示需原生确认;保留文件,NOT_IN_USE 只能由严格绑定的 inspection 得出|
|41|备份恢复异常|源路径/清单预检失败也报 unsafe、UI 挂锁|修正沿用是否实际启动/释放事实;目标改动与内容冲突仍阻止安装|
|42|批量状态读取协议|A 响应不匹配就把未查的 B 等都标 MALFORMED|修正重启桥接并逐项读取,不伪造后续状态|
|43|状态缓存身份|只凭字节/名称复用另一库的结果|修正规则版本 + canonical library / block type 身份|
|44|映射关闭|先置 closed,再取消映射;失败以后不能重试|修正确认成功才关闭;保留用户/被改指向的映射|
|45|无关盘枚举失败|另一个盘查询失败阻止目标 share 复用|修正独立确认目标盘;未知盘只作为不可用盘符|
|46|真实地址/共享库占用|同目标 / 实际共享数据库活动 writer|保留实际写互斥;普通错误不得扩展成永久禁用其它独立功能|
|47|未知执行异常|无法知道是否已经启动/改数据|保持未确认;修已知 prelaunch/reporting 根因,不能把一切异常默认放行|
|48|退出后 cscript owner 查询|3 秒包含 helper/Python/COM 启动,现场实际约 4.75 秒,误报查询超时|真实隔离批次复现;改为统一 10 秒有上限查询,超时仍是未确认而非确认占用|
|49|导出全进程 WMI|直接 WMI 卡住会让 supervision 长期保留 worker|移入有上限只读 helper,保留所有名称的真实后代;启动前查询失败不会制造 native owner|
|50|操作后内容/状态失败|移除/新增已返回安全退出,但后续读取失败又丢失释放结论|内容恢复状态与进程释放分别保留,不能用 mutated 标签代替活动会话证明|
|51|启动前证据磁盘错误|目录/request/worker.log 创建失败,被通用异常当原生会话未释放|明确尚未创建 worker 的失败可继续;runtime 指纹写入为可选,不掩盖真实结果|
|52|证据查看器|打开证据路径失败进入通用 worker 错误处理,错误保留映射|只显示查看失败,不创建数据库未释放状态|
|53|维护库归属|目标 RepairTarget 可证明所属库,却仍全局检查 retained worker/export|从原 PRJ 的唯一有效 CentralLibrary 解析身份,复用统一归属检查;普通扫描保持独立|
|54|同库路径别名|同库 drive/UNC/大小写别名被当成另一个库,或 UI resolve 访问失联网络|使用已有 lease 映射和本地规范键,不在控件更新中访问网络|
|55|同盘符多个 retained 库|A 释放后所有权转到当前 A,关闭 A 会断掉仍有 B 会话使用的盘符|关闭前把所有权交给同盘符/共享路径的 retained B;B 未释放不取消映射|
|56|目录输入的库身份|原始输入为目录,释放同目录任意 LMC 都可能误匹配|读取时绑定选中的实际 LMC;未绑定目录不接受 sibling LMC 释放证明|
|57|已释放但待数据恢复|同目标正确阻止重放,却把旧 safe=true 改成 false,进而停止独立目标|保留同目标数据恢复要求,同时保留已释放事实,独立目标可继续|
|58|真实 EDIF 参数形式|程序发出 `-icdb <目录>`,归属解析器只认 `-icdb=<目录>`|兼容实际分离参数并按完整目录边界匹配,不把明确无关的导出器当未知 owner|
|59|Add 的 pre-capture 准备|原始库引用/清单/工程对齐失败发生在导出前,却继承 unsafe|准备阶段失败明确未启动;实际 capture 后始终沿用真实释放证据|
|60|最后一笔导出进程记录|native 和数据库已释放,但 finally 中的 process.json 写入异常覆盖结果|最终报告 I/O 保留真实 release;无退出证明的分支仍保留 owner|
|61|冻结代理启动与 PCB 门禁|代理十秒未就绪可误报,工作进程又只等十五秒|代理启动 30 秒、工作进程门禁 45 秒共享常量;日常 IPC 仍为 2 秒,监听以 0.2 秒轮询响应取消|
|62|连接握手/门禁取消竞态|用户在 handshake 时取消,父进程仍可能先放行 PCB|认证后、门禁放行前、worker 执行原生前都再次检查取消;等待门禁时也响应取消|
|63|残留释放入口|只显示存在进程即拒绝,没有正常关闭、保存提示或确认后结束的路径|增加精确名单、逐项关闭、保存选择、独立强制确认、重新证明与继续未执行项;未知归属也显示供定位|
|64|释放后的全库证明|对普通输出逐个重复 resolve,同一副本 23,218 条记录导致原生退出后仍等待约 97 秒|一次规范库路径;复用已拒绝链接的扫描 metadata,先筛原生数据库候选再独占读取;同一库证明约 6.9 秒|
|65|管理启动前历史阻止|旧记录让新任务尚未启动便失败,却被 LibraryBusy 类型强制改成 unsafe,UI 永久禁用;扫描也不显示记录和无 JSON 的 retained lease|新任务在互斥内先做完整复核,归档旧记录并恢复句柄/映射;未启动不制造 owner,UI 旧标志转为重查要求,历史/内存记录可见;同目标未恢复修改仍禁止重放|
## 执行与验收顺序
1. 保存现场弹窗、PID、request、release 和主要文件 hash;正常取消本次遗留打开。
2. 各域先写失败回归,分别修空壳/弹窗、UI/批次、库协调/管理、导出/恢复、网络/状态桥接。
3. 统一串行运行全量测试,避免测试之间竞争真实 Windows named mutex。
4. 在新隔离库运行应用检查入口:故障/提示项 → 正常项;确认无自动 Repair,无遗留本次 native owner,原来源文件保持不变。
5. 新目录生成候选冻结包;校验冻结源码一致、启动/退出、x86 桥接、冻结检查链路,再打开供用户测试。
## 验证记录(持续更新)
- 现场遗留读写确认:真实 No 取消,PCB 60544 / launcher 63824 退出,6 项 hash 一致。
- startup / desktop / native repair 相关回归:194 passed(后续统一全量再次覆盖)。
- 第一轮真实副本批次:**未通过**。成功复现原始读写确认;新取消路径正常退出,但随后 cscript owner helper 的 3 秒超时将结果重新标为 unsafe,后两项 SKIPPED。原库 6 项 hash 一致。保留 `blocking-live/result.json` 与完整原始记录。该误阻塞已新增回归修正,原副本重跑结果见下一条。
- 修正后的真实副本批次:**通过退出与继续验收**。`blocking-live/retry1/result.json`:现场副本 `FAIL / DATABASE_SAFETY_STOP / safe=true`(No 正常取消);故意损坏副本 `FAIL / ICDB_REPAIR_REQUIRED / safe=true`;健康副本 `CHECKED / safe=true`。三项各自 `session-release.json` 均为 worker 已退出且 remaining_pids=[]。源 6 项 hash 一致,无 Repair/Restore/保存集成。此结果证明失败项能够退出并继续,不证明故障内容已修复。
- 第一轮统一全量:1452 passed / 3 failed。三项为旧测试的相对路径归属 fixture、缺 library 的内部请求 fixture,以及新增的应用 JSON 回收 callsite 未纳入静态审计白名单。已限定修正 fixture 和 JSON callsite;未放宽未知原生归属、未允许删除 Mentor 文件。审查额外发现的 55–60 项已补回归并纳入下述复跑。
- 统一全量复跑:**1479 passed**,1 条既有 pywinauto STA warning;`blocking-final-pytest.xml`。编译与 diff 检查通过。
- 后续冻结实测:现场副本 FAIL 退出后继续;损坏样本遇到 `DIALOG_AGENT_START_TIMEOUT`,同样已释放并继续;健康样本 CHECKED。该结果只通过退出/继续验收,损坏样本本轮未进入 PCB,不能算内容检查通过。由此补充 61–62 的启动/取消修正。
- 启动修正增量:**239 passed**,含代理、worker 门禁、普通验证、批次、保存/退出、修复凭证与 UI 回归;`startup-delta-pytest.xml`。本机另有独立 pytest 使用相同产品 mutex,增量回归使用专有测试 mutex 名,产品互斥名不变。
- **最终 r3 冻结链路通过**:`blocking-live/frozen-r3/result.json` 中现场副本 `FAIL / DATABASE_SAFETY_STOP / safe=true` → 损坏样本 `FAIL / ICDB_REPAIR_REQUIRED / safe=true` → 健康样本 `CHECKED / safe=true`。三项实际启动 PCB,三项本次会话均退出、没有未执行项、没有 Repair/Restore,原库 6 项主要文件 hash 一致。
- 冻结验证使用与包内字节码逐项比对一致的控制层,worker、弹窗代理、原生日志和进程探测均运行实际冻结 exe;没有把源码子进程冒充冻结子进程。24 个 Python 模块及 2 个 VBS 内容一致;实际 GUI 启动/退出、x86 codec、包内 VC runtime 加载、完整进程快照 helper 通过。
- 最终独立只读复核:已报告的 P2 修正后无剩余 P1/P2;后续启动预算/取消竞态也单独复核通过。当前实测环境为 Windows 11 / Mentor EE7.9.5,Windows 10 未实机验收。
- 候选包目录:`dist/blocking-locks-20261105-candidate-r3/CbbRefreshVerify-Background`。`BUILD.json` 记录源码基线/实际文件 hash,`VALIDATION.json` 记录全量与增量测试和原生链路结果;仍是候选包,不合并发布、不关闭工单。
本次不宣称 ICDB 自动修复已通过验收。生产库的其它原生 Project Backup 会话保留;它们连接失败不等于数据库损坏,也不等于可强夺写权限。
## 确认释放会话(后续增量)
用户反馈 r3“释放检查”遇到两个 Project Backup 进程只返回不能解除,没有关闭入口。本次增加“释放残留会话…”:先列当前库关联进程和工程,确认后逐项正常关闭。已加载 PCB 允许正常关闭;标准保存问题由用户明确选择保存、放弃或取消。已知启动确认可正常取消。单项未退出继续处理其它项;全部退出后再证明库文件可独占读取,归档本应用旧 checkpoint,解除保留句柄与映射,恢复按钮。
正常关闭无法完成时,提供独立的“强制结束所列会话…”警告确认,默认取消,列出项目、进程和 PID,明确未保存修改丢失及写入中断风险。仅对已证明属于当前库、重新核对 command/creation/image/进程句柄身份的列表发送结束请求;不是全机按进程名清理。此显式功能不改变普通检查/关闭/取消任务的合作退出策略,不删除 Mentor 地址、数据库或锁文件。
不安全批次结果可直接进入释放;取得当前库释放证明后,重新显示结果框,用户可点击“继续未执行项”。只恢复原检查/校验中 SKIPPED 的条目,重新套用当前可见性及非 READY 资格。失败项不自动重放,库增删改不自动重放。
原生实际验证:`diagnostics/icdb-authority-20261105/confirmed-release-b00c3b7c/result.json`。在 `C:/Temp/CbbBlocking-15cb2f99/healthy` 一致副本运行真实 Qt 确认按钮,正常关闭 Project Backup PID60404,监督 worker PID13008 随后退出,full release proof 通过;6 个主要文件 SHA-256 前后一致,未执行 Repair/Restore。截图 `preview.png` / `released.png` 保留。初次试验遇到不存在的 pywin32 查询 API,已改用 Windows QueryFullProcessImageNameW 并增加真实句柄回归后重跑通过;初次证据不覆盖。
当前边界:跨机器会话、不可读归属、操作系统拒绝权限或原生数据库地址未释放,不能通过本机结束请求制造成功。最终证明失败会保留原始原因与证据。强制结束仅证明进程结束,不证明工程数据完好;ICDB 自动修复仍未验收。标准 PCB 保存提示有精确消息/按钮与 UI 单元验证,尚无真实未保存 PCB 工程的端到端保存验收。
验证记录:统一全量 **1509 passed**(`residual-release-pytest.xml`,328.25 秒,1 条既有 STA warning);最新释放/未知归属/确认名单竞态/界面继续入口/库协调/文件保护增量 **83 passed**(16.42 秒)。未知归属进程现在显示为诊断行;只有预览时和执行时都证明属于当前库的项才可以关闭,执行时恢复可读不能扩大用户确认过的名单。代码复核发现的结果框隐藏及确认名单竞态均补先失败的回归后修正,最后只读复核未发现剩余 P1/P2。编译与 diff 空白检查通过。
r4 冻结增量的第一次实际试验没有通过 100 秒测试门槛:原生 PID25260 / worker PID61916 已正常退出,旧 checkpoint 已完整归档且 release proof 成功,但全库证明对 23,218 条记录反复路径解析约 97 秒,测试控制器已提前超时。证据 `confirmed-release-dfb29e88` 保留,不能算该轮通过。新增失败回归后修正了普通文件的多余解析,同一副本实际 proof **6.906 秒 / 12 个独占文件**;最新涉及库释放、维护范围、网络映射和 UI 的回归 **105 passed**(17.09 秒)。新的 r5 候选需再次执行完整冻结释放链路。
**r5 冻结链路已通过**:`confirmed-release-99f3f4e0/result.json`。真实 Qt 确认入口关闭 Project Backup PID48080,包内监督 worker PID23040 退出,旧 checkpoint 归档并清除,12 个原生文件独占读取证明通过,6 个主要文件 hash 一致。整个试验约 27 秒(含启动后等待 15 秒、列表查询、关闭及复查),原生退出至释放完成约 6 秒。随后立即使用同一 r5 exe 运行真实 PCB 检查,`next-check-result.json` 返回 `CHECKED / safe=true`,实际打开正常并退出、remaining_pids=[]、没有 Repair/Restore,证明清理后可以继续后续任务。此检查耗时约 100 秒,其中 PCB 启动约 86 秒;不能把该启动耗时称为释放卡死。
r5 包内 26 个 Python 模块及 2 个 VBS 与当前源码一致,实际主窗口启动/退出、x86 codec、包内 runtime 和冻结进程探测 helper 均通过。性能改动经独立只读复核,未发现新 P1/P2。只交付候选包,不关闭工单或宣称 ICDB 数据已自动修复。
r6 用户中心库误阻止的根因、源日志、修正和冻结管理链路见 `2026-11-05-management-phantom-blocker.md`。统一全量 1518 passed;完整原生移除与重新读目录验证通过。历史记录本身不再被直接当作活跃会话,管理入口允许重新复核;生产库工程未改动。
@@ -2,7 +2,7 @@
## 现象
原生 worker 创建私有 Windows Desktop 后,如果 Job Object 创建或 worker 进程启动失败,构造函数直接抛出异常,已创建的 Desktop 和 Job 句柄没有释放。重复触发失败会留下孤立的私有桌面对象,表现为桌面资源持续增加、后续同名或同类原生任务启动异常。
代码与回归测试确认:原生 worker 创建私有 Windows Desktop 后,如果 Job Object 创建或 worker 进程启动失败,构造函数直接抛出异常,已创建的 Desktop 和 Job 句柄没有释放。重复失败存在累积私有桌面资源的风险;本次未对现场所有桌面异常作归因。
状态读取路径还有一个相同的边界:`read_library_flag` 在进入资源清理保护前加载 UI 依赖;依赖加载异常时,刚创建的私有桌面无法执行 `CloseDesktop`。
@@ -11,26 +11,37 @@
1. `DesktopWorker.__init__` 的资源申请跨越多个可能失败的步骤,但没有事务式回滚。
2. 进程句柄、Job 句柄和 Desktop 句柄的释放顺序没有覆盖构造阶段异常。
3. `read_library_flag` 的依赖导入位于 `CreateDesktop` 之后、`try/finally` 之前。
4. `read_library_flag` 和 `read_library_tree_image_map` 都在清理保护外先创建 Desktop,再执行 `Path(library)` 等准备逻辑;准备阶段抛异常时也会泄漏。
## 解决方案
- 在 `DesktopWorker` 构造开始时初始化资源字段,并把 Desktop、Job、worker 创建放入同一异常边界。
- 构造失败时按 worker、Job、Desktop 顺序回滚;已启动的 worker 先终止,避免留下无父级的原生进程。
- 在状态读取创建 Desktop 前完成 UI 依赖加载,确保依赖异常不会产生未受保护的 Desktop。
- 增加回归测试,模拟 worker 启动失败并断言 Desktop、Job 均已关闭。
- 两处状态读取均在准备逻辑完成后、`try/finally` 内创建 Desktop,清理时检查资源是否存在。
- 测试中分离 Desktop、Job、Process、Thread 的接口,断言失败后关闭顺序;覆盖四个构造失败点与进程创建后的终止分支。
## 影响范围
- 只影响原生 worker 启动失败和状态读取依赖加载失败时的清理路径。
- 覆盖原生 worker 构造失败,以及状态读取准备和启动阶段的资源清理路径。
- 正常任务的私有桌面、Job Object、进程树和数据库释放策略不变。
- 不会切换用户输入桌面,也不会改变中心库写入策略。
## 验证
- `python -m pytest -q tests/test_desktop_worker.py tests/test_dialog_agent.py tests/test_native_inspection_release_regressions.py tests/test_verify_diagnostics.py`:33 passed。
- 修复分支完整 `python -m pytest -q --tb=short`:480 passed。
- 同环境、同工作树切换到 `origin/main`(`39985f7`)完整运行相同命令:471 passed。两次运行依次执行,均仅有一条 pywinauto STA 警告。
- `tests/test_desktop_worker.py`:9 passed。两处状态读取准备阶段泄漏的新增测试在修复前 2 failed,修复后通过。
- 变异验证:把回滚中的 `self.job.Close()` 故意改为 `self.job.CloseDesktop()`,测试产生 2 failed;恢复正确代码后通过。
- `python -m compileall -q src tests`:通过。
- `git diff --check`:通过。
- `git -c core.whitespace=cr-at-eol diff origin/main HEAD --check`:通过。命令显式识别 Windows CRLF;仓库 `.gitattributes` 为 `* -text`,保留源文件换行,不依赖本机 `core.whitespace` 配置。
此前 #7 将一次运行中的 `test_backup_recovery.py::test_invalid_source_does_not_touch_target` 失败称为既有基线问题,缺少同环境对照,撤回该归因。本次该用例在 main 与修复分支均通过。审查者在旧 WIP 基线测得的 19 个失败不属于本次从 main 独立提交后的测试范围。
## 合并范围修正
原 PR #7 以 ICDB WIP 分支为父提交,却以 main 为合并目标,实际携带 59 个提交、118 个文件。现从 main 单独 cherry-pick 桌面修复并补充上述审查修正;相对 main 只有本工单、两个源码文件和一个测试文件,共 4 个文件。独立代码复审未发现待处理问题。
## 分支
`codex/desktop-leak-pr`
`codex/desktop-leak-main`
@@ -0,0 +1,47 @@
# 导出身份比对与备份恢复:阻塞审计及修正
本记录只覆盖 `native_engineering_identity.py`、`native_export_guard.py`、`backup_recovery.py`。没有启动原生工具,没有修改生产库、Mentor 地址、锁或数据库文件。完整应用现场验证由主任务单独记录。
## 从前到后的阻塞点
| 阶段/触发条件 | 修正后行为 | 保留原因/证据 |
|---|---|---|
| 输入工程身份、库引用、工程完整性或独立目录不符 | 本项拒绝;没有创建原生会话时可继续下一项 | 防止替换错误工程;不是占用锁 |
| EE7.9.5 导出程序缺失或二进制版本未通过能力探测 | 本项拒绝;未启动即不宣称会话未释放 | 不猜测原生接口 |
| CBB 操作互斥量实际被其它操作持有 | 不并行启动;保留原有串行约束 | 原生 COM/共享库任务串行运行 |
| 库协调句柄确实独占或已有未释放任务 | 保留互斥 | 句柄/释放证明由 `library_lock` 负责 |
| 机器上存在其它工程的 PCB、LM 或原理图进程 | 不再单凭进程名阻塞恢复 | 完整工程/数据库/库路径参数匹配;`WorkOld` 不匹配 `Work` |
| 有原生进程但没有可识别的工程参数 | 本项返回 `TARGET_NATIVE_OWNER_UNCONFIRMED` | 尚不能通过进程名称判断它打开的文档;不称为确认占用 |
| 进程参数明确属于目标工程或目标库 | 本项返回 `TARGET_NATIVE_OWNER_ACTIVE` | 不抢占用户或其它任务会话 |
| 目标/共享数据库仍有 `sAddress.adr` | 本项返回地址状态未确认;保留地址原件 | 地址存在不等于活动占用;目前该写入入口仍需原生判断 |
| 目标 LMC、工程/PCB 的 `icdb.dat` 或 Layout 无法独占读取 | 拒绝安装 | 补上实际文件句柄探测,包含共享 `__ReusableBlocksSymb` |
| 源、目标、验证副本内容在复制/验证期间变化 | 本项拒绝覆盖;已证明释放时不制造未释放状态 | 保留原文件与清单 |
| 其它工程的 iCDB 导出器/服务器存在 | 不再全机拒绝导出,也不收养为当前任务后代 | 仅追踪本次后代和命令明确指向本次分析副本的原生进程 |
| 当前导出器或后代仍活跃/句柄未释放 | 保留导出任务及它所属库的排除 | 不杀原生进程;库 A 的已绑定记录不排除 B |
| 创建导出进程遭明确 FileNotFound/Permission/CreateProcess 拒绝 | 返回失败且本任务会话安全;可重试 | `native_started=False` 及 `export-launch=false` 绑定证明 |
| 导出构造异常但可能已经启动原生进程 | 保持未确认,不能因没有返回 Python 对象就放行 | 明确区分未启动和部分启动 |
| 导出前完整进程树查询超时/失败 | 本项失败且明确未启动;释放本任务未启动的 guard,可继续独立项 | 不把查询失败视为空进程列表或数据库无人占用 |
| 已启动导出后的完整进程树查询超时/失败 | 10 秒以内返回未确认,保留本次进程与已记录句柄 | 不无限等待 WMI,也不强杀原生进程 |
| 导出器已退出,但输出错误或不完整 | 内容失败与释放结论分开 | 不因业务失败制造活动会话 |
| 导出后的汇总报告写入失败 | 保留已确认释放;报告写入错误另行返回 | 先完成释放证明及清除本次导出 guard,再写可选汇总 |
| 恢复开始前检查失败或恢复结果报告写入失败 | 保留 `release_confirmed`;没有原生会话则允许其它独立项继续 | 不再统一返回 `safe_to_continue=False` |
| 程序重启后导出 PID 被其它进程复用 | 新记录比对 creation identity 与 executable;不同身份不当成旧 owner | 旧记录没有创建身份时仍需新鲜独立证明 |
| 正常导出 checkpoint 存在 | 重新核对库、lease、token、独立副本、进程身份、地址和数据库句柄 | 不把其它同名服务器当作当前副本 owner |
| 应用 checkpoint 损坏 | 默认拒绝;显式释放入口持有完整库/原生证明时可使用 `allow_unknown_checkpoint` | 此路径仍要求全机已知导出 owner 和 retained registry 均空;调用者必须归档未变化的原字节 |
## 回归验证
首次新增的六项失败测试均先观察到旧实现失败,然后实现修正。领域测试曾串行运行 `81 passed in 62.78s`;后续增加了未知启动、损坏 checkpoint 显式释放、无 PID 证明、报告 I/O 及 PID 复用的正反测试,新增子集 `12 passed`,汇总报告 I/O 正反 `2 passed`。主任务须再进行最终全量运行;多进程并行运行这些测试会争用产品全局 named mutex,不能把相互争用的结果当成业务失败。
全进程树查询现已移入现有 `--owner-probe` 只读 helper,使用版本化 snapshot 协议并保留 PID、parent、未知可执行文件名称、命令、路径和创建身份。上限统一为 10 秒,为 helper 启动及 WMI 查询保留余量;超时只结束读取进程的 helper。回归覆盖启动前与启动后的查询超时、未知后代、PID 0 系统项、协议版本、重复 PID、缺少创建身份等。新增相关子集 `17 passed`。实际本机只读 helper 返回 457 个进程(包括非 `.exe` 系统名称),457 项创建身份,耗时 4.75 秒;该数值是一次观察值,并非上限,没有启动 Mentor。
最后补充了 `capture_identity` 的启动前异常边界:来源校验、当前 lease 证明、证据目录范围、导出器能力检查、目录创建或 mutex 创建失败,均返回明确未启动的安全失败,不把被调用者异常变成活动会话。六项对应的失败测试先观察到旧实现抛异常,再验证修正;与查询预算、已启动失败的正反测试合并为 `11 passed`。这些测试没有取得产品全局 mutex。
最后一次只读复审补齐了进程诊断记录写入边界:已证明导出进程/后代退出且副本数据库已释放后,最终 `*-process.json` 写入失败只附加报告警告,不能抹去释放事实;仍有活跃原生进程时保持未确认。首个 `export-launch.json` 在任何启动前写入失败,则返回明确未启动;仅凭本进程当前 lease、匹配的一次性 token、独立副本 owner 和零 process records,清除此任务尚未启动的 guard。缺少日志不能由重启后的无身份记录借用此通路。三项报告异常回归先失败,修正后与未释放正反测试合计 `7 passed`,未使用产品全局 mutex。
## 仍需明确保留的验收缺口
- 没有在真实 Mentor 环境验证本轮恢复/导出功能。上述证据为回归测试及只读代码审查。
- 恢复写入入口中的旧地址,需要原生 inspection 确认;本次没有自动删地址,也没有仅凭文件不存在就确认释放。
- 无工程参数的外部原生会话仍是未确认;不能自动 Attach 到全局 COM 并操作他人的文档。
- 当前没有解决所有原生内容失败;本轮不表示 ICDB 自动修复或现场问题验收成功。
@@ -0,0 +1,124 @@
# ICDB 判定接口与本次检查失败的只读调查
日期:2026-11-05。基于 `1acfbd6`。本轮新增离线元数据探测脚本,读取安装的
类型库、PE 导出和既有日志;没有改变应用判定、打开工程、调用 Repair/Restore、
删除地址文件或终止原生进程。自动修复仍未验收通过。
## 结论
“打不开 PCB 并出现弹窗”是有价值的直接观察,但一般数据库打开失败、服务器
连接失败、版本恢复选择,都不能单独等同于 ICDB 数据损坏。现阶段最可靠的
ICDB 维护依据,是**本次打开的同一工程/数据库发出的明确原生诊断**。
尚未找到已验证、无需打开数据库、能直接预测 PCB 将发生 ICDB 故障的独立接口。
底层确有原生错误类别和错误码,但只有导出名称,没有取得对应 SDK 声明和
完整枚举,也没有建立真实故障与数字的映射;不能猜测参数调用后接入产品。
## 1. 已解析的 PCB 原生类型库
前一轮探测对 PCB 启动器执行 LoadTypeLib 失败,并不意味着安装缺少类型库。
本轮按 32 位注册表定位到:
```text
HKCR\TypeLib\{EDEDED00-D5F6-4B04-8FE7-EDEDEDEDED00}\1.0\0\win32
D:\Mentor\7.9.5EE\SDD_HOME\wg\win32\bin\XtremeDesignSessionWG.exe
```
成功读取 `MGCPCB (AutoActive series) Type Library`,共 614 个类型。没有创建
COM 实例,也没有连接正在运行的 PCB。
| 元数据 | 实际声明 | 能证明什么 |
| --- | --- | --- |
| `IMGCPCBApplication.OpenDocument(filename)` | 返回 Document,没有 ICDB 专用返回码声明,`scodeArray` 为空 | 可作为实际打开尝试;不能据声明直接解码 ICDB 故障 |
| `ActiveDocument` | 返回当前 Document | 空文档只表示尚未取得文档,不说明失败原因 |
| `IMGCPCBError` | `ErrorString`、`Severity`,没有 ErrorCode 属性 | 提供文字/严重级别;不是统一的打开错误码接口 |
| `EPcbErrorSeverity` | `epcbWarning=0`、`epcbError=1`、`epcbFatalError=2` | 是错误严重级别,不是三种 ICDB 健康状态 |
| `Errors` 集合使用方 | CopperBalancing、DFFSliverAnalysis、IDFImport、DXFImport | 不能把这些集合当成 PCB 启动错误队列 |
| 应用事件 | 打开前/打开后事件;`OnOpenDocument` 标志为 Open=1、Reload=2 | 没有声明 ICDB 打开失败事件和专用数字码 |
类型库没有专用错误表,不代表运行时绝不会返回 HRESULT/EXCEPINFO。真正执行
调用时仍应记录 HRESULT、EXCEPINFO 的 wCode/scode/source/description 和调用阶段,
但尚未实测的 HRESULT 不能赋予“数据库损坏”的含义。
## 2. 底层 iCDB 候选接口
静态解析 `iCDBPI.dll`、`iCDBDiagPI.dll`、`iCDBCAPI.dll`,三者均为 x86 PE。
`diagnostics/icdb-authority-20261105/metadata.json` 保存实际文件 SHA-256、
原始符号名称、类型库说明及常量,避免跨版本混用。
| 接口线索 | 已证实内容 | 未证实内容 |
| --- | --- | --- |
| `Result_get_iErrorCategory` / `Result_get_iErrorCode` / `Result_get_sResult` | 原生 Result 有错误类别、编号、文字 | 对象布局、调用约定、错误枚举及故障映射 |
| `CDBPI_IsOpenCDB` / `IsOpenCDBEx` | 存在数据库打开状态查询符号 | 未打开、不可达、损坏之间如何区分;不是已验证的健康检查 |
| `PIMgr_GetStatus` | C++ 对应声明返回/填写 `ESynchStatus` | 不能把同步状态当 ICDB 健康状态 |
| `Diag.GetProjectStatus` / `GetActiveIssues` / `GetStatus` | 存在诊断、工作区及项目状态接口 | 结构体、枚举、初始化方式、查询范围及副作用 |
| `Diagnostic.Check` / `CDBPI.IsDatabaseAvailable` | 存在相关诊断符号 | 参数语义、是否只读、超时/释放行为;未执行 |
导出名的存在不证明这是厂商支持的外部自动化 API。不能用 ctypes 默认 int
返回类型和猜测指针去调用这些接口;尤其不能在主界面线程内调用。
## 3. 本次“全部失败”的具体证据
旧记录:`diagnostics/live-open-20261105/evidence/icdb-check-1738989c1dae/`,
原生 PID `63480`,目标为 IP101GR 的 `Work/database`。
这次补读该 PID 的原生进程日志:
```text
D:\Mentor\7.9.5EE\SDD_HOME\standard\iCDB\ProjectBackup\Logs\
2026-11-05 08.08.40-63480\2026.11.05.iCDBProjectBackup.txt
```
日志从 08:08:41 读取该数据库的 ServerProperties,依次尝试所记录的多个地址
和主机名,端口 10002。08:09:14 报 `Network: ERROR: Could not connect to IP [...]`
后,再次读取同一数据库的服务器属性并继续重试。调查时读到的后续日志中仍有
相同循环。摘录和来源保存在 `diagnostics/icdb-authority-20261105/network-evidence.json`。
因此,这一次没有取得 Output 的直接线索是**原生工具正在反复连接数据库记录的
服务器地址**,不是仅仅窗口标题不匹配。此前允许通用标题的修改不足以解决它。
尚不能仅凭这些日志确定是服务器已退出、地址过时、监听配置或防火墙问题,
也不能证明数据库内容损坏,更不能据此确认目标无人占用。
同批另两项缺少唯一 PRJ,不能把它们与此连接失败合并成一个损坏结论。
本轮没有再次启动原库检查,也没有强制结束 PID 60000/63480。
## 4. 检测规则的遗漏与证据等级
安装 DLL 包含以下明确的原生维护消息:失联、整理中断、升级中断、重要操作
中断、跨机失联、自动修复中断,均带厂商的 Repair project 指引。
把这六段原生模板传给当前 `parse_session_loss`,只有跨机失联模板被识别。
这确认了规则覆盖范围偏窄;**模板存在不证明此次工程触发了这些错误**。
| 本轮实际证据 | 合理结论 |
| --- | --- |
| 同一数据库的完整失联/不一致/检查失败原生诊断 | 明确 ICDB 异常,记录子类、数据库路径、会话及证据 |
| 服务器地址重试、连接失败、一般 Could not run server | 连接/启动失败,尚不能认定损坏或无人占用 |
| `problem encountered while opening ...` / `loading ... failed` | 一般打开失败,还需具体数据库日志/弹窗 |
| Database Restoration 的手动/自动保存版本选项 | 需要用户选择版本,不自动判为损坏或代选 |
| Project Backup 未占用、备份目录已加载 | 占用/备份信息,不保证 PCB 完整打开或设计健康 |
| 无 Output、超时、未知错误 | 无法确认,不能推成健康或需要 Repair |
| READY/PENDING、地址文件、历史日志 | 均不能单独决定本次 ICDB 是否有问题 |
## 5. 建议的接入顺序与验证缺口
普通检验继续实际打开工程;同步捕获本次弹窗、增量日志和同一工程身份。
打开成功就继续普通检验,不以 Project Backup 预检作为其前置门槛;打开失败
仅在有明确 ICDB 原生诊断时标注 ICDB 子类,其他错误分别记录。检查结果不应
禁用普通检验或原生版本管理,也不据单项诊断设置永久的全库失败标志。
独立批量检查仍只做用户明确启动的检查。原生日志补采应绑定启动时间/PID、
目标路径和读取预算,保持增量及有界;网络日志中的“连接失败”不等同会话释放。
如果新检查不能退出,应显示实际滞留目标和原生连接阶段,不能靠伪造释放证明
继续对同一目标执行写操作。
下一轮动态探测应使用完整隔离库,分别记录正常打开、会话失联、文件损坏、
连接不可达、版本恢复选择的 PCB/原生证据;只运行诊断打开,不执行 Repair/Restore。
提取真实 HRESULT/EXCEPINFO 与原生文字的对应关系,并验证取消与正常关闭。
即使没有数字码,已确认的原生消息仍可用于分类;不能为等一个状态码阻塞普通流程。
本轮只验证元数据提取、既有日志关联和解析器覆盖范围,没有执行新的 PCB 打开
对照,没有验证底层 C API、解决原生网络重试或修复 ICDB。没有生成新应用候选包。
后续经用户授权完成了真实损坏/正常/缺文件对照,见
[损坏弹窗与错误码实测](2026-11-05-icdb-corruption-error-code-probe.md)。取得了 PCB
COM 两层错误码,但损坏与缺文件返回相同值,仍不能把该码当 ICDB 健康状态。
@@ -0,0 +1,102 @@
# ICDB 损坏弹窗与错误码实测
2026-11-05,Mentor EE7.9.5,PCB `2013.0.543.380`。按用户授权,只在新建的
`C:\Temp\CbbCodeProbe-7f1b0307` 内制造损坏。应用判定代码没有修改,没有执行
Repair、Restore、版本回退、集成或保存。
## 实测结论
成功触发数据库打开失败弹窗,并取得 `OpenDocument` 的 HRESULT 和 EXCEPINFO。
但**数据库损坏与文件不存在返回完全相同的两层错误码**,因此这两个数字不能
单独用于判断 ICDB 异常,更不能单凭它们自动请求 Repair。
| 对照 | `OpenDocument` 结果 | 外层 HRESULT | 内层 EXCEPINFO.scode | 原生证据 |
| --- | --- | --- | --- | --- |
| 未修改副本 | 返回正确工程文档 | 没有失败异常 | 不适用 | 工程身份匹配,数据库 GUARD OK,连接成功 |
| 损坏 `icdb.dat` 前 64 字节 | 失败,ActiveDocument 为空 | `0x80020009` / `-2147352567` | `0x8004021E` / `-2147220962` | 数据库文件打开失败弹窗;本轮日志明确指出损坏副本的数据库格式不受支持,并建议 Repair |
| 不存在的 PCB 路径 | 失败,ActiveDocument 为空 | `0x80020009` / `-2147352567` | `0x8004021E` / `-2147220962` | 文件不存在弹窗,没有数据库损坏依据 |
两次失败的 EXCEPINFO 全部字段相同:
```text
wCode: 0
source: Expedition PCB - Reusable Block
description: Failed to open document.
helpFile: ...\hssd\win32\bin\mgcpcbautomation.hlp
helpContext: 10187
scode: -2147220962 (0x8004021E)
```
外层 `0x80020009` 是 COM 调度异常。内层码在本次对照中表示通用的
`Failed to open document.`,尚不能赋予它 ICDB 专用含义。这里取得的是 PCB
自动化接口的错误,**没有调用未验证的 iCDB C API,也没有取得其内部 Result
错误枚举的可靠映射**。
正常、损坏、缺文件三种情况在等待 COM 注册时,都出现过
`0x800401E3` / `-2147221021`。正常副本随后成功打开,故此码也不能视为 ICDB
故障;它记录的是当时 COM 对象尚不可取得。
## 复现条件与证据
种子来自已有的隔离测试库:
`diagnostics/icdb-auto-recovery-20261027/verify-library`,不使用生产库作破坏对象。
完整复制得到 control 和 damaged 两份库,每份 21,029 个文件、557,187,955 字节。
复制后核对 LMC、PRJ、DXP、icdb.dat、PCB、LayoutDB 六个主要文件的 SHA-256。
仅在 damaged 副本清零工程 `database/icdb.dat` 的前 64 字节,其余字节保留。
正常副本和损坏副本在启动前彼此独立。
通过同一个原生入口启动空白 Reusable Block PCB,然后在专用私有桌面执行
`app.OpenDocument(绝对PCB路径)`。调用前核对 COM 主窗口 PID 的启动器父子关系,
记录 InstanceGuid,确认没有已有文档,且 PRJ/PCB 工程配置的 CentralLibrary
均解析到该副本自己的 LMC。没有创建生产库 COM 写会话。
正常对照 PID 15536,10:16:44 开始 OpenDocument,10:17:06 返回正确文档。
随后授权匹配的关闭助手执行 `doc.Close False`,未保存,正常退出。
损坏对照 PID 63384,10:18:10 开始 OpenDocument,10:18:51 取得异常。
弹窗原文:
```text
Problem encountered while opening one of the database files.
See PCBLogFile.txt for details. Loading of the job has failed.
Do you wish to view the log file now?
```
截图:
`diagnostics/icdb-authority-20261105/live-damaged/database-failure-75566502.png`。
先记录弹窗,再点击“否”关闭可选的查看日志提示,未发送修复指令。
本轮 `PCBLogFile.txt` 增量为:
```text
Database cannot be loaded and the server cannot be started, the project will
not be opened. Please use iCDB Project Backup 'Repair project' functionality
to fix this issue. (Database [C:\Temp\CbbCodeProbe-7f1b0307\damaged\
ReusableBlocksLP\IP101GR\Work\database\icdb.dat] in old, unsupported format)
```
“旧格式”是原生对被清零文件头的诊断文字;本试验没有证明实际存在版本过旧。
iCDBManager 同一 PID 的日志也指向该损坏副本。原生层先重复读取损坏数据库后
才显示弹窗;COM 异常在确认弹窗后返回,故不能只等 COM 异常而省略限时的
独立弹窗/日志观察。
缺文件对照最终 PID 61776,记录为 `live-missing-3174b7/com-result.json`,
10:25:08 发起调用,10:25:10 取得异常。首轮缺文件试验的单按钮提示未被助手
识别,助手按 180 秒上限停止,原生会话保留;之后核对窗口 PID 和明确的缺文件
文字,确认原生按钮 ID 为 2,正常确认提示并关闭空白 PCB。修正试验助手后重跑
取得上述完整错误码。该超时不是 ICDB 故障证据,首轮记录保留。
## 验证结果与边界
`diagnostics/icdb-authority-20261105/verify_live_probe.py` 对真实记录执行断言,
汇总保存在 `live-comparison.json`:正常文档身份、两种失败完全相同的 EXCEPINFO、
损坏弹窗与数据库路径、来源六文件哈希、损坏文件未被修复、测试进程退出及
已有原生会话保留均核对通过。三个最终试验都正常结束,前台窗口句柄前后相同。
没有剩余测试 PCB 或 iCDBNetServer;既有 Project Backup PID 60000、63480 保留。
检测可使用“实际打开失败 + 本次明确 ICDB 诊断 + 数据库目标一致”的组合证据。
文件不存在、一般打开失败和 COM 注册等待要分别记录。明确的单项故障只应
成为该项备注,不据此永久禁用普通检验或版本管理。
本次验证了**文件头损坏**,没有复现现场“先前服务器会话失联”,也没有验证自动
修复、READY 或完整产品维护入口。没有运行/宣称全量应用测试;本轮只增加
调查脚本与记录。不能将本次复现算作现场 ICDB 修复通过。
@@ -0,0 +1,86 @@
# ICDB 扫描按实际打开错误内容分类
## 当前行为
批量检查复用 PCB 诊断打开路径,不再先启动 Project Backup 检查所有条目。
诊断请求固定关闭自动修复、原生 Verify 菜单及恢复确认;只打开、核对目标并
正常关闭,不执行集成、保存、Repair 或 Restore。普通检验和原生版本管理
继续使用独立入口,界面既有 READY 排除规则保留。
只有实际打开错误中包含 iCDB 内容,或本轮原生日志明确给出 iCDB 错误,才
将该条目标为 ICDB 问题。通用错误码、工具内部类别名称、路径中出现 icdb、
成功输出以及历史错误均不能单独决定分类。没有 iCDB 具体原因的一般打开
失败仍保留原始错误和未确认结果,不显示成数据库损坏或健康。
正常打开只返回 CHECKED,不推定 CBB READY。单项失败在实际会话已释放后
继续后续项;未确认释放或用户取消时仍保留真实状态。没有增加永久的故障
锁,也没有把扫描结果作为普通检验或版本管理的前置条件。
## 重复扫描的证据
实机发现同一工程连续打开失败时,PCBLogFile.txt 可能逐字相同。仅比较内容
会把本次故障误当成旧日志。补充采集当前 PCB PID 对应的 iCDBManager 日志,
核对日志创建/修改时间、PID 头和目标数据库路径后,保留具体 ERROR 内容。
采集在既有的限时只读助手中执行,保留 16 KiB 头和最多 128 KiB 尾的读取
预算,不在 Qt 线程中等待日志,不递归扫描整个库,也不调用修复接口。
文件日志仍采用启动前检查点及有界增量读取;未知基线、无法确认历史的日志
重写,不提升为本轮证据。扫描只读取本次诊断子目录的错误证据,不能引用
上一次操作的证据目录来标记当前条目。
## 实际验证
使用上一轮完整复制的两个隔离库,通过应用的 `run_verify(maintenance=True)`
及 `run_batch` 执行真实生产调度路径。请求故意带旧 `auto_repair=True`,内部
诊断仍强制关闭修复。
- 损坏副本:保留清零前 64 字节的 icdb.dat。原生再次弹数据库打开失败提示;
扫描读取本轮 PID 3060 的日志,返回 FAIL / ICDB_REPAIR_REQUIRED,并保留
`in old, unsupported format`、目标数据库路径及日志证据。
- 随后正常副本:原生 PID 28948 打开正确工程,正常关闭,返回 CHECKED。
- 两项均 `safe_to_continue=true`、`library_flag=UNKNOWN`、恢复尝试为空;没有
请求 Project Backup、Repair、Restore、集成或保存。测试 PCB 和数据库服务器
均退出;没有终止已有 Project Backup 会话。
证据:`diagnostics/icdb-authority-20261105/product-scan-comparison.json`,以及
相邻的 `product-scan-damaged` 和 `product-scan-control`。首次复测暴露的重复日志
漏判记录也保留,没有清空故障日志来让测试通过。
回归覆盖实际损坏报文、未知 iCDB 错误内容、相同错误码的非 ICDB 失败、正常
输出、旧日志残留、日志重写、旧操作证据隔离、当前/其他 PID、日志新旧、
其他数据库路径、取消、会话未释放、失败后继续,以及原有界面/报告行为。
最终测试命令为 `python -X utf8 -m pytest -q`,1326 项通过,436.32 秒,保留一条
既有 pywinauto STA 提示;详细结果保存在同目录 `pytest-scan-utf8.xml`。
系统默认 GBK 模式的一次运行中,既有测试用默认编码读 UTF-8 报告失败;该项
在 UTF-8 模式下单独复测通过,未因此修改原生修复逻辑。
冻结候选包构建及启动/正常退出冒烟通过,16 个关键 Python 模块和两个 VBS
与源码一致,包内 x86 运行库及 Unicode/UNC 协议检查通过。使用冻结程序的
限时日志采集入口回放当前 PID 证据,确认能保留实际 ICDB 原因;未在冻结包中
再次执行整条原生扫描批次。独立候选包不覆盖此前的程序目录,未合并发布。
只读独立复核未发现必要 P1/P2,指出批量报告残留旧的 Project Backup 流程名称。
随后仅将该报告名称改为实际的 PCB 打开检查,并增加报告输出断言,104 项相关
回归通过;同步重建并复核最终 r2 候选包。完整回归记录保留在上述 XML,文案
修正后的相关回归记录为 `pytest-scan-post-review.xml`。
## 文件头损坏与版本管理
“版本历史与备份…”打开原生 iCDB Project Backup,入口只核对 LMC 和唯一 PRJ,
不以 icdb.dat 文件头可读作为准入条件。有效的完整历史备份包含数据库文件,
原生 Restore 可以替换损坏状态,而不是计算或补写丢失的 64 字节。
这份隔离样本的备份配置为相对 `./backups`,列出的多个历史 ZIP 中包含
`database/icdb.dat`。文件头非零仅是清单检查,不能代替原生验证备份有效性。
清单保存在 `version-backup-inventory.json`。
已有同类隔离实测中,原生 Repair 从已有备份恢复了该文件头损坏,随后重新
打开并验证成功;没有有效备份的对照则明确失败。见
[此前有备份的恢复记录](../releases/2026-10-28-icdb-open-repair-reopen-validation.md)
及[无备份的失败记录](../releases/2026-10-28-icdb-no-history-repair-validation.md)。
那次是 Repair 调用原生备份恢复,不等同于本轮已通过当前版本管理按钮执行
Restore;本轮没有选择恢复版本或发起恢复。
恢复会回到所选备份的内容版本,应保留原始损坏副本,恢复后重新打开并单独
检验。PCB 启动时“最后手动保存/自动保存”的选择不应直接视作工程 iCDB 已
恢复。没有有效恢复源时不能承诺恢复。现场会话失联和自动修复仍未验收通过。
@@ -0,0 +1,58 @@
# 管理、协调租约与状态扫描阻塞审计
审计范围:`library_lock.py`、`management.py`、`native_management.py`、`management_transaction.py`、`native_backup.py`、`network_path.py`、`batch_status.py`、`status_bridge.py`、`owner_probe.py`、`native_export_guard.py`、`icdb_ownership.py`、`native_occupancy.py`。本文仅说明这些边界,原生窗口清理、普通校验入口与界面由本轮其他修改覆盖。
本轮没有在该审计任务中启动 Mentor、Repair、Restore,未修改现场中心库,也没有删除原生地址、锁或数据库文件。测试使用临时文件和替代原生边界。
## 已落实的误阻塞修正
| 入口或条件 | 原误阻塞 | 修正及验证重点 |
| --- | --- | --- |
| `native_management.NativeOperationError`、`management.run_management.invoke` | 原生目录失败已退出,异常丢失释放事实 | 操作成功与释放证据独立传播;已退出目录失败可以重新取得协调锁 |
| `management_worker_main` | 桌面、映射、能力或来源校验在原生启动前失败,却报告未释放 | 原生启动前保持 safe;进入实际原生边界前才变为未确认 |
| `NativeManagementAdapter._worker`、`_launch_worker` | 请求准备或桌面构造失败制造假会话 | 准备失败为 safe;构造失败仅消费明确的 process_created=false、worker_exited=true、remaining_pids=[]、safe=true 协议;部分启动和缺证据仍未确认 |
| `NativeManagementAdapter._add_transaction` | Add 已退出后,路径、引用对齐、被动占用检查失败再次保留库锁 | 仅实际启动下一原生调用前重置释放状态;异常在租约退出前传播已知释放事实 |
| `NativeManagementAdapter.prepare_add_transaction` | 来源缺 CentralLibrary、staged/asset/lease 校验失败制造假未释放 | 原生 capture 前的完整准备块返回 safe=true 失败;capture 后准备事务记录失败保留 capture 的实际释放事实 |
| `management_worker_main` 的 CleanWGLib remove | wait 已退出但输出文件读取失败,或明确创建拒绝仍 unsafe | wait 确认退出立即记录 safe=true 与 exit_code;FileNotFoundError/PermissionError 的创建拒绝为 safe,未知创建错误和 timeout 保持未确认 |
| `management.run_management` | 移除后被动检查异常、报告写入异常把已释放改成未释放 | 保留释放事实;报告存储错误写入返回 detail;不重复抛出最终报告异常 |
| `management.RECOVERY_REQUIRED` | 目标数据恢复要求被混同为原生会话仍存活 | 可同时返回 RECOVERY_REQUIRED 和 safe=true;同目标禁止自动重放,明确不同目标不被旧数据恢复记录误挡 |
| `management._may_have_mutated` | 末阶段看似 prewrite 即忽略曾经写入的 history | 仅完整且一致的 prewrite history 可确认未写入;缺失、混合历史保持未知 |
| `LibraryLock.__enter__` | retained checkpoint 租约永远不再检查回调 | 同一租约 release probe 证明退出后,可清理身份相符且无未决 export/pending 的应用检查点并重新取得锁 |
| `NativeManagementAdapter.release_confirmed` | 超时 worker 后来正常返回仍永久保留 | 按库跟踪原 worker;只有它已退出且其独立结果 safe=true 才回收;缺失/损坏/unsafe 结果不推定释放 |
| `LibraryLock`、`release_retained_library` 的 CreateFile | 权限、网络、路径失败统一称 competing writer | 仅 Windows sharing/lock violation 32/33 分类为 LibraryBusy,其余保留实际错误 |
| `has_blocker` | 完整快照中的异库 pending 一律阻止 | 主记录与 pending 采用一致规则:只有完整、有效且明确搬迁的源记录可忽略 |
| `release_retained_library` | torn、malformed 应用检查点没有恢复路线 | 显式释放强制全库 fresh owner/native-file/export proof,复核原字节未变化,先归档再清应用 JSON;未知 proof 标记不足时拒绝归档 |
| `release_retained_library` 的范围 | 选择 Target 却清除全库旧检查点,可能漏 Other 真实会话 | 只有检查点和进程内 owner 都绑定相同 block_name 才使用所选目标范围;旧无 scope、范围不符、损坏记录一律 full proof |
| `_prove_library_released` 的目录范围 | 目标释放仍扫描所有 CBB,且漏共享 SymbolLibs | 已绑定目标时仅扫描目标 Work、共享 SymbolLibs 和所选 LMC;共享地址与数据库始终参与证明 |
| `_prove_library_released` 的 retained registry | 库 A 的 worker 阻止库 B | 新 worker 绑定 cbb_library;明确异库可排除;旧无 scope registry 保留未确认 |
| `_owner_matches_scope` | Work 子串匹配 WorkOld,worker 只带 job 路径时反而漏检 | Windows argv 精确路径边界;读取有大小限制的 worker 请求以绑定 library/project/database;没有可判定路径时不推定无关 |
| `_owner_matches_scope` 的 EDIF 命令 | 不识别产品实际 `-icdb <database>`,异库 exporter 被当全局未知 owner | 同时支持实际位置参数与 `-icdb=`、`-project=`,带空格路径和大小写按 Windows argv 解析,目录前缀相似不当作同库 |
| full known library process proof | 全机同名原生程序都被称本库占用 | 命令行明确指向异库可排除;无范围证据仍保持未确认;损坏检查点 full recovery 不使用范围排除 |
| `_persist_blocker`、`_recorded_supervisor_alive` | maintenance 只凭旧 PID 存在,PID 复用会误挡 | 新 version=1 记录附 machine 和 supervisor_created;只有进程创建身份匹配才称原监督进程仍在;旧身份缺失需额外 fresh native-idle 和原独占/目标 guards |
| maintenance 新 owner | retained 后缺 release_probe | 保存明确的 release callback,复用协调租约的再证明路径 |
| `BatchStatusSession.scan` | 单个请求/CBB/path mismatch 把后续未查询项全部标错 | 错配记录只属于当前项;支持 server 时先 restart 再逐项查询;已确认基础设施不可用仍批量标 UNAVAILABLE |
| `fingerprint_library`、batch record | 相同字节/时间的另一库或另一 block_type 复用旧 READY/PENDING | cache 绑定绝对库路径和 block_type;不 resolve 映射路径,不递归扫描工程负载 |
| `LibraryPathLease.close` | 映射查询或取消一次失败后 closed 永久为 true | 成功后才结束租约,瞬时失败可再次 close |
| `release_after_recovery`、`release_retained_path_leases` | retained mapping 永久不释放,旧界面丢引用后也不可回收 | 核对 RELEASE_CONFIRMED 与该库 display/mentor/已知 UNC 身份;registry 保留旧对象;共享当前映射先转移所有权再释放旧租约 |
| `WindowsWNetBackend.mapping`、`mappings` | 无关盘符读取失败让目标映射不可用 | 目标盘直接查询;全枚举跳过不可查盘但把它保留为已占用,避免抢占未知盘符 |
对应集中回归:`tests/test_management_retry_blockers.py`。既有管理、事务、租约、网络路径、状态扫描、导出与维护测试也纳入定向回归。
## 仍保留的真实互斥和未确认边界
| 条件 | 当前行为与可用路线 |
| --- | --- |
| uploader 或真实同库任务仍持有独占协调句柄 | 继续互斥;不删除 `.cbb-upload.lock`,文件中的旧 PID 文本不作为当前持锁证据 |
| 全局 `Local\\CbbRefreshVerify795` mutex | 仍序列化本应用原生 COM/管理操作。尚未证明所有 Mentor 全局状态可按库并发,未直接删掉此保护 |
| `native_management._catalog_com` 已有 LibraryManager.exe | 仍在 COM 激活前保守拒绝。尚无证据证明本机 DispatchEx 不会复用已有服务、Quit 不会关闭他人窗口;本轮仅修为 prelaunch safe,不声称已支持并存 |
| `_catalog_native` 激活后无 native_pids、原生 PID/退出状态缺失 | 仍未确认,可能 COM 已启动不可追踪服务。新请求不自动把空 PID 列表视为已释放;提供显式 fresh release 路线 |
| `native_management._wait_native_exit` | 运行时仍以已观察 PID 等待退出,未新增 Library Manager 全部后代的 JobObject/creation-time 追踪;超时转显式释放证明 |
| 旧无创建身份的外部/导出进程记录 | 不能把活 PID 直接证明是旧进程,也不能无条件忽略;新记录创建身份由本轮相应 exporter 修改提供。无法独立确认时保留现场与明确未确认原因 |
| `sAddress.adr` 仍存在 | 地址存在本身不证明当前活跃,但显式“释放”也不冒充原生 NOT_IN_USE。应走既有 inspection-only 的精确 project/database 原生占用检查;本任务不手删地址,也不自动 Repair/Restore |
| `owner_probe` 超时、访问受限、空 command、请求文件不可读 | 不把它称已确认活跃;也不猜已退出。明确异库证据可以排除;范围不明时保留未确认,并允许独立只读检查 |
| export ownership 元数据有效但副本目录/数据库缺失,或 token/lease 不匹配 | 仍拒绝自动清理,因为无法证明原副本的原生写入已结束;未知元数据恢复通路不能覆盖掉已知但尚未证明释放的 export copy |
| native backup 的自身 JobObject/子进程仍活动或查询失败 | `native_backup.py` 已使用 checkpoint=false;等待其自身会话树退出,不因主窗口/launcher 先退就推定全部释放,不 terminate/kill 原生事务 |
| 二进制 hash、flow、ASCII removal contract、来源/目标/证据隔离、Add nonce 和 duplicated lease handle | 继续作为能力或身份校验;它们不是 native busy。失败不授权绕过未验证原生写入路径 |
| 损坏检查点的全局 fresh-proof 路线 | 为弥补未知 scope,可能要求全机已知 native/app workers 均退出;这是显式未知记录恢复的限制,不适用于具有明确异库范围的普通重试 |
维护入口的 PID 创建身份、原生文件独占和 fresh guards 是不同证据:创建身份匹配只说明原监督进程仍在;创建身份不同/旧 schema 缺失只消除错误 PID 归属判断,不能替代原生或数据库释放证明。
@@ -0,0 +1,37 @@
# 中心库无活跃会话却被阻止(r5 现场反馈)
## 现场证据
用户运行 `blocking-locks-20261105-candidate-r5` 后反馈中心库操作仍提示上次未安全释放,扫描看不到活跃会话。
- r5 `data/logs/application.log` 中,15:54(北京时间)的 IP101GR PCB 检查因原生读写确认退出;相应 `session-release.json` 为 `safe_to_continue=true`、`worker_exited=true`、`remaining_pids=[]`。这次打开失败不等于释放失败。
- 随后版本工具 `native-backup-6bc7146a0eee/result.json` 为 `CLOSED / safe_to_continue=true / native_exit_code=0`,原生 PID60660 已退出。
- 中心库仍存在 11 月 4 日的 `.mentor_lib.cbb-upload.blocked.json`:PID57076、lease `eae80a06f23e4664b00c70b6bf81b01d`、`OWNER_RELEASE_UNCONFIRMED`。这是一条旧操作记录,并非当前进程存在的证据。
- 本轮只读实查:关联进程列表为空,协调文件可独占读取,库内没有 `sAddress.adr`,原有完整释放证明也通过。未为诊断删除源库任何记录或工程文件。
## 根因
`run_management` 在 `LibraryLock.__enter__` 因历史记录拒绝时尚未启动任何原生会话,但异常收尾用 `isinstance(error, LibraryBusy)` 强制把结果改成 unsafe。互斥通道尚忙也返回相同 unsafe。UI 随后把这一结果记为按库保留的 unsafe 标志,直接禁用后续管理;此时没有 PID 可供扫描。失败发生在创建本次 checkpoint 之前,也没有对应的管理尝试日志。
原残留扫描只列进程,不列历史 JSON 或 `checkpoint=False` 的进程内 retained lease,进一步造成“明明没有会话但永远不能继续”的表现。
## 修正
1. 有旧 JSON、同库进程内 retained lease、旧管理标志或未确认释放的历史修改时,在操作互斥范围内、任何新原生调用前执行已有完整释放证明。证明通过才归档本应用旧记录并释放本应用句柄;真实 owner、地址、共享访问冲突仍拒绝。未知归属不假定无人占用。
2. 管理尝试先记录 `release_check`;未启动的新操作被拒绝时保留本次 `safe=true`,不制造新的原生 owner。已启动原生后仍沿用实际退出证据,不能把超时统一放行。
3. UI 的旧 unsafe 标志成为重新复核要求,不再直接禁用管理。匹配当前库的证明且本批结果全部安全后,先恢复 retained 路径 lease,再清旧标志。映射恢复失败保留重试状态。
4. 扫描明确显示历史记录日期与内存保留记录,和活跃进程分开。空进程列表不再隐藏历史阻止原因。
5. 同目标历史修改未完成仍保留“需恢复,不能重放”;会话已释放与数据是否恢复分别处理。没有修改的 BLOCKED 结果不再触发额外原生目录读取。
6. 管理结果记录操作类型、CBB、状态、安全状态、原因及证据位置,包含启动前拒绝。
## 验证
- 按现场旧记录结构新增回归,初次 4 项均失败;修正后通过。覆盖无 owner 继续、真实地址保留、执行通道忙不制造 owner、空列表显示历史记录。
- 后续覆盖无 JSON 的 retained Verify 锁、匹配证明后恢复路径 lease、UI 可以再次发起复核。相关 **119 passed**。
- 实际隔离 UI 链路 `diagnostics/icdb-authority-20261105/legacy-management-a1c609ec/result.json`:从已验证原生 CRUD fixture 创建新隔离副本,预置相同旧记录与 UI unsafe 状态,通过应用“移除 CBB”确认入口完成复核、旧记录原字节归档、原生移除 IP101GR、重新读取目录确认缺失、UI 与路径 lease 恢复。**REMOVED / safe=true,59.016 秒,原始 fixture 全文件哈希不变**。生产库工程与旧记录未改动。
- 统一全量 **1518 passed**(`management-release-pytest.xml`,268.10 秒,1 条既有 STA warning);编译与 diff 检查通过。
- **r6 冻结复验通过**:`legacy-management-b3a65315/result.json`。控制层源码与包内字节码一致,真实管理 worker 与进程探测运行 r6 exe;相同 UI 链路返回 `REMOVED / safe=true`,原生目录再次确认缺失,旧记录归档、保留映射与 UI 标志恢复,原 fixture 未修改;96.313 秒。包内 26 个 Python 模块及 2 个 VBS、实际 GUI 启停、x86 codec 和包内 runtime 加载通过。
- 独立复核确认没有新增 P1/P2。候选包为 `dist/blocking-locks-20261105-candidate-r6/CbbRefreshVerify-Background`。
生产库的旧记录保留到用户下一次实际管理操作或确认释放时,由同一完整证明流程复核、归档。没有直接删除源库锁文件,也没有替用户重放增删改。不把本次阻塞修复算作 ICDB 数据自动修复验收。
@@ -0,0 +1,27 @@
# r9 补充审查:4 个已复现问题
本次只审查和构造回归复现,未修改产品源码、未启动 Mentor、未操作真实库。源码与 r9 BUILD.json 的 SHA-256 比对另行确认。先前 1,564 项测试通过没有覆盖以下组合场景。
复现:`python -X utf8 -m pytest -q diagnostics/icdb-authority-20261105/test_r9_remaining_review.py --junitxml=diagnostics/icdb-authority-20261105/r9-review.xml`。四项均在预期断言处失败,收尾完成,无残留测试线程。
1. **P2:停止整队列不会取消后台检查。** 后台 Beta 检查期间提交前台任务,再点击“取消当前任务并停止队列”。`request_stop()` 在 busy=true 时只设置前台 stop,没有设置 IcdbCheckTask.stop。前台任务取消后后台继续执行 Gamma。实测 calls=['Beta', 'Gamma'],期望仅 ['Beta']。位置:`ui/main_window.py:1484`。
2. **P2:排队项目变成 READY 后仍重复检查。** 检查启动时捕获目标列表,后续权威刷新将 Gamma 更新为 READY,恢复执行时只检查 invalidated,未重新核对 READY 操作资格。实测 Gamma 仍进入 verifier。用户要求 READY 即使被勾选也应排除。位置:`ui/icdb_check_task.py:66`、`ui/main_window.py:781`。
3. **P2:释放失败被误显示成读写确认。** 注入 `DATABASE_SAFETY_STOP` 和原始错误 `PCB_RELEASE_UNCONFIRMED: launcher still running`,表格固定显示“未确认:需处理读写确认”。该分类也包含普通会话释放失败,不能据此断定存在读写提示。位置:`ui/main_window.py:819`。
4. **P3:普通校验排队时隐藏后台检查进度。** 检查 Beta 仍运行时提交普通 Gamma 校验;`_run(maintenance=False)` 无条件隐藏 icdb_progress 和 icdb_phase,后台任务仍活跃。实测 isHidden=true。位置:`ui/main_window.py:1215`。
建议先修 1、2,再修 3、4;把这四个复现并入正式 tests 后复测。没有依据将这些结果扩大成数据库损坏或“所有释放又失败”。
仍保留此前验收缺口:ICDB 自动修复、现场服务器失联完整恢复以及 Windows 10 关键流程没有因本轮检查验收而变成通过。
## r10 修复
- “停止队列”同时取消后台 ICDB 任务和当前前台队列;独立“停止检查”仍只取消 ICDB 检查。
- 每项在真正取得原生执行机会后检查最新 READY 集合。权威状态读取在让出队列之前发布状态,不等待界面回调;重新加载和局部状态刷新均按库匹配,其他库的同名 CBB 不受影响。
- 仅本次错误明确含 read/write access permission 时显示读写确认;释放未确认和其他安全状态分别显示,原始错误保留在提示中。
- 普通校验排队时保留活跃 ICDB 任务的进度条和阶段标签。
4 个原始复现已转入 `tests/test_background_queue_regressions.py`,修复前均红,修复后均绿;另补权威状态在 Qt 回调前发布、跨库同名隔离以及不同安全原因显示,共 10 个回归。第一轮相关 49 项通过。产品源码仅修改 `ui/main_window.py`、`ui/icdb_check_task.py`;具体最终测试、冻结和原生验收见 r10 的 VALIDATION.json。
最终全量 1,574 项通过(0 失败、0 错误、0 跳过);有既有 pywinauto STA COM 初始化提示 1 条。编译和 diff 检查通过。r10 冻结包 37 个模块/脚本与源码比对、GUI 启动冒烟和包内 x86 状态桥运行库加载检查通过。
r10 实际冻结执行器隔离批次通过:损坏项 FAIL / ICDB_REPAIR_REQUIRED / safe=true(68.563 秒),随后正常项 CHECKED / safe=true(26.812 秒);无关的同桌面模态窗口未被操作,来源六文件哈希不变。实际界面按钮检查及排队状态读取通过,顺序 check-start → check-end → status-start → status-end,筛选可用,最终 READY,50ms 定时器最大间隔 79ms,整体 23.656 秒。目录行的初始 PENDING 为测试注入,原生打开和状态读取实际执行;不冒充现场服务器失联恢复验收。
@@ -0,0 +1,90 @@
{
"review_comment": "https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/3#issuecomment-249",
"version": "0.2.4b1",
"base_main": "a56fd17b60d9f67997d7f48c862f3367d4502f84",
"tests": {
"infrastructure": {
"tests": 491,
"errors": 0,
"failures": 0,
"skipped": 0,
"xml_sha256": "12c2cd40542cb01d4fa374e98908c8a77a8deb83f95d0833645b1666ea5574e9",
"xml_file": "codex-review-infrastructure-pytest.xml",
"seconds": 52.986
},
"native_backend": {
"tests": 1376,
"errors": 0,
"failures": 0,
"skipped": 0,
"xml_sha256": "8db65d11dd918d2a1b141d72627654a990ea903b68e20511e982dfb927d6be58",
"xml_file": "codex-review-native-backend-pytest.xml",
"seconds": 275.957
},
"ui_integration": {
"tests": 1545,
"errors": 0,
"failures": 0,
"skipped": 0,
"xml_sha256": "5a3e6d58d611240c6e34cc9bd2350a8bf605516ef453240333836c1f88233af6",
"xml_file": "codex-review-ui-integration-pytest.xml",
"seconds": 325.065
},
"final_stack": {
"tests": 1601,
"errors": 0,
"failures": 0,
"skipped": 0,
"xml_sha256": "bab74328facfbf512416b62a074fb5cb65ea39b04f0039d42083f96b5349c351",
"xml_file": "codex-review-residual-release-pytest.xml",
"seconds": 246.75
},
"umbrella": {
"tests": 1601,
"errors": 0,
"failures": 0,
"skipped": 0,
"xml_sha256": "566f50b4ba537d4e54939946a5b48cad9ed6dbd8fb69c5ffbabcc5da0fbaf507",
"xml_file": "r10-pr-pytest.xml",
"seconds": 313.479
},
"force_boundary": {
"tests": 43,
"errors": 0,
"failures": 0,
"skipped": 0,
"xml_sha256": "832425306b5fea3a150cc2188e00c97a771bb6d953e54b69c091d36e8062dca9",
"xml_file": "pr249-force-focused.xml",
"seconds": 3.231
}
},
"compileall": "passed",
"git_diff_check": "passed",
"final_content": {
"file_count": 195,
"sha256": "2784b81ff0ae4d47f85ac8ea4b217292bbd9a1e716667e729863bfd4a6a372cd",
"umbrella_equals_stack": true
},
"post_test_formatting": "MainWindow line endings aligned with UI review base (AST equality asserted); unchanged README lines retain main line endings",
"regression_red": {
"baseline": "e4266ae",
"selected_new_ownership_tests": 3,
"expected_failures": 3,
"method": "in-memory baseline module; no production files or native processes modified"
},
"force_contract": {
"clients_only": true,
"same_user_sid_and_windows_session": true,
"servers_never_terminated": true,
"durable_consent_before_action": true,
"independent_library_release_proof_required": true
},
"not_validated": [
"field server-session loss recovery",
"Windows 10",
"cross-machine shared-library concurrency",
"actual Mentor force termination",
"new frozen package"
],
"delivery_policy": "Draft review only; no merge, publish, automatic issue closure or native production-library operation"
}
@@ -0,0 +1,127 @@
{
"candidate": "r10",
"candidate_base_commit": "1acfbd637452e8f64d96981044194661dfb4ec1b",
"python_full_suite": {
"name": "pytest",
"errors": "0",
"failures": "0",
"skipped": "0",
"tests": "1574",
"time": "207.493",
"timestamp": "2026-11-05T21:09:55.032807+08:00",
"hostname": "Abs"
},
"frozen_modules_and_scripts_compared": 37,
"native_check_statuses": [
"FAIL / ICDB_REPAIR_REQUIRED / released",
"CHECKED / released"
],
"native_check_seconds": [
68.563,
26.812
],
"foreign_same_desktop_modals_untouched": true,
"source_primary_files_unchanged": true,
"gui_native_queue_order": [
"check-start",
"check-end",
"status-start",
"status-end"
],
"gui_cached_filter_available": true,
"gui_max_timer_gap_seconds": 0.079,
"automatic_icdb_repair_acceptance": false,
"field_server_session_loss_acceptance": false,
"tested_environment": "Windows 11 / Mentor EE7.9.5",
"candidate_code_sha256": {
"src\\cbb_refresh_verify\\application.py": "babeefe4ddc2fb78b4af3a0413f60cb351eeca57f6f2862e2bc1b8a94de59f32",
"src\\cbb_refresh_verify\\backup_recovery.py": "ea7bb90151e825efb6163e43ed344b62407565005f0a0459c423ba2205539373",
"src\\cbb_refresh_verify\\batch_status.py": "918c80dd4f75146baa6b7616ce415ac89ee28f6c1f46b2bfda77bacc998b5b62",
"src\\cbb_refresh_verify\\close_failed_block.vbs": "e5f0d9678e07a239e96e634b90cdc761de52fc136311fae31484fa38d6fd9781",
"src\\cbb_refresh_verify\\com_runtime.py": "f0f3802d16c791a8bf9dfef44a94917072bfbe8fb3a927d76be8ef0341c7ed35",
"src\\cbb_refresh_verify\\content_check.py": "8239c17b6e2452917ed573a7deb3e4c8583eaef91bf3630eff5f424d2e398aa9",
"src\\cbb_refresh_verify\\defaults.py": "19aaec03822468b18099f358a5aac3433346ff81345b58a896545bf508840e3d",
"src\\cbb_refresh_verify\\desktop_worker.py": "3b79dcab16b67b68550957c09c9de1ba08d85109356274c8cff4d0ae02dec3fb",
"src\\cbb_refresh_verify\\diagnostic_open.py": "d6c49ffe5a9969d95edbbc896d47a7e100f248380a287f99ba7bb8b92dbb3072",
"src\\cbb_refresh_verify\\dialog_agent.py": "6a9b9802912afc4dd670ee9b28845d97bd64fe55fd577815ef729850fcff78db",
"src\\cbb_refresh_verify\\dialog_models.py": "f68cee01d927b48364276d3c1f79e63ce8af231878db0b907d292780cfbafb56",
"src\\cbb_refresh_verify\\dialog_probe.py": "d61f0c955c961980c9635940eefb9c7947c1fe90b4b6fd844720df41ddb0aa0e",
"src\\cbb_refresh_verify\\dialog_rules.py": "3189b476f7ff4678b59212ed7b4081164e038a7cf7662cd895f9019cb3c1edc5",
"src\\cbb_refresh_verify\\dialog_supervisor.py": "15c415c2d2e916f0b64ba04416164ff3b50996d70b22d3b4ac2226b5547ac247",
"src\\cbb_refresh_verify\\failed_session.py": "7e6f7fa5ebcf2d49f00c5efc5f5d26412704b68e9122e476ac1f672009a98c09",
"src\\cbb_refresh_verify\\icdb_maintenance.py": "6ab1d469b09e92641ff60a4b6acd4cc13d40cc200a6fae880e864181526994c4",
"src\\cbb_refresh_verify\\icdb_ownership.py": "7bf2342c1efa34a2ebcbf27964a40a490220f7146fbb6ef957b8d8dd1e511f03",
"src\\cbb_refresh_verify\\icdb_recovery.py": "358c0e9a004f1e0b2127513286e1dffd94b0180a8b7314aa9655462f570b6360",
"src\\cbb_refresh_verify\\integration.py": "86669ad73ae1d2d724d406c3644bb0c297236a4ffdc4b66117930ec660613e88",
"src\\cbb_refresh_verify\\library_lock.py": "315b3cba411713aea870b0a773b3cb986be73b85b88732d8cd82862f22f650db",
"src\\cbb_refresh_verify\\library_manager_verify.py": "71e16a8581ad44b646be78b480f9812d596a6d162b3eec03d078bf4ea4503195",
"src\\cbb_refresh_verify\\library_reader.py": "47d58d1518ba7e69c22b41db2d34c0a61e2cc1362cf28414f0dae43d4a011753",
"src\\cbb_refresh_verify\\maintenance_entry.py": "3b93c25f69d7a961facd8004e20e155a84cc18f06d57b741a9055786ed058990",
"src\\cbb_refresh_verify\\management.py": "64e6fc1382092f1d2154052ec456b9c511934d3f2b231a163be28dca27eb3873",
"src\\cbb_refresh_verify\\management_transaction.py": "32c53caaead05d5db3a31d3e30721e65434e5837095dc0a7812f825603aa37a0",
"src\\cbb_refresh_verify\\mentor_adapter.py": "26a73074ef162d240c80747e371f357c8ec87b00cdbbd3509acc4ac58e68e955",
"src\\cbb_refresh_verify\\mentor_paths.py": "ba48c0003b3eacc88906497b2e43b299ca4536a9f7dae44bbacf6a6ec694c116",
"src\\cbb_refresh_verify\\models.py": "9e8209a8b006df923f1d07ccab64bbcfe5d674c4fc5f5426bfbc9b11da4ee105",
"src\\cbb_refresh_verify\\native_add_failures.py": "192202732423dc624c5548c75e80bb20973e5087c21d4798c9637f25b6de7e94",
"src\\cbb_refresh_verify\\native_backup.py": "7276fa3d9298e85550151cf0c7191d84bd08a18a72df5d854a3c38f1a88b001d",
"src\\cbb_refresh_verify\\native_engineering_identity.py": "a241bc3be7dc5a922ce728f1febd3684e79e6e45935589c80a4b1a5933e05df5",
"src\\cbb_refresh_verify\\native_export_guard.py": "5120c13e4ef1255d4c61316eb40005c4039c23e41edca4437bd0ef28b543a959",
"src\\cbb_refresh_verify\\native_management.py": "bdb42f0bbbc964e56cfe08f66804a03dfc73157eb98911531c9d4c8f940c7595",
"src\\cbb_refresh_verify\\native_occupancy.py": "7d686079d12217382adace6361cd343d977b91075e3b157cf410ca92d07ffc49",
"src\\cbb_refresh_verify\\native_repair.py": "65d3e5b28b1f375402b002f56527dafbcc79e21cea66822e3bc26332d59e76d7",
"src\\cbb_refresh_verify\\native_repair_ui.py": "3eb2af181befea94af8b3b3131edff09589f340bcba777abbe3b67fd31b80393",
"src\\cbb_refresh_verify\\network_path.py": "6b5b159502d42a46bf4ee26d5698fd2c6498b359088e2138eae46e9065b13c6f",
"src\\cbb_refresh_verify\\operation_queue.py": "57ad9da8aeedd9cd75a8c5b5fda44044bcb457670a4a323e9e8bb247198074ab",
"src\\cbb_refresh_verify\\owner_probe.py": "8dcfd60246cb2979db34d02e0a2d605c334b6f7e2666f22daa2f50c0408bc903",
"src\\cbb_refresh_verify\\pcb_bridge.py": "745ef117e984021e8a3d9087a2783c1efe19bed97edcfbcfae7703a291cc6eaa",
"src\\cbb_refresh_verify\\pcb_restore_prompt.py": "5342612ad666e3faf5a9a4ce4710e9af458d357c44037135f2837dad436d5cf7",
"src\\cbb_refresh_verify\\post_repair_consent.py": "3b1eaff1db58fd311a9dbc5fdcd850911b04cc5afd4e747f70a62b721b8fafa4",
"src\\cbb_refresh_verify\\post_verify_diagnostics.py": "2442de485bc0d6a34c4b2d9b10bc5f323ea38f9ff66674119a50f5c6043bac63",
"src\\cbb_refresh_verify\\reimport_recovery.py": "66b2ba8f8c2ec9ae8c386d6b1aa74a1fbb6bfb509c4df0a262f8d7a0ca48469e",
"src\\cbb_refresh_verify\\residual_sessions.py": "5fa8a0f87d6aa4c7e0dec4ab8eb1d649d20fdbea35016fbd65fefefd42607eef",
"src\\cbb_refresh_verify\\runtime_logging.py": "44ac25430a89c5ffc9e48e88c227e88052f4c815365e8556f67918f3edaba39e",
"src\\cbb_refresh_verify\\runtime_timing.py": "c4ed4057b5000e0b86ccc7dd1fe2004d6e0dae29f397fec77b01ad41a81c6367",
"src\\cbb_refresh_verify\\save_verified_block.vbs": "2ede0a1297e44533cd6333b775cdd70c86b5f2aaaaa18ccd6df158379ef1fd31",
"src\\cbb_refresh_verify\\scoped_manifest.py": "54cf4288e7850e220b394ae971ea98785eba631f80784a3f9354209e649f6498",
"src\\cbb_refresh_verify\\status_bridge.py": "05d627d02934a14f5de409472c291d37565a95b836f9c130111aae80612e9196",
"src\\cbb_refresh_verify\\verify_diagnostics.py": "041b77d7d49e4c86338ed5e5741d0d27a2c42bc358f4da54b3efe64af847e9b8",
"src\\cbb_refresh_verify\\verify_probe.py": "3ed743df49d195c74f5045985ab23491d103ee2c42c07b558c1e037176402ed2",
"src\\cbb_refresh_verify\\verify_runner.py": "bb5bae2616e3440a6adc083a2a6941176205f574fde077cf89a6121077833fed",
"src\\cbb_refresh_verify\\__init__.py": "be4027b9b6c4659a7f7d499a14f75ec84c410bd9a706d9ffd9f97d44a52b3806",
"src\\cbb_refresh_verify\\__main__.py": "dc4be034a0e43d4b659a9300137c29a6a6b7a32b7eaef02b6dff9a88c41305c0",
"src\\cbb_refresh_verify\\assets\\dialog_rules.v1.json": "3911080fc799a819662e225a7ad73e883abd8dcc9936cc30ec1b2d6d51a03c6e",
"src\\cbb_refresh_verify\\ui\\backup_notice.py": "1a3fa5ac18a267907faf95f36d61b9f08bcbb939eef4d480dd06af1391a81599",
"src\\cbb_refresh_verify\\ui\\backup_recovery_dialog.py": "621ea2efea85367b0c9bb9790040860a1ac001f40f8834289cb67c96ea0648d1",
"src\\cbb_refresh_verify\\ui\\icdb_check_task.py": "03e6416ee3d2fa11ecedb624a558516eeb56e634f1f22e41e57e10db6a13d656",
"src\\cbb_refresh_verify\\ui\\main_window.py": "4526f59b7f26374ff119d45453b278780a51ef7632eb1611c4e06a205cd30795",
"src\\cbb_refresh_verify\\ui\\management_dialog.py": "27d920284e508f19fae71b04a3b3d81a251559d52c4cc8d691c41186686572b7",
"src\\cbb_refresh_verify\\ui\\pcb_restore_dialog.py": "b7228d828df235ac51502dc68760d28c4896af02c780e9806e11c32dba7f62da",
"src\\cbb_refresh_verify\\ui\\residual_release_dialog.py": "515aaa236ad6a0bee9e5383c6c9d093baeb8bc7dc98f1b978b1197ce672c6764",
"src\\cbb_refresh_verify\\ui\\workers.py": "4cc9360a4820ded27a7d5ef709fc9f04b1d4804cb1e7a614f6c357ed547da888",
"src\\cbb_refresh_verify\\ui\\__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
},
"scope_note": "Candidate icons are local workspace assets and are excluded from this change. Native evidence files and engineering copies remain local.",
"pr_integration": {
"base_commit": "a56fd17b60d9f67997d7f48c862f3367d4502f84",
"python_full_suite": {
"name": "pytest",
"errors": "0",
"failures": "0",
"skipped": "0",
"tests": "1582",
"time": "194.662",
"timestamp": "2026-11-05T21:54:33.463053+08:00",
"hostname": "Abs"
},
"compile_check": true,
"diff_check": true,
"candidate_code_differences": [
"src\\cbb_refresh_verify\\mentor_adapter.py"
],
"difference_scope": "Merge main: defer private desktop allocation into protected cleanup in the two legacy status readers; retain PCB diagnostic/integration behavior.",
"both_cleanup_regression_suites_preserved": true,
"candidate_rebuilt_after_main_integration": false,
"automatic_icdb_repair_acceptance": false,
"field_server_session_loss_acceptance": false
}
}
@@ -0,0 +1,45 @@
{
"date": "2026-11-06T09:27:28.849706+00:00",
"source_commit": "44bb734de80afbe3ff04ba72d52a465ff3b91684",
"version": "0.2.4b2",
"reason": "New user icon was uncommitted in original workspace and absent from isolated worktree; synchronize all three assets.",
"assets": [
{
"name": "cbb_refresh_verify.ico",
"sha256": "62d4472ebaa94f9b39aebae06459685d3322041d0e1a7ada9f368cfd816ba1d6",
"bytes": 123139
},
{
"name": "cbb_refresh_verify.png",
"sha256": "840f657fd7c622d9afd78625a64ed8ad04c90a895dd3c9fd37eebae543cdebe8",
"bytes": 1187049
},
{
"name": "cbb_refresh_verify.svg",
"sha256": "cbba5b3f339cbeb5eb43012709237359442ef67c69fc5b6679a8ee4c0a1b7328",
"bytes": 1582896
}
],
"exe_icon_frames_match_current_ico": true,
"exe_icon_frame_count": 7,
"qt_window_icon_matches_current_ico": true,
"frozen_product_module_bytecode_matches_source": true,
"module_count": 62,
"packaging_tests": {
"passed": 13
},
"frozen_smoke": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\runtime-lifecycle-b2-iconfix\\smoke-334dde93\\result.json",
"exe": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\runtime-lifecycle-b2-iconfix\\dist\\CbbRefreshVerify-Background\\CbbRefreshVerify.exe",
"exe_sha256": "7c4a78d9d2116f637c2a236a82979db2ce49b9b5fa6493a16e66fc803b38d999",
"zip": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\runtime-lifecycle-b2-iconfix\\CbbRefreshVerify-0.2.4b2-iconfix-44bb734.zip",
"zip_sha256": "4a0a96321801fdebf13fed0f850a0c4b32d6928480288fd35c2d36f534e6e5b5",
"zip_bytes": 56846008,
"runtime_changes": false,
"full_runtime_suite_rerun": false,
"prior_runtime_suite": {
"passed": 1717,
"source_commit": "5873b7146b61d0cdd0eb22f6f2941797b02f38fc"
},
"pr": 13,
"draft": true
}
@@ -0,0 +1,75 @@
{
"date": "2026-11-06",
"review_comments": [
275,
283,
286
],
"tests": {
"backend_full": {
"tests": 1385,
"errors": 0,
"failures": 0,
"skipped": 0,
"xml_file": "cbb-refresh-verify-795-backend-fixes-fixes.xml",
"xml_sha256": "b1af9d0025abbbbcf25fe134b02da980c9402dd22c0fa39e1f96b9f8124183ee"
},
"backend_read_guard": {
"tests": 9,
"errors": 0,
"failures": 0,
"skipped": 0,
"xml_file": "cbb-refresh-verify-795-backend-fixes-fixes-focused.xml",
"xml_sha256": "25dac539945d9ba71f2f9dafeb342e8f6110e094c62364369013e16e551b1f94"
},
"ui_integration": {
"tests": 29,
"errors": 0,
"failures": 0,
"skipped": 0,
"xml_file": "cbb-refresh-verify-795-ui-fixes-fixes.xml",
"xml_sha256": "9eeeff28c6fcd31d8f2e6d7fb38e0bdf01b0058ba7b7f9e5df2f173f71a4fca0"
},
"final_integrated": {
"tests": 1631,
"errors": 0,
"failures": 0,
"skipped": 0,
"xml_file": "cbb-refresh-verify-795-review-stack-fixes-full.xml",
"xml_sha256": "7d75049e4feecb4a6026062321d4a0fcb1c92e30f8f1ba23d11a4ea28ee19c9c"
},
"final_release_focused": {
"tests": 63,
"errors": 0,
"failures": 0,
"skipped": 0,
"xml_file": "cbb-refresh-verify-795-review-stack-fixes-focused.xml",
"xml_sha256": "469923faf0919ab2e3dffac183f2cb44a6b606aa15edacce92a88da1bde81424"
}
},
"compileall": "passed",
"git_diff_check": "passed",
"source_and_tests_sha256": "6fe6f02a3e0e731c4c9d2ff2f2c5e86e537139ef5e59f15441a9059522109526",
"post_full_suite_change": "GUI proof path comparison now uses lexical absolute paths, with no filesystem resolution; final 63 release regressions cover this one-line change. No further full-suite run was performed for this isolated UI comparison.",
"ui_preview": {
"file": "2026-11-06-release-preview.png",
"sha256": "e13645ed6b8234a2d46f6c462a262c7ef2092459c8e94f60a9fa273fd02c3613",
"mode": "offscreen simulated sessions; local Windows fonts loaded for visual inspection"
},
"red_regressions": {
"initial_backend": 6,
"initial_release": 12,
"ui_outcomes": 2,
"additional": "report storage, prelaunch evidence/probe timeout, cancel release propagation"
},
"scope": "Historical records trigger actual ownership proof; no automatic repair, server termination, native lock/address deletion or library backup.",
"native_cancel": "Only a unique enabled standard Cancel/取消 (IDCANCEL=2) on a current matching Project Backup window; identity, user/session, title, desktop and button rechecked; request does not prove cancellation.",
"not_verified": [
"actual Mentor native cancellation",
"actual Mentor force termination",
"field ICDB recovery",
"new frozen package",
"Windows 10",
"cross-machine shared-library concurrency"
]
}
@@ -0,0 +1,43 @@
# 0.2.4b2 原生操作生命周期候选
本候选处理 PR #12 评论 297 的三条 P1 及直接相关的停止、释放记录问题。ICDB 自动 Repair 仍关闭,现场失联恢复尚未验收。本轮不合并、不发布正式版、不关闭工单。
差分审查入口:[Draft PR #13](https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/13),依赖 #12。
**当前图标修正版:** `44bb734` 将原工作区的新版蓝色对勾/环形箭头 ICO、PNG、SVG 同步进候选分支。已核对 EXE 内嵌的七个尺寸图标、Qt 窗口图标及三份打包资源一致,13 项打包测试和冻结入口冒烟通过。62 个产品模块字节码保持与当前源码一致,未为资源替换重跑业务全量。当前便携包为 `CbbRefreshVerify-0.2.4b2-iconfix-44bb734.zip`(56,846,008 字节),SHA-256:`4a0a96321801fdebf13fed0f850a0c4b32d6928480288fd35c2d36f534e6e5b5`。实际程序路径及验证明细见[图标修正记录](2026-11-06-icon-candidate-validation.json)。下文首次候选包哈希保留用于历史追溯。
## 行为
- 停止、超时、失败收尾和监督器失联只请求停止后续动作,不强杀可能正在 Save/Close 的 worker 或 cscript。保存助手先持久登记精确身份,再通过启动门进入 COM。保存返回、关闭返回和独立资源释放分别记录;进程消失不能把未知保存升级为成功。
- 普通停止等待 5 秒;可能正在写入时最多等待 120 秒后转后台观察。相关库的租约、映射和精确进程观察继续保留,不占全局队列票据;独立库可以运行。用户取消的批次不自动恢复。
- 管理 catalog/add/remove 使用统一监督器身份、取消信号、启动门和会话记录。更新在 remove 后停止,记录需恢复,不继续 add。异步 Add 提交前必须成功保存 NATIVE_CALL;提交后记录失败仍保留监督。未知 Add 结果不触发自动 Close/Quit,明确且已应答的原生失败可正常收尾。
- 重启在后台核对受控记录,不执行记录中的命令或重放写操作。PID 复用只观察替代进程,继续独立资源核对;证明成立才解除对应等待。报告错误、损坏 JSON 本身不生成全局占用。
- 权威状态桥保持默认;legacy 读取只观察已打开、匹配库且可读取的 LM,不启动程序、不回答提示、不关闭用户窗口。无法观察保持 UNKNOWN。
界面保持三个页签。新增说明写在本文和代码中;界面仅呈现当前状态和操作结果。原工作区、用户图标改动和已运行程序未替换。
## 实现与审查
基线为 `5eb4b2d`,独立分支为 `codex/runtime-lifecycle-safety`。保存与只读接线共用 mentor_adapter,审查顺序为:事实/会话协议 → 保存与停止 → 管理监督 → 界面交接 → 独立审查修正。共享文件的 A/C 接线合并在同一提交,候选仅在三单元联通后生成。
独立审查发现并修正:IDENTITY_CHANGED 提前结束观察、异步 Add 阶段不准确、权限 COM 查询后未再次检查取消,以及写入阶段记录失败后仍继续提交 Add。新增回归先复现失败,再验证修正。该审查不等于真实 Mentor 数据恢复验收。
## 验证口径
可机读结果、源码身份、XML 与候选包哈希见 [验证记录](2026-11-06-runtime-lifecycle-validation.json)。Python 全量回归、实际 32 位 cscript 的 FakeDoc Save/Close、真实 Windows 测试进程的父进程退出,以及冻结入口冒烟分别记录。
最终源码提交 `5873b71`:**1,717 passed**,编译与 diff 检查通过。正式候选 GUI 及四类 worker 入口冒烟通过;62 个产品模块的冻结字节码与当前源码一致。独立冻结夹具的保存/管理 × 用户取消/父进程退出共四个场景通过。唯一已知测试警告为 pywinauto 的 STA 模式提示。
首次候选便携 ZIP 为 `CbbRefreshVerify-0.2.4b2-runtime-lifecycle-5873b71.zip`,54,227,147 字节;SHA-256:`5983a4e99884d9551d1dba3d81af88f754ce876c19a780fe82e8ea9fd58a566b`。未上传为正式 Release。构建、XML 和各专项结果文件的实际路径保存在验证 JSON。
冻结测试夹具是独立测试程序,使用同一批产品模块并替换外部原生边界;不随候选交付,也不是应用实际按钮操作或真实数据库修复。冻结产品的 GUI、启动前取消、拒绝错误桌面的管理入口、进程探针及仅观察的弹窗代理单独冒烟。
可复现脚本在 `tests/support/runtime_lifecycle_frozen_*.py`。它们是实际执行脚本的逐字节副本,最终套件完成后归档,未改变产品源码或 pytest 收集范围。先按现有 spec 构建产品,再以 `runtime_lifecycle_frozen_fixture.py` 为独立 PyInstaller onedir/console 入口、产品 `src` 为搜索路径构建夹具;`runtime_lifecycle_frozen_smoke.py <worktree> <artifact-root>` 和 `runtime_lifecycle_frozen_boundaries.py <worktree> <artifact-root>` 分别验证产品与夹具。产物目录结构见验证 JSON,不应将测试夹具放入便携包。
本轮没有在生产库执行原生写入、删除地址/锁文件、强杀 Mentor、选择历史版本或重放旧管理任务。下列项仍为 **未验收**:
- 使用本候选实际按钮完成真实 Mentor 隔离副本的 Save 中停止、延迟 Close、管理失联及独立库继续。
- Windows 10、共享网络库和跨机器并发。
- 现场故障一致副本上的 ICDB 失联恢复、数据版本和 READY 验收。
原生 COM 调用本身可能不可取消。后台观察提供状态与资源保护,不承诺把任意卡死进程自动安全结束。回退程序不撤销数据库事务;工程恢复仍依靠人工一致备份或明确选择的原生历史版本。
@@ -0,0 +1,387 @@
{
"date": "2026-11-06T08:53:02.058981+00:00",
"version": "0.2.4b2",
"base_commit": "5eb4b2d7ed09947f19310ba80ba3b9a2a09b6f7d",
"source_commit": "5873b7146b61d0cdd0eb22f6f2941797b02f38fc",
"src_tree": "ef5ef877665fff179dd6d21d8d6f86e8de0a7e5b",
"pytest_tests_tree": "35f5ad42f2c6150bf5762a46cdeb9880d1fc0d0c",
"environment": {
"platform": "Windows-11-10.0.26200-SP0",
"python": "3.12.10",
"mentor_library_manager_file_version": "2012.0.522.442",
"mentor_library_manager_sha256": "a9c4c3d806794dbdc56d2bc4c637a1cf25e87e98644a7d5fd4eb0067ca5845e0",
"mentor_home": "D:\\Mentor\\7.9.5EE\\SDD_HOME",
"matches_validated_mentor_library_manager_binary": true
},
"tests": {
"final_full": {
"status": "passed",
"tests": 1717,
"failures": 0,
"errors": 0,
"skipped": 0,
"xml": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\runtime-lifecycle-final-1717.xml",
"xml_sha256": "1d427f97dd04afbf8912bbdf787b0b9dfd1ed383d32bfe07f7a94cc24f9c5292"
},
"known_warning": "pywinauto: Revert to STA COM threading mode",
"compileall": "passed",
"git_diff_check": "passed",
"independent_review": "important findings resolved through 051498b"
},
"frozen_product": {
"build": "passed",
"smoke": {
"gui": {
"status": "passed",
"exit_code": 0
},
"verify-worker": {
"exit_code": 1,
"status": "passed",
"evidence": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\runtime-lifecycle-b2\\smoke-74da6fd5\\verify-worker"
},
"management-worker": {
"exit_code": 2,
"status": "passed",
"evidence": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\runtime-lifecycle-b2\\smoke-74da6fd5\\management-worker"
},
"owner-probe": {
"exit_code": 0,
"status": "passed",
"evidence": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\runtime-lifecycle-b2\\smoke-74da6fd5\\owner-probe"
},
"dialog-agent": {
"status": "passed",
"exit_code": 0,
"scope": "empty PID set, observe only"
},
"module_origins": {
"status": "passed",
"module_count": 62,
"source_root": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795-review-stack\\src"
},
"version": "0.2.4b2"
},
"evidence": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\runtime-lifecycle-b2\\smoke-74da6fd5\\result.json",
"evidence_sha256": "4f354f3700d09e299e6437d36492fd8b380440eae7e5ec859c90645f16004ac2",
"module_bytecode_matches_source": true,
"module_count": 62,
"modules": [
{
"module": "cbb_refresh_verify",
"source_sha256": "dd960394d88657820dba858792abf0de258e05efceb08ae2f0b39c6ad78bc469"
},
{
"module": "cbb_refresh_verify.application",
"source_sha256": "433c22b01df8376651c43cc9fe7f9551b531dcd3b3e590b77cb3ec1bfbfd2de6"
},
{
"module": "cbb_refresh_verify.backup_recovery",
"source_sha256": "ea7bb90151e825efb6163e43ed344b62407565005f0a0459c423ba2205539373"
},
{
"module": "cbb_refresh_verify.batch_status",
"source_sha256": "918c80dd4f75146baa6b7616ce415ac89ee28f6c1f46b2bfda77bacc998b5b62"
},
{
"module": "cbb_refresh_verify.com_runtime",
"source_sha256": "f0f3802d16c791a8bf9dfef44a94917072bfbe8fb3a927d76be8ef0341c7ed35"
},
{
"module": "cbb_refresh_verify.content_check",
"source_sha256": "8239c17b6e2452917ed573a7deb3e4c8583eaef91bf3630eff5f424d2e398aa9"
},
{
"module": "cbb_refresh_verify.defaults",
"source_sha256": "a1c03192f4a61ad0924b24bb635705b1ee4811d35483a02864200f95edbbebe3"
},
{
"module": "cbb_refresh_verify.desktop_worker",
"source_sha256": "1dd449e9885b32cb479970ddb9d1e3b580bc6339f9b491be0a414a1de8054efa"
},
{
"module": "cbb_refresh_verify.diagnostic_open",
"source_sha256": "d6c49ffe5a9969d95edbbc896d47a7e100f248380a287f99ba7bb8b92dbb3072"
},
{
"module": "cbb_refresh_verify.dialog_agent",
"source_sha256": "6a9b9802912afc4dd670ee9b28845d97bd64fe55fd577815ef729850fcff78db"
},
{
"module": "cbb_refresh_verify.dialog_models",
"source_sha256": "f68cee01d927b48364276d3c1f79e63ce8af231878db0b907d292780cfbafb56"
},
{
"module": "cbb_refresh_verify.dialog_probe",
"source_sha256": "d61f0c955c961980c9635940eefb9c7947c1fe90b4b6fd844720df41ddb0aa0e"
},
{
"module": "cbb_refresh_verify.dialog_rules",
"source_sha256": "3189b476f7ff4678b59212ed7b4081164e038a7cf7662cd895f9019cb3c1edc5"
},
{
"module": "cbb_refresh_verify.dialog_supervisor",
"source_sha256": "15c415c2d2e916f0b64ba04416164ff3b50996d70b22d3b4ac2226b5547ac247"
},
{
"module": "cbb_refresh_verify.failed_session",
"source_sha256": "7e6f7fa5ebcf2d49f00c5efc5f5d26412704b68e9122e476ac1f672009a98c09"
},
{
"module": "cbb_refresh_verify.icdb_maintenance",
"source_sha256": "e81d253d8b4d3e7c1000c45335a392c72b110748675c2c3324ebe524a7252e98"
},
{
"module": "cbb_refresh_verify.icdb_ownership",
"source_sha256": "7bf2342c1efa34a2ebcbf27964a40a490220f7146fbb6ef957b8d8dd1e511f03"
},
{
"module": "cbb_refresh_verify.icdb_recovery",
"source_sha256": "358c0e9a004f1e0b2127513286e1dffd94b0180a8b7314aa9655462f570b6360"
},
{
"module": "cbb_refresh_verify.integration",
"source_sha256": "f188856af69ba25e265988c51db7a6d4263326440d972ca78378b733b891c1ed"
},
{
"module": "cbb_refresh_verify.library_lock",
"source_sha256": "eea9021300410cc810eaa0e18aac546c94139f89528f5d68784ef3f943b65ce4"
},
{
"module": "cbb_refresh_verify.library_manager_verify",
"source_sha256": "71e16a8581ad44b646be78b480f9812d596a6d162b3eec03d078bf4ea4503195"
},
{
"module": "cbb_refresh_verify.library_reader",
"source_sha256": "d4007caa361edd467773e49f721e02671d5a9bcb30741769a57349850b12c438"
},
{
"module": "cbb_refresh_verify.maintenance_entry",
"source_sha256": "3b93c25f69d7a961facd8004e20e155a84cc18f06d57b741a9055786ed058990"
},
{
"module": "cbb_refresh_verify.management",
"source_sha256": "3d6e2505ce6e21acdea88d59b920229ba1ce2cee6c914ac1b0b8267cf0d34ffc"
},
{
"module": "cbb_refresh_verify.management_transaction",
"source_sha256": "32c53caaead05d5db3a31d3e30721e65434e5837095dc0a7812f825603aa37a0"
},
{
"module": "cbb_refresh_verify.mentor_adapter",
"source_sha256": "15bde6706c6533d839646b894896237921b89a8a3ce3016aa8161c591dab314f"
},
{
"module": "cbb_refresh_verify.mentor_paths",
"source_sha256": "ba48c0003b3eacc88906497b2e43b299ca4536a9f7dae44bbacf6a6ec694c116"
},
{
"module": "cbb_refresh_verify.models",
"source_sha256": "9e8209a8b006df923f1d07ccab64bbcfe5d674c4fc5f5426bfbc9b11da4ee105"
},
{
"module": "cbb_refresh_verify.native_add_failures",
"source_sha256": "192202732423dc624c5548c75e80bb20973e5087c21d4798c9637f25b6de7e94"
},
{
"module": "cbb_refresh_verify.native_backup",
"source_sha256": "7276fa3d9298e85550151cf0c7191d84bd08a18a72df5d854a3c38f1a88b001d"
},
{
"module": "cbb_refresh_verify.native_engineering_identity",
"source_sha256": "a241bc3be7dc5a922ce728f1febd3684e79e6e45935589c80a4b1a5933e05df5"
},
{
"module": "cbb_refresh_verify.native_export_guard",
"source_sha256": "5120c13e4ef1255d4c61316eb40005c4039c23e41edca4437bd0ef28b543a959"
},
{
"module": "cbb_refresh_verify.native_management",
"source_sha256": "27a835bc31cfe0e1a09d767977e5c348d40e246c58d58ce1ef085411afe6d56e"
},
{
"module": "cbb_refresh_verify.native_occupancy",
"source_sha256": "7d686079d12217382adace6361cd343d977b91075e3b157cf410ca92d07ffc49"
},
{
"module": "cbb_refresh_verify.native_repair",
"source_sha256": "9f024dcb5e3a76369191033474fefd30787b6ae350c315f3cf38e92e5a1f3e98"
},
{
"module": "cbb_refresh_verify.native_repair_ui",
"source_sha256": "3eb2af181befea94af8b3b3131edff09589f340bcba777abbe3b67fd31b80393"
},
{
"module": "cbb_refresh_verify.network_path",
"source_sha256": "6b5b159502d42a46bf4ee26d5698fd2c6498b359088e2138eae46e9065b13c6f"
},
{
"module": "cbb_refresh_verify.operation_lifecycle",
"source_sha256": "28b6cdebd22caaf369eeb8b7236e258b8160a8fa7c7d8bfbf0980979e9eba3ae"
},
{
"module": "cbb_refresh_verify.operation_queue",
"source_sha256": "57ad9da8aeedd9cd75a8c5b5fda44044bcb457670a4a323e9e8bb247198074ab"
},
{
"module": "cbb_refresh_verify.owner_probe",
"source_sha256": "8dcfd60246cb2979db34d02e0a2d605c334b6f7e2666f22daa2f50c0408bc903"
},
{
"module": "cbb_refresh_verify.pcb_bridge",
"source_sha256": "745ef117e984021e8a3d9087a2783c1efe19bed97edcfbcfae7703a291cc6eaa"
},
{
"module": "cbb_refresh_verify.pcb_restore_prompt",
"source_sha256": "5342612ad666e3faf5a9a4ce4710e9af458d357c44037135f2837dad436d5cf7"
},
{
"module": "cbb_refresh_verify.post_repair_consent",
"source_sha256": "3b1eaff1db58fd311a9dbc5fdcd850911b04cc5afd4e747f70a62b721b8fafa4"
},
{
"module": "cbb_refresh_verify.post_verify_diagnostics",
"source_sha256": "2442de485bc0d6a34c4b2d9b10bc5f323ea38f9ff66674119a50f5c6043bac63"
},
{
"module": "cbb_refresh_verify.reimport_recovery",
"source_sha256": "66b2ba8f8c2ec9ae8c386d6b1aa74a1fbb6bfb509c4df0a262f8d7a0ca48469e"
},
{
"module": "cbb_refresh_verify.residual_sessions",
"source_sha256": "69d39648a2f3c98a297847baf51373632acd881f034a2a011d0c3d27dbec3dfc"
},
{
"module": "cbb_refresh_verify.runtime_logging",
"source_sha256": "44ac25430a89c5ffc9e48e88c227e88052f4c815365e8556f67918f3edaba39e"
},
{
"module": "cbb_refresh_verify.runtime_timing",
"source_sha256": "c4ed4057b5000e0b86ccc7dd1fe2004d6e0dae29f397fec77b01ad41a81c6367"
},
{
"module": "cbb_refresh_verify.scoped_manifest",
"source_sha256": "54cf4288e7850e220b394ae971ea98785eba631f80784a3f9354209e649f6498"
},
{
"module": "cbb_refresh_verify.session_record",
"source_sha256": "394719d125b91ff124d707229f4599dcca9229d2c6e653840ef7e02f48c642b6"
},
{
"module": "cbb_refresh_verify.status_bridge",
"source_sha256": "05d627d02934a14f5de409472c291d37565a95b836f9c130111aae80612e9196"
},
{
"module": "cbb_refresh_verify.ui",
"source_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
},
{
"module": "cbb_refresh_verify.ui.backup_notice",
"source_sha256": "1a3fa5ac18a267907faf95f36d61b9f08bcbb939eef4d480dd06af1391a81599"
},
{
"module": "cbb_refresh_verify.ui.backup_recovery_dialog",
"source_sha256": "621ea2efea85367b0c9bb9790040860a1ac001f40f8834289cb67c96ea0648d1"
},
{
"module": "cbb_refresh_verify.ui.icdb_check_task",
"source_sha256": "01985644641ec623ed440433248d9445ab5b9f37d27a843ec7f26267895de397"
},
{
"module": "cbb_refresh_verify.ui.main_window",
"source_sha256": "a625b2934a34c86ff30eaf115e50b0560cbb24f3dccf15bbb51574a9808c660d"
},
{
"module": "cbb_refresh_verify.ui.management_dialog",
"source_sha256": "93b7a7464161e7448aa78c42ca8742c970e80387f9c6c1b9b4b6a07532da3ebf"
},
{
"module": "cbb_refresh_verify.ui.pcb_restore_dialog",
"source_sha256": "b7228d828df235ac51502dc68760d28c4896af02c780e9806e11c32dba7f62da"
},
{
"module": "cbb_refresh_verify.ui.residual_release_dialog",
"source_sha256": "ca0e18886457800cae7b543286395f05db1a74eb41e53fc54cfee57f93294368"
},
{
"module": "cbb_refresh_verify.ui.workers",
"source_sha256": "4cc9360a4820ded27a7d5ef709fc9f04b1d4804cb1e7a614f6c357ed547da888"
},
{
"module": "cbb_refresh_verify.verify_diagnostics",
"source_sha256": "041b77d7d49e4c86338ed5e5741d0d27a2c42bc358f4da54b3efe64af847e9b8"
},
{
"module": "cbb_refresh_verify.verify_runner",
"source_sha256": "2c7c37d2b510ab5735cd4a71cb834768c711e31b0ce5dd28b23acb8dcfe22ffb"
}
],
"exe": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\runtime-lifecycle-b2\\dist\\CbbRefreshVerify-Background\\CbbRefreshVerify.exe",
"exe_sha256": "9051ab52acb67d0d291ae230e7d55d1ce6f580f653d58a82d463b7c42cebb90f",
"zip": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\runtime-lifecycle-b2\\CbbRefreshVerify-0.2.4b2-runtime-lifecycle-5873b71.zip",
"zip_sha256": "5983a4e99884d9551d1dba3d81af88f754ce876c19a780fe82e8ea9fd58a566b",
"zip_bytes": 54227147
},
"frozen_native_boundaries": {
"scope": "separate frozen fixture; fake native COM/management boundary",
"product_native_database_recovery_validated": false,
"results": [
{
"mode": "save",
"stop_kind": "cancel",
"status": "passed"
},
{
"mode": "save",
"stop_kind": "supervisor-loss",
"status": "passed"
},
{
"mode": "management",
"stop_kind": "cancel",
"status": "passed"
},
{
"mode": "management",
"stop_kind": "supervisor-loss",
"status": "passed"
}
],
"evidence": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\runtime-lifecycle-b2\\frozen-boundaries-cdb9d884\\result.json",
"evidence_sha256": "9cdc149a6fa545206c361aaba278ef54c9ee2529e9a09b440c5fcf9378fc8529",
"module_bytecode_matches_source": true,
"module_count": 47,
"validation_helpers": [
{
"file": "tests\\support\\runtime_lifecycle_frozen_fixture.py",
"sha256": "b87761c06a300f0d93bce864d3d31e045f2135324b2a9a42f8682ccbff09ad22"
},
{
"file": "tests\\support\\runtime_lifecycle_frozen_boundaries.py",
"sha256": "bf98533a1c06ccf4453d20b3804323779719e697141b8dc64fa679960a979cfd"
},
{
"file": "tests\\support\\runtime_lifecycle_frozen_smoke.py",
"sha256": "b11e0aac71c1079625ac9832dc5c38197bc80110478c4d9e02c2a3bc58a2a2d0"
}
],
"helpers_copied_after_full_suite": true,
"note": "Tracked helper files are byte-identical copies of the scripts executed from diagnostics. Not shipped; do not replace real application-button acceptance."
},
"not_run": [
"Real Mentor isolated-copy application-button Save/Close and management-loss scenarios",
"Windows 10 and cross-machine shared-library concurrency",
"Field ICDB disconnected-session recovery and data-version/READY acceptance"
],
"auto_repair_enabled": false,
"production_native_writes_performed": false,
"merged_or_published": false,
"issues_closed": false,
"pull_request": {
"number": 13,
"url": "https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/13",
"draft": true,
"base": "codex/review-residual-release",
"creation_head": "8e946c6e7a115a1ac22101b97df9b3fc6e2862f0"
}
}
@@ -0,0 +1,109 @@
{
"date": "2026-11-09",
"scope": "New full reviews #344/#345/#348 and runtime re-review #326",
"branch_heads_before_validation_commit": {
"codex/review-native-backend": "e8a1714d956e733648dacda1bfdce387e07e4cba",
"codex/review-ui-integration": "0fca6abb4e1b54bd06e6986782ccf45383306e43",
"codex/review-residual-release": "a7df105616a8f5cfab61b85e0c5d93aa74ded415",
"codex/runtime-lifecycle-safety": "a045683c8a6d6918ec701e2944d0e20c25296b65"
},
"environment": {
"python": "3.12.10",
"windows": "10.0.26200"
},
"full_suite": {
"tests": 1775,
"failures": 0,
"errors": 0,
"skipped": 0,
"duration_seconds": 273.089,
"xml": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\2026-11-09-full-review-final-1775.xml",
"xml_sha256": "a993473b7784748f671a3d38a0e80f1aa7dfaf714b23d6378ed26d0481c1b904",
"warnings": [
"Existing pywinauto STA COM threading warning"
]
},
"focused_runs": [
{
"scope": "Backend",
"passed": 287
},
{
"scope": "UI standalone",
"passed": 65
},
{
"scope": "Residual standalone forward integration",
"passed": 41
},
{
"scope": "Final runtime/UI integration before last merge",
"passed": 211
},
{
"scope": "Terminal-publication fix lifecycle/management group",
"passed": 159
}
],
"compileall": "passed",
"diff_check": "passed",
"independent_review": "No remaining important introduced findings after fixing two queue timing findings",
"tested_files_sha256": {
"src/cbb_refresh_verify/application.py": "62b9b013563a216d020c8366cccb8f949b02afa72dc63c26940804441cc23057",
"src/cbb_refresh_verify/batch_status.py": "55a4007cf54491ca8e045a5bf86aa4353ac8437cca5a76cd79304962e3d3fa0a",
"src/cbb_refresh_verify/desktop_worker.py": "27d751816552d215dd8ee9cd67941d824f7b4493120ead7c3f87e5eba5ec4a92",
"src/cbb_refresh_verify/library_lock.py": "636e0b1c84061796424da6289361c059fecacad31802df2edc3f3455ec8c40d8",
"src/cbb_refresh_verify/management.py": "29ecb101dacf2bb23b8b8931c01078956dd6c641444e82a69a62e811c7261fc5",
"src/cbb_refresh_verify/native_management.py": "7598a18da12dd527e9403a956b1ade76b253f89a9ee4790f0802244b54616844",
"src/cbb_refresh_verify/operation_queue.py": "a6932d8bc68ac436091f6d362557c66d274ec52d0f295a22f014a3396845c79f",
"src/cbb_refresh_verify/pcb_bridge.py": "2f6c2b09642100468814500569dbfc6695fc2a46b9a3b8333daa8e1530cdeef5",
"src/cbb_refresh_verify/scoped_manifest.py": "26ba8303e2de294374443513d667570abed2f0f2563a2e969a9cb35e74d65c00",
"src/cbb_refresh_verify/session_record.py": "fbf31fb38894f75d725b269229c7d928be54136b491cc5339f7a4b018d7b7f5e",
"src/cbb_refresh_verify/ui/icdb_check_task.py": "cc108974e843f31e59e90acd18ad2d810e6bcd67d7cf303b2b772501d95b6d8e",
"src/cbb_refresh_verify/ui/main_window.py": "a05d9d838be4b9e9f6e2b612c25186e77f23899bffc59b67f50fea597eab47c9",
"src/cbb_refresh_verify/ui/management_dialog.py": "f3fc740197fc109ae3ddd8ead556a8847d5e00ec9a97d5cc2e166d656fb861ef",
"src/cbb_refresh_verify/ui/residual_release_dialog.py": "82382af142379c540721dbfa7ba5a5d766130d89807e1f34ad032a0ad1411f5c",
"tests/test_batch_status.py": "c1f9af0c7c6f3171b3726fb53346f480b13b8d9cd88111a9c12e2fae2392bbff",
"tests/test_batch_status_faults.py": "314e890c9aee439c8a75b4dfd605685240c74f6212da975fd50526bd70b4a0ff",
"tests/test_management_retry_blockers.py": "124dbef291dbe331ba0c3e7de1665c910ea43e37b1563429b387f0f2233953ee",
"tests/test_management_supervision.py": "2dcadd6551a94c52e5d4e74285959f10144a89c48af87e508abaea945652d1f6",
"tests/test_pcb_bridge.py": "527fb63147a203a2e9419da989aff600d3e461580989e39e8ede3a8cc57b901a",
"tests/test_residual_release_dialog.py": "10b23d009703d542e5ce8ecb35a4833e0135bb8852b4116e49eedc7010397a1e",
"tests/test_review326_dispatch_and_receipts.py": "ce4d529246da1c4d53c3dec70ef49542165064404301dec3e32851df62492ecf",
"tests/test_review326_queue_guidance.py": "c2c97aceddd9713618afdd5f2a7be4764b56a1b62fc672282fb8c218b4969c74",
"tests/test_review345_ui_cli.py": "795e7e4f9bda8874c6a65a08863cab19f4adb3412287276ee2960588f0fd2fa8",
"tests/test_review348_release_comparison.py": "ceccdb3d6f7bc6cb65562c059f13aa2bc5d89d98c6f64b0f11d9b943ebbde165",
"tests/test_review_backend_followup.py": "c10704fae250c9e204c995843a2baab0a2562e51984ae4d429aa438ba1c88113",
"tests/test_review_management_ui_followup.py": "746602a84c3ba9cb041c5dd354b39d344355b71a3666b27a47eb9aee231e171c",
"tests/test_scoped_manifest.py": "d22749cd240ab365f91ebb9d0469dc3336150338a84a77c1aeb6fbc20ccd439f"
},
"icon_sha256": {
"ico": "62d4472ebaa94f9b39aebae06459685d3322041d0e1a7ada9f368cfd816ba1d6",
"png": "840f657fd7c622d9afd78625a64ed8ad04c90a895dd3c9fd37eebae543cdebe8",
"svg": "cbba5b3f339cbeb5eb43012709237359442ef67c69fc5b6679a8ee4c0a1b7328"
},
"design_choices": [
"Production status always fresh; low-level checkpoint reuse explicit",
"Final source and target hashes retained; copy hashed in stream",
"Optional CLI limit; default all requested targets",
"Release mapping cleanup serialized before waiting task admission",
"New native calls vetoed on supervisor loss; accepted pending calls may record continuation identity"
],
"candidate_rebuilt": false,
"automatic_icdb_repair": false,
"merged": false,
"issue_closed": false,
"not_verified": [
"Real Mentor/native database writes during loss/cancel",
"Frozen execution of this revision",
"Windows 10",
"Shared network/cross-machine release"
],
"first_full_run": {
"tests": 1774,
"failures": 1,
"xml": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\2026-11-09-full-review-first-failure-1774.xml",
"xml_sha256": "9d7275b0a4122db12b8367e4367759f3051ce29a56645da5098776d8b9ca2bc8",
"reason": "FINISHED permitted watchdog orphan exit before result publication; receipt now precedes terminal phase"
}
}
@@ -0,0 +1,43 @@
# Full-review follow-up: comments 344 / 345 / 348 / 326
This revision follows the new November 9 full reviews. The old November 6 deep-review backlog remains outside the user's previously selected scope.
## Backend: #10 / 344
- Status scans default to fresh queries. The low-level explicit checkpoint reuse API remains for compatibility with controlled read-only scans; production status queries continue to use `resume=False`. Its lightweight fingerprint is not described as proof that payload databases are unchanged.
- Change reports include rule version, compared scope, excluded directory categories/suffixes and positional runtime rules from the classifier constants. Unrelated CBB content is still explicitly outside comparison. Counts in the management result refer to compared engineering files.
- Removed the tests-only `pcb_bridge.execution_outcome` helper with optimistic release defaults; active production release checks remain authoritative.
- SHA-256 is computed during streaming of unlocked primary files instead of a separate preliminary source pass. The final source SHA-256 and target SHA-256 remain. This reduces three source reads to two without dropping same-size/same-mtime mutation detection. The review's suggested removal of final source hashing was not adopted because it would weaken the requested source-change check.
Backend commit `e8a1714`; 287 focused regressions passed. Details and exact command are in `docs/reviews/2026-11-05-native-backend.md`.
## UI/CLI: #11 / 345
- Replaced repeated same-purpose result windows with lifecycle-managed dialogs. Task recovery associations survive independent result windows; deleted dialogs/actions are forgotten. Forward integration keeps #12's enabled release/resume paths and recreates a resume action if its old summary was closed.
- Closing waits visibly, including a persistent still-waiting notice after 120 seconds. It does not promise a kill deadline or interrupt a write. The final idle check includes the independent lifecycle pool.
- Optional positive CLI `--limit` bounds requested names in order. Omission retains the existing all-target behavior rather than silently dropping work. Logs distinguish selected/total/remaining and explain retained mappings after exceptions or unsafe results, pointing to confirmed release instead of an unconditional network-drive deletion.
- Removed `if False` branches/dead reload field in the standalone UI PR; forward integration retains the later enabled release behavior. File line-ending rules from the prior tip remain.
UI commit `342a726`; 65 targeted UI/CLI tests passed on its branch. Integration keeps PR13 stop propagation; CLI test doubles were updated to accept and assert that stop input.
## Residual release: #12 / 348
The unreadable-window poisoning and oversized repeated-output findings target the old `5eb4b2d` tree and are already fixed in `651b539`. This revision additionally displays confirmation scope and actual execution result as separate columns, and counts allowed-but-unreleased rows with their reasons. Preview consent never overrides a fresh activity/identity refusal. Residual commit `14c9a74`; 35 targeted tests passed.
## Runtime and queue: #13 / 326
- Native management launch/open/create/close/quit/Add submission/Remove launch and editor entry use `dispatch_native`. It records the pending phase before callback entry, synchronous return records POSTCALL, callback failure records UNCONFIRMED, and asynchronous submission requires its existing explicit result acknowledgment. Save/failure-close helpers retain their atomic registration/start gate.
- Identity recording for an already-started Remove uses explicit continuation semantics: it must match the operation's current pending phase, cannot authorize a new call, and may continue tracking/waiting after supervisor loss. A real boundary regression demonstrates that loss immediately after Popen cannot abandon its wait. `required=False` controls journal errors, not permission to dispatch after supervisor loss.
- Pre-proof snapshots freeze alias generation, block and operation ID. Full-library proofs and operation receipts list `reconciled_blocks` and exact retained sessions; a replaced operation is not acknowledged.
- Cached queue reasons name the retained CBB and expose a release entry in verification, ICDB and management waiting states. NativeQueue emits an updated notice only when the cached wait reason changes, so a task transitioning from running to retained cannot leave a hidden dead-end action. Queue predicates do no filesystem/native I/O.
- The queued release's mapping finalizer runs inside the serialized release action before it admits the waiting native ticket. The completion signal skips later destructive mapping cleanup. This prevents an old proof from clearing a new task's mapping retention. Both normal confirmation and native Cancel use this finalizer. Closing the residual panel does not mark the still-queued parent action finished.
## Validation and limits
The final focused integration run passed 211 tests. New tests were first observed failing for missing dispatch/receipt facts, dispatch veto after accepted Remove, missing queue explanations, changed wait reasons, stale mapping cleanup and dialog lifecycle boundaries. Independent review's two queue timing findings were corrected and re-reviewed with no remaining important introduced finding.
The first full run exposed one terminal-publication race (1,773 passed, one failed): management wrote FINISHED before its result receipt, allowing the orphan watchdog to exit between them. The result is now flushed before FINISHED. The regression accepts either the normal failure exit or orphan exit only after checking the saved result and the absence of a subsequent native action. A new test first reproduced the missing receipt at the FINISHED transition; 159 focused lifecycle/management tests then passed. The failed full XML is preserved separately, and the full suite is rerun on the corrected tree.
This does not require a final result before every orphan exit: PRECALL/POSTCALL may deliberately exit during an idle non-COM wait after supervisor loss, as requested by review 319. Such an exit is neither operation success nor resource-release proof. A missing final result is returned as `success=False, safe_to_continue=False` by the launcher, with the last call facts retained in the session journal for explicit reconciliation. FINISHED specifically is now published only after its final receipt; requiring a final receipt before all POSTCALL orphan exits would leave the original idle-orphan problem unresolved.
Full-suite counts, source hashes and immutable XML copies are in the accompanying validation JSON. Compilation/diff checks and icon equality are checked on the integrated tree. No frozen package was rebuilt; the old iconfix package does not contain this revision. No real Mentor Save/Close/Add/Remove or shared/network/cross-machine acceptance was performed. ICDB automatic Repair remains off. No merge to main, formal release, issue closure, native address/lock deletion, foreground mouse control, or native process termination is authorized by this validation.
@@ -0,0 +1,85 @@
{
"date": "2026-11-09",
"scope": "Latest comments #309/#310/#313 only; older deep-review backlog excluded by user",
"integration_head_before_validation_commit": "b1c36e3ca6d6680d0b656a725a1db871d5ffabbd",
"branch_heads": {
"codex/review-native-backend": "b7baa24b5ed87256fb846236b08c75ff02887e9c",
"codex/review-ui-integration": "e73f9c6edcb36a18fdd8a17b565cc1fb7c97f26b",
"codex/review-residual-release": "651b539d0ef491b3fd410031b9f981d71272aac9",
"codex/runtime-lifecycle-safety": "b1c36e3ca6d6680d0b656a725a1db871d5ffabbd"
},
"environment": {
"python": "3.12.10",
"windows": "10.0.26200"
},
"full_suite": {
"tests": 1742,
"failures": 0,
"errors": 0,
"skipped": 0,
"duration_seconds": 293.056,
"xml": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\2026-11-09-latest-stack-full-1742.xml",
"source_xml": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\cbb-refresh-verify-795-review-stack-fixes-full.xml",
"xml_sha256": "e00376d01fa26dd8e62d065d700f8cc200000665a6958536d109336119f8a0c0"
},
"focused_suites": {
"cbb-refresh-verify-795-backend-fixes": {
"tests": 94,
"failures": 0,
"errors": 0,
"skipped": 0,
"duration_seconds": 41.802,
"xml": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\2026-11-09-cbb-refresh-verify-795-backend-fixes-latest-focused-94.xml",
"source_xml": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\cbb-refresh-verify-795-backend-fixes-fixes-focused.xml",
"xml_sha256": "d8580bcc766c7ad6d3ce1d9de7d44b496afaaa8c3489cf01245f1d8a7d15500c"
},
"cbb-refresh-verify-795-ui-fixes": {
"tests": 43,
"failures": 0,
"errors": 0,
"skipped": 0,
"duration_seconds": 5.554,
"xml": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\2026-11-09-cbb-refresh-verify-795-ui-fixes-latest-focused-43.xml",
"source_xml": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\cbb-refresh-verify-795-ui-fixes-fixes-focused.xml",
"xml_sha256": "0ed470d756f0b387ca9fbf4f4891bd60abfc3faf535e67c06d6640557d627753"
},
"cbb-refresh-verify-795-review-stack": {
"tests": 136,
"failures": 0,
"errors": 0,
"skipped": 0,
"duration_seconds": 16.117,
"xml": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\2026-11-09-cbb-refresh-verify-795-review-stack-latest-focused-136.xml",
"source_xml": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\cbb-refresh-verify-795-review-stack-fixes-focused.xml",
"xml_sha256": "2087bfa02a0251deee535acda7454c0967b1f11aa5c7125296efc78c141fec59"
}
},
"compileall": "passed",
"diff_check": "passed",
"independent_review": "No important introduced findings before or after integration",
"controlled_private_desktop_fixture": "passed",
"tested_files_sha256": {
".gitattributes": "a68c9f317ed6e4131c72269b98403c7a3f9c08d5d03e710425301bd50273e791",
"src/cbb_refresh_verify/application.py": "b76ff891d77f0a7dba8feeb1f5ae5745ee83dee2fa46c585c187dbf9bb1c39d2",
"src/cbb_refresh_verify/library_lock.py": "8b3286d36ad686f65ec7e30eded9c4d0f8a629ca214b39812dfba496292da976",
"src/cbb_refresh_verify/residual_sessions.py": "b5d38a06cb2902c3612edb75b1ef83596524f02b6cf7890f6541f210d3e380ce",
"tests/test_residual_sessions.py": "f34c77a11d2a2b08c5670a9809474afd84b212016a27b104249e5fddb4675f1f",
"tests/test_review313_residual_isolation.py": "4052ca51c35787a8df9b63e79febd38b666da98c09976fffb34f4361381d8b4d",
"tests/test_review_foreign_release_evidence.py": "cbf2af02418f6bd3ec720a7a7658cf97f06d1a6e12009fdff2a99f465711591e"
},
"icon_sha256": {
"ico": "62d4472ebaa94f9b39aebae06459685d3322041d0e1a7ada9f368cfd816ba1d6",
"png": "840f657fd7c622d9afd78625a64ed8ad04c90a895dd3c9fd37eebae543cdebe8",
"svg": "cbba5b3f339cbeb5eb43012709237359442ef67c69fc5b6679a8ee4c0a1b7328"
},
"automatic_icdb_repair": false,
"candidate_rebuilt": false,
"not_verified": [
"Real Mentor operations",
"Frozen execution of this revision",
"Windows 10",
"Shared/network/cross-machine release"
],
"merged": false,
"issue_closed": false
}
@@ -0,0 +1,24 @@
# Latest PR review follow-up: 2026-11-09
The user selected the November 9 latest re-review comments, excluding the older November 6 deep-review backlog.
| PR/comment | Result | Source/validation |
| --- | --- | --- |
| #10 / #309 | Proof names foreign CBB checkpoints and distinguishes requested block from effective BLOCK/LIBRARY scope. | `b7baa24`; 94 backend/lease regressions passed. See `2026-11-09-pr10-review309.md`. |
| #11 / #310 | No new blocking finding; backend synchronized in `e73f9c6`. Existing pause semantics retained. | 43 UI/background/queue regressions passed on the UI branch. |
| #12 / #313 | Incomplete window reads isolated by PID; bounded repeated evidence; native Cancel retains preview scope. | `651b539`; 116 relevant regressions passed, including a real controlled private desktop/window. See `2026-11-09-pr12-review313.md`. |
| #13 integration | Both sides of the residual preview conflict retained: runtime write protection and per-PID read/scope metadata. | Integration merge `b1c36e3`; 136 focused integrated regressions passed. Full result and file hashes in the accompanying JSON. |
## Line endings
At the chain tip only, `application.py` is normalized from CRLF to LF. Replacing its LF bytes with CRLF reproduces its prior bytes exactly, and `git diff --ignore-space-at-eol` is empty for this file. `main_window.py` already uses LF. Two specific `.gitattributes` rules pin these files to LF; the global `* -text` rule and every other source file's line-ending policy remain unchanged. No repository-wide normalization or EditorConfig layer is added.
## Independent review
The reviewer found no important introduced P1/P2 in the backend/residual changes. Its focused check passed 45 tests with the private-desktop fixture excluded; the coordinator separately ran that actual fixture successfully. A second inspection of the integration confirmed that PR13's active-write, retained-session acknowledgment and durable release records coexist with the new evidence and window-read boundaries. No tests or native operations ran during that integration inspection.
## Reproduction
Use the existing `diagnostics/run_review_fixes.py` wrapper from the original workspace against each linked worktree; it isolates the test mutex and sets the intended source path/Qt offscreen. The final full run, compileall and git diff --check results are in `2026-11-09-latest-pr-stack-validation.json`, including evidence hashes and source/test file hashes.
Current ICO/PNG/SVG hashes match the previously corrected blue check/circular-arrow assets. No candidate executable or ZIP was rebuilt: the previous iconfix package does not contain these fixes. These are source/PR validations, not real Mentor/Windows 10/shared-network/cross-machine acceptance. ICDB automatic Repair remains OFF. PR #3 remains a historical umbrella already covered by #10–#13. The PRs stay Draft; no merge to main, formal release or issue closure is performed.
@@ -0,0 +1,7 @@
# PR #10 comment 309
Release proof now records the requested CBB, effective BLOCK/LIBRARY proof scope, identities/statuses of archived checkpoints and retained owners, and `reconciled_foreign_blockers`. For example, a BlockA operation which safely reconciles BlockB's historical checkpoint explicitly lists BlockB and the full-library proof. `data_recovery_confirmed` remains false. No extra gate, automatic backup or native action was added.
Three regressions first failed on missing evidence fields. After the fix, 94 backend/lease/release regressions passed; compileall and git diff --check passed. The tests use real temporary file/lease handles with the process-query boundary replaced; no production library was opened. Full regression is run again at the integrated chain tip. Real Mentor/shared-network acceptance and a rebuilt package remain outstanding.
Review: https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/10#issuecomment-309
@@ -0,0 +1,38 @@
# PR11 review 345 follow-up
Repeated CBB verification and ICDB check runs replace the previous result dialog
for the same operation. Closing a result retires its Qt widgets and action
references, while the check task keeps its requests, results, invalidated targets,
and recovery context. A result for the other operation stays available.
During application shutdown, the title and status show that background work and
path release are pending. After approximately 120 seconds a second notice says
the application is still waiting. This is a notice interval, not a deadline;
shutdown still waits for existing workers and never forces native tools to stop.
The batch CLI accepts an optional positive `--limit`:
```powershell
python -m cbb_refresh_verify --batch-lmc 'Z:\library\library.lmc' --limit 2
```
The limit selects the first requested names in their existing order. With no
`--block`, the existing non-READY selection supplies those names; repeated
`--block` options retain their explicit order. Omitting `--limit` retains the
existing all-requested behavior. `batch_selection` logs selected, total, and
remaining counts. On batch exceptions or unconfirmed native release,
`batch_drive_retained` identifies the drive, remote share, display library and
Mentor path, and directs the operator to the confirmed-release entry. It never
disconnects a mapping while a native writer may still be active.
Management change counts explicitly cover the compared engineering files only.
Removed from the standalone PR11 head: all five `if False` blocks for the later
release flow, and `_reload_after_maintenance` assignments and its unused reload
branch. The hidden release control and existing method signatures remain for
the later PR12 implementation. When integrating forward, preserve PR12's live
release/resume behavior and retire its dialog-owned action references as well.
Preserve PR13's asynchronous close and lease guards when adding the wait notice.
Validation uses the isolated offscreen review wrapper with fake native boundaries;
it does not open Mentor, kill native processes, invoke Repair, or back up a library.
@@ -0,0 +1,21 @@
# PR #12 comment 313
## Behavior
Window inspection takes a Windows thread snapshot for the requested processes, finds the desktops assigned to those threads, and opens only those desktops. Failures are recorded per affected PID outside the real window list. Unrelated inaccessible desktops no longer disable healthy rows. Preview, cooperative close, forced close, native Cancel and save-prompt responses all use the same per-PID incomplete-read fact during their fresh action-time observation.
Unknown or active native writes still prohibit forced termination. The review's suggested blanket force override for unreadable windows is not implemented: exact identity and private-desktop ownership do not establish that Save/Repair has returned. A failed read names that session and can be retried, while other sessions remain actionable.
Disk evidence retains window identity and a UTF-8 output excerpt of at most 4,096 bytes, original output length and SHA-256. It omits child-control trees and duplicated native-activity output. The complete preview remains in memory for the existing details action.
Native Cancel carries the preview's `block_names` and checks the same Work + SymbolLibs scope before the exact-process/control checks. It no longer silently widens a scoped preview to the full library.
## Verification
The initial five regressions failed against the old implementation. After the fix, 116 residual/backend/UI/background-queue regressions passed; compileall and git diff --check passed. This includes a real controlled Windows window on a private CbbManagement desktop, read from another process, with clean fixture exit and no Mentor/native database actions.
The real fixture caught a platform limitation missed by API doubles: EnumThreadWindows silently omits windows on another desktop. The final implementation maps threads to desktop names and uses OpenDesktop/EnumDesktopWindows for only those mapped desktops. GetThreadDesktop's borrowed handle is never closed; owned OpenDesktop handles and the thread snapshot are closed.
Independent review and the full integrated-chain run are recorded at the final PR #13 tip. Older deep-review comments #297-#300 remain outside this explicitly selected scope. No candidate was rebuilt; real Mentor, Windows 10 and shared/network release scenarios remain unverified.
Review: https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/12#issuecomment-313
@@ -0,0 +1,7 @@
# PR #12 comment 348 follow-up
The review's process-isolation and oversized-output findings target the earlier `5eb4b2d` tree; both are addressed by `651b539` and remain covered by the real private-desktop and bounded-evidence regressions.
The remaining consent/result presentation gap is fixed here. The residual table retains its original six columns and adds confirmation scope and actual result. A row that was allowed at confirmation but rejected after a fresh activity/identity check now shows both facts beside its reason. The result message also counts confirmation-allowed rows and how many remain unreleased; it does not reinterpret preview consent as successful execution or bypass an active-write guard.
The new regression failed before the change. The release-dialog, residual follow-up and private-desktop regression group passed 35 tests. No production/native database action was performed. Integrated full validation is recorded at the PR #13 tip; no candidate build, merge or field acceptance is implied.
@@ -0,0 +1,53 @@
{
"date": "2026-11-09",
"review_url": "https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/13#issuecomment-319",
"base_commit": "96e179ba6529eefd1401a012155cd7b7cbc0ffc3",
"environment": {
"python": "3.12.10",
"windows": "10.0.26200"
},
"full_suite": {
"tests": 1732,
"failures": 0,
"errors": 0,
"skipped": 0,
"duration_seconds": 501.166,
"xml": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795\\diagnostics\\cbb-refresh-verify-795-review-stack-fixes-full.xml",
"xml_sha256": "578dd4b01db5c18ce657430ce3b2cd2dba8516d8191bdd31bf4e0df78187b3ba",
"warnings": [
"Existing pywinauto Revert to STA COM threading mode warning"
]
},
"focused_suite": {
"passed": 79
},
"compileall": "passed",
"diff_check": "passed",
"independent_review": "No remaining important findings in the scoped diff; durable completion feedback addressed",
"tested_files_sha256": {
"src/cbb_refresh_verify/desktop_worker.py": "173a951d510e78e088c439768ecfe31ab198723e7717b314b4a74a9d590a1fe0",
"src/cbb_refresh_verify/library_lock.py": "391fb2f0b25f400676ba230472fb3bbd4a0d994ed5a8dfee46f8f32c7f198219",
"src/cbb_refresh_verify/mentor_adapter.py": "1fc8c98d5508120e03bf001a42dd3e0da7b165eb18c90f1aed83147d17db8638",
"src/cbb_refresh_verify/native_management.py": "96c2fe613a657d110302c8f7cc033b91ef256e168036f4794eb39be475e69de8",
"src/cbb_refresh_verify/residual_sessions.py": "a19622be3e959a7e64bf9de953450203499fb82618da8537307e5f61ef4d793b",
"src/cbb_refresh_verify/session_record.py": "c4c5771d11d7469980c02f2fb264b26e5ee26634a37daae008dbb65097ab4e0c",
"src/cbb_refresh_verify/ui/main_window.py": "e46096ce784933f3ab47e4c316f3c84e3e5cf07242489323f6234b61aeb22f80",
"src/cbb_refresh_verify/ui/residual_release_dialog.py": "ec7566bac6d92845f98933bffd4c06f31eeec42946e69699aff94e31f9a8baef",
"tests/test_incident_safety.py": "52f3ccb04db8971bf38628c2bd334d4fd79795e97de6538f6f66cb35f7123ece",
"tests/test_integration.py": "9e9d30692607c889e1e3744a63c79c68e3af7d4c86804e0bcebe48e68765086c",
"tests/test_management_supervision.py": "107196ce8a00dd76e7e3e83fb47ae2e67c857fa234492ba593bf71a7d7e1c5e5",
"tests/test_residual_release_dialog.py": "7d2bf95afc2f9112b9433c7d2fefde3b35b5e85a70a6bd506471c039e2505509",
"tests/test_review_319.py": "831fd34d426abb4436b9132bfb7ed754994155a34ac60a59378dee0964b54d28"
},
"not_verified": [
"Frozen execution of this revision",
"Real Mentor native writes/save/close during supervisor loss",
"Windows 10",
"Shared network/cross-machine release",
"ICDB field recovery"
],
"automatic_icdb_repair": false,
"candidate_rebuilt": false,
"merged": false,
"issue_closed": false
}
@@ -0,0 +1,33 @@
# PR #13 review 319 validation
Review: https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/13#issuecomment-319
Baseline: `96e179ba6529eefd1401a012155cd7b7cbc0ffc3`.
## Changes
- Successful explicit resource proof acknowledges the retained queue entries captured before proof. The alias-set identity is a generation token: unrelated libraries and newer handoffs remain retained. Scoped proof acknowledges only the matching block.
- Final observer outcomes remain available in the residual dialog after its observer ends. The existing refresh/confirm actions provide a running-session retry path. Successful release records a matching operation's RELEASED observation, so history refresh does not resurrect it as unconfirmed. Journal failure is reported separately from the established release fact.
- Discovery observes process identities only. It never opens production databases exclusively, deletes coordination checkpoints, closes clients, or replays operations. Startup observers use the same read-only mode; only explicit release or already-authorized task reconciliation can establish resource release.
- Supervisor loss sets cancellation independently of evidence-write success. An idle worker exits with code 75 only when its matching journal and synchronized in-memory phase both confirm no pending native write and every registered helper has exited. Missing/unknown/stale-active state retains the worker. The watchdog rechecks returned calls and cannot authorize a subsequent native write/helper. Native children are never terminated by this exit; worker exit is not database-release or Save-success evidence.
- Worker phase persistence and watchdog journal reads share a gate to avoid a Windows sharing violation between reading the JSON and atomically replacing it. The dead `_terminate_pid` function is removed; cleanup tests observe actual process-termination boundaries.
## Verification
Commands run through the existing test wrapper with a dedicated mutex namespace and Qt offscreen; no user/native Mentor process is part of these fixtures.
```
python -u -X utf8 diagnostics/run_review_fixes.py "F:/Sync Tool Skyworth/Work Maps/cbb-refresh-verify-795-review-stack"
python -X utf8 -m compileall -q src tests
git diff --check
```
The focused session/watchdog/management/dialog/lease regression run passed 79 tests. The initial regressions were observed failing before the corresponding fixes: explicit proof left a library unavailable, history scan attempted release, orphan idle workers stayed alive, residual refresh omitted runtime records, and successful release reappeared as unconfirmed. Real harmless Windows subprocess fixtures cover PRECALL/POSTCALL, NATIVE_CALL/CLOSING, unknown/missing/stale journals, failed cancel/evidence writes, and rechecking after return.
Full-suite result is recorded in `2026-11-09-review-319-validation.json`. Independent review found no remaining P1/P2 in this scoped diff; its observation about durable completion was addressed and regression-tested.
## Limits
This is source/PR validation. The earlier `0.2.4b2-iconfix-44bb734` executable has not been rebuilt with these changes and must not be used as evidence for this revision. Its icon assets are preserved.
Not validated here: frozen execution of this revision, real Mentor Save/Close or library writes during supervisor loss, Windows 10, shared/network or cross-machine release. ICDB automatic Repair remains disabled. No field-recovery acceptance, merge, formal publication or issue closure is implied.
File diff suppressed because it is too large. Load diff
@@ -0,0 +1,76 @@
# 2026-11-11:逐按钮模拟操作验证
基线:42e87190263453c8e3a7001d5768741b3b3a2c38。
全量 Python 测试 **1,835 passed**,只有既有的 STA COM 模式警告;Windows 私有桌面的界面组 **276 passed** 并正常退出。记录了 **281 次点击信号**、**78 种控件标识**和 **28 张代表性截图**。新增测试包含真实 Qt 文件选择和确认按钮、连点、停止、取消、窗口关闭和动态结果入口。
## 逐按钮模拟操作结果
Windows 原生 Qt 控件,独立私有桌面。48 个应用按钮/复选项全部触发点击信号,另验证 3 个菜单页签;276 项界面相关测试通过。
文件选择和确认框使用真实 Qt 对话框及其按钮;原生读写、恢复、保存、取消、强制结束均使用隔离替身。通过表示界面连接、参数、状态及错误处理通过,不表示真实 Mentor 操作通过。
| 控件 | 按钮 | 结果 | 验证点 |
| --- | --- | --- | --- |
| BackupRecoveryDialog.browse_button | 选择… | 通过 | 真实备份工程选择框打开/取消 |
| BackupRecoveryDialog.restore_button | 恢复并校验 | 通过 | 无效来源拒绝;确认/拒绝;显示受控恢复结果 |
| BackupRecoveryDialog.close_button | 关闭 | 通过 | 空闲关闭恢复窗口 |
| BackupRecoveryDialog.cancel_queue_button | 取消排队 | 通过 | 取消等待票据,不调用恢复后端 |
| MainWindow.native_lm_verify_check | 使用 Library Manager 原生 Verify(Win10 兼容测试) | 通过 | 勾选状态切换,原生 Verify 选项保持独立 |
| MainWindow.advanced_toggle | 高级设置 | 通过 | 展开和收起高级设置 |
| MainWindow.select_visible_button | 选择可见项 | 通过 | 只勾选可见条目 |
| MainWindow.clear_selection_button | 清空选择 | 通过 | 清空勾选 |
| MainWindow.load_button | 读取 CBB | 通过 | 读取当前库并更新清单 |
| MainWindow.browse_button | 选择 | 通过 | 真实选择框打开/取消;取消不触发读取 |
| MainWindow.selected_batch_button | 批量校验已选 | 通过 | 真实确认/拒绝;仅执行可见非 READY 已选项 |
| MainWindow.batch_button | 校验可见非 READY | 通过 | 真实确认/拒绝;只执行可见非 READY 项 |
| MainWindow.icdb_repair_button | ICDB 批量检查 | 通过 | 开始独立 ICDB 检查,不启用自动 Repair |
| MainWindow.backup_restore_button | 从备份恢复 CBB… | 通过 | 为当前非 READY 目标打开备份恢复窗口 |
| MainWindow.reimport_button | 选择工程副本恢复失败项… | 通过 | 副本恢复仍绑定失败 CBB |
| MainWindow.stop_button | 当前项完成后停止 | 通过 | 设置停止标志,后续目标不再执行 |
| MainWindow.logs_button | 打开日志目录 | 通过 | 打开正确目录;目录及系统打开失败时显示错误 |
| MainWindow.recovery_button | 打开异常证据 | 通过 | 打开正确证据路径;系统打开失败时显示错误 |
| MainWindow.native_backup_button | 版本历史与备份… | 通过 | 只启动一次;收到关闭回执后恢复可用 |
| MainWindow.icdb_stop_button | 停止检查 | 通过 | 停止当前后台检查,不再打开后续项 |
| MainWindow.add_button | 新增 CBB… | 通过 | 打开新增操作窗口 |
| MainWindow.update_button | 用工程替换已选 CBB… | 通过 | 打开所选目标的更新窗口 |
| MainWindow.remove_button | 移除已选… | 通过 | 打开所选目标的移除窗口 |
| MainWindow.release_library_button | 释放残留会话… | 通过 | 打开当前库的释放预览 |
| MainWindow.queue_release_button | 查看残留会话… | 通过 | 打开等待库的释放预览,关闭面板不误解锁任务 |
| MainWindow.runtime_log_toggle | 查看运行日志 | 通过 | 显示/隐藏运行输出 |
| ManagementDialog.sources_button | 选择来源工程… | 通过 | 真实工程选择框打开/取消并更新来源行 |
| ManagementDialog.auto_verify | 本批次上传后自动校验 | 通过 | 切换本批次自动校验选项 |
| ManagementDialog.release_wait_button | 查看残留会话… | 通过 | 传出等待释放的目标库 |
| ManagementDialog.logs_button | 查看日志 | 通过 | 显示/隐藏管理日志 |
| ManagementDialog.execute_button | 预览并确认执行… | 通过 | 真实确认/拒绝,重复点击不重复执行 |
| ManagementDialog.stop_button | 当前项结束后停止 | 通过 | 设置管理任务停止标志 |
| ManagementDialog.close_button | 关闭 | 通过 | 空闲关闭,执行中禁用点击不打断任务 |
| PcbRestoreDialog.manual_button | 最后手动保存 | 通过 | 只选择最后手动保存版本,关闭不覆盖已有选择 |
| PcbRestoreDialog.auto_button | 自动保存 | 通过 | 只选择自动保存版本,关闭不覆盖已有选择 |
| PcbRestoreDialog.cancel_button | 取消 | 通过 | 取消选择,不默认回退版本 |
| ResidualReleaseDialog.refresh_button | 重新检查 | 通过 | 重新扫描预览,不执行释放 |
| ResidualReleaseDialog.confirm_button | 确认释放 | 通过 | 传递当前确认范围,显示实际释放结果 |
| ResidualReleaseDialog.prompt_button | 处理保存提示… | 通过 | 真实是/否/取消,传递对应保存选择 |
| ResidualReleaseDialog.details_button | 查看原生提示… | 通过 | 显示原生提示全文和详细信息 |
| ResidualReleaseDialog.cancel_native_button | 请求原生取消… | 通过 | 真实确认/拒绝后发送受控取消请求 |
| ResidualReleaseDialog.force_button | 强制结束客户端… | 通过 | 真实确认/取消后传递强制标志;未调用实际强杀 |
| ResidualReleaseDialog.close_button | 关闭 | 通过 | 空闲关闭;执行中请求停止并等待当前项返回 |
| Result.batch_release | 查看残留并释放… | 通过 | 从批量失败结果打开当前库释放入口 |
| Result.batch_resume | 继续未执行项 | 通过 | 释放确认后只执行此前未执行的目标 |
| Result.batch_copy | 选择副本恢复… | 通过 | 兼容维护结果中的副本恢复仍绑定失败工程 |
| Result.icdb_release | 查看残留并释放… | 通过 | 从 ICDB 结果打开对应库释放入口 |
| Result.icdb_copy | 选择副本恢复… | 通过 | 从 ICDB 结果打开对应失败工程的副本恢复 |
每项对应的测试编号、点击次数、完整事件记录及代表性截图见同目录 button-matrix.json / windows-all.json。
本轮额外修复:打开日志目录时,目录创建失败或系统拒绝打开不再向 Qt 事件循环抛出未处理异常,改用现有错误提示。
未验证:真实 Mentor 写库、ICDB 修复、Windows 10、跨机器共享库。
## 验证边界与构建记录
实际原生写库、历史回退、保存、原生取消及强制结束均使用隔离替身。未触碰现场工程或中心库,未证明 ICDB 原生修复成功;Windows 10 和跨机器共享库仍未验收。
首次候选构建因 C 盘空间不足在收集文件阶段失败,未计为通过。本轮未完成构建已迁至 F 盘继续,原候选和测试证据保留。最终构建及冻结验证结果见相邻验证 JSON。
逐按钮的测试编号与点击次数、界面 XML、截图路径和文件哈希保存在相邻验证 JSON 及其引用的本机证据中。截图只提供代表性界面,逐项结论以测试断言和点击记录为依据。
@@ -0,0 +1,210 @@
{
"date": "2026-11-11",
"baseline": "fdd07224aac99bee8d1f29e8e82d6a2c49964c87",
"review_comment": "https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/13#issuecomment-381",
"full_suite": {
"tests": 1789,
"failures": 0,
"errors": 0,
"skipped": 0,
"seconds": 341.04
},
"focused_final": {
"tests": 57,
"failures": 0,
"errors": 0,
"skipped": 0,
"seconds": 10.935
},
"windows_native_direct_runs": [
{
"tests": 21,
"failures": 0,
"errors": 0,
"skipped": 0,
"seconds": 14.625
},
{
"tests": 21,
"failures": 0,
"errors": 0,
"skipped": 0,
"seconds": 11.765
}
],
"windows_platform": {
"platform": "windows",
"desktop": "CbbUiReview381_ea1f0bcdc2",
"exit_status": 0,
"total": 21,
"failures": 0,
"mentor_calls": false
},
"inconclusive_windows_controller_run": {
"controller_exit_code": 1,
"assertion_total": 21,
"assertion_failures": 0,
"counted_as_pass": false,
"controller_cause_confirmed": false,
"followup": "Direct pytest launch and stack diagnostics; two direct runs exited 0."
},
"negative_control": {
"expected_failure": true,
"result": {
"tests": 1,
"failures": 1,
"errors": 0,
"skipped": 0,
"seconds": 3.75
},
"method": "Only baseline _queued_action restored in test memory; no source mutation."
},
"compileall_passed": true,
"diff_check_passed": true,
"tested_production_sha256": "a18d36f893f6e23b8e52b3d7e0b8cfba9ebfa08abb44acb54219cc092b318942",
"full_run_file_hashes": {
"src/cbb_refresh_verify/ui/main_window.py": "a18d36f893f6e23b8e52b3d7e0b8cfba9ebfa08abb44acb54219cc092b318942",
"tests/test_review381_queue_release.py": "56661d607b55d1284fad69105d4457ab6816696a998188d6986ed67765b326e0",
"src/cbb_refresh_verify/assets/cbb_refresh_verify.ico": "62d4472ebaa94f9b39aebae06459685d3322041d0e1a7ada9f368cfd816ba1d6",
"src/cbb_refresh_verify/assets/cbb_refresh_verify.png": "840f657fd7c622d9afd78625a64ed8ad04c90a895dd3c9fd37eebae543cdebe8",
"src/cbb_refresh_verify/assets/cbb_refresh_verify.svg": "cbba5b3f339cbeb5eb43012709237359442ef67c69fc5b6679a8ee4c0a1b7328"
},
"final_file_hashes": {
"src/cbb_refresh_verify/ui/main_window.py": "a18d36f893f6e23b8e52b3d7e0b8cfba9ebfa08abb44acb54219cc092b318942",
"tests/test_review381_queue_release.py": "e568a2701337be068af91a03ff3451d82c705109a7b62c3c11663a42bcf552b5",
"src/cbb_refresh_verify/assets/cbb_refresh_verify.ico": "62d4472ebaa94f9b39aebae06459685d3322041d0e1a7ada9f368cfd816ba1d6",
"src/cbb_refresh_verify/assets/cbb_refresh_verify.png": "840f657fd7c622d9afd78625a64ed8ad04c90a895dd3c9fd37eebae543cdebe8",
"src/cbb_refresh_verify/assets/cbb_refresh_verify.svg": "cbba5b3f339cbeb5eb43012709237359442ef67c69fc5b6679a8ee4c0a1b7328"
},
"test_refinement_after_full": "Only one test gained an admission barrier; production unchanged; focused and Windows rerun.",
"candidate_rebuilt": true,
"candidate_executable": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-candidate\\dist\\CbbRefreshVerify-Background\\CbbRefreshVerify.exe",
"bundle": {
"module_count": 62,
"source_and_frozen_code_equal": true,
"source_sha256_checked": true,
"packaged_assets_equal": true,
"exe_icon_images_equal": true,
"executable_sha256": "71812c1fdfa6522fad4468f4ca4c7d8f77aa9c6880b70f0d8b7ad58a31d55ec2"
},
"frozen_smoke": {
"gui": {
"status": "passed",
"exit_code": 0
},
"verify-worker": {
"exit_code": 1,
"status": "passed",
"evidence": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-candidate\\smoke-319c4ff4\\verify-worker"
},
"management-worker": {
"exit_code": 2,
"status": "passed",
"evidence": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-candidate\\smoke-319c4ff4\\management-worker"
},
"owner-probe": {
"exit_code": 0,
"status": "passed",
"evidence": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-candidate\\smoke-319c4ff4\\owner-probe"
},
"dialog-agent": {
"status": "passed",
"exit_code": 0,
"scope": "empty PID set, observe only"
},
"module_origins": {
"status": "passed",
"module_count": 62,
"source_root": "F:\\Sync Tool Skyworth\\Work Maps\\cbb-refresh-verify-795-review-stack\\src"
},
"version": "0.2.4b2"
},
"independent_review": {
"scope": "review381 and directly related stale/cross-task release UI",
"reviewer": "review381_final",
"reviewed_files": [
"src/cbb_refresh_verify/ui/main_window.py",
"tests/test_review381_queue_release.py"
],
"finding": "foreground completion erased independent background wait",
"followup": "fixed; both completion orders reproduced failing then passing",
"final_review": "no additional findings",
"review_native_calls": false
},
"native_mentor_operations_verified": false,
"icdb_repair_accepted": false,
"field_library_modified": false,
"windows10_verified": false,
"shared_library_verified": false,
"merged": false,
"artifacts": {
"review381-baseline.xml": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-baseline.xml",
"sha256": "f224548b02b5841f7d3148375eea50f6505c83af912b66a7d67c5d572bc287b3"
},
"review381-red-controlled.xml": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-red-controlled.xml",
"sha256": "15c620cb4ec251501d83fd7ebc4ee654df8b0e4b7aa8f8ccb2d7af30ec162c5f"
},
"review381-interleaved-red.xml": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-interleaved-red.xml",
"sha256": "fea4342c4bdab3b03e70f70585e66281188ac80ae7613e4bf7db9fcd616ce1f5"
},
"review381-completion-red.xml": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-completion-red.xml",
"sha256": "6dab11fbba7878db81885f5d11fadc5c80728a9dab3cab8edd0883cb7015d3e2"
},
"review381-full.xml": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-full.xml",
"sha256": "9f73a2dddae8831c2bad4ae1fb2f2f509b1981508a512fffb208dce1a858860f"
},
"review381-focused-barrier.xml": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-focused-barrier.xml",
"sha256": "ec7fd3dc3ce869b97ed3f22e7e9528f60dc8512a89e451eae434fe665bae838a"
},
"review381-admission-negative-control.xml": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-admission-negative-control.xml",
"sha256": "ecb66304ee5cc1be9c19c4ba33c57a7bd86e65b156a959fa5665cba2274679e9"
},
"review381-windows-direct-first.xml": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-windows-direct-first.xml",
"sha256": "3ba672a9190a5ec715ae99db01c0faca1b3bfcfe9d25a25ba46309af7f5c1880"
},
"review381-windows-direct-first-platform.json": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-windows-direct-first-platform.json",
"sha256": "a255f39b48fe69656c553b6a4feff6f905beb1e18ba2de9724454f15d3d8ea56"
},
"review381-windows.xml": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-windows.xml",
"sha256": "a2d53895c5afaf0907a670d544cd98db6c786bbf75f2e520ac0355d3c4a88ff2"
},
"review381-windows-platform.json": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-windows-platform.json",
"sha256": "e06d76c1c2c297f6d4e11f76a903666a1bdf47cdc81a26a295c2040ef9381bbb"
},
"review381-independent-review.json": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-independent-review.json",
"sha256": "155c0b1c4ae3ecb5b94b1a2d321e47b96a4885f18cc369f36c82408b20959981"
},
"review381-candidate/source-sha256.json": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-candidate\\source-sha256.json",
"sha256": "6028d5cb3f6eeec68ae938fb73223d5343de7c8671ba3b77d9a1d1e6faa37608"
},
"review381-candidate/bundle-verification.json": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-candidate\\bundle-verification.json",
"sha256": "08117158e59192c7ee7f0538968b24d5e4cb8e9f8f230636bd009aa39ffbc4b5"
},
"review381-candidate/smoke-319c4ff4/result.json": {
"path": "C:\\Users\\ABS_Windows\\.codex\\visualizations\\2026\\11\\02\\01a19b86-b23f-7fe0-9407-45a4303e10e3\\review381-candidate\\smoke-319c4ff4\\result.json",
"sha256": "3cefd58746044c0f48455aba9796b0a30687ee5a30084fa1948449e5094c1a10"
}
},
"committed_file_hashes": {
"src/cbb_refresh_verify/ui/main_window.py": "a18d36f893f6e23b8e52b3d7e0b8cfba9ebfa08abb44acb54219cc092b318942",
"tests/test_review381_queue_release.py": "e0d62424cbf70ec9573db78ede0a4c6534775399194eed4c6b28ba8d468307be",
"src/cbb_refresh_verify/assets/cbb_refresh_verify.ico": "62d4472ebaa94f9b39aebae06459685d3322041d0e1a7ada9f368cfd816ba1d6",
"src/cbb_refresh_verify/assets/cbb_refresh_verify.png": "840f657fd7c622d9afd78625a64ed8ad04c90a895dd3c9fd37eebae543cdebe8",
"src/cbb_refresh_verify/assets/cbb_refresh_verify.svg": "cbba5b3f339cbeb5eb43012709237359442ef67c69fc5b6679a8ee4c0a1b7328"
},
"post_validation_whitespace_cleanup": "Trimmed one trailing blank line in test file; compiled test code equal; staged diff check rerun."
}
@@ -0,0 +1,43 @@
# PR #13 评论 381:等待入口与交错操作验证
对应审查:<https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/13#issuecomment-381>。基线为 `fdd07224aac99bee8d1f29e8e82d6a2c49964c87`。
## 修复行为
校验获得原生执行机会后,立即刷新执行状态和残留入口。入口显示及点击均复核现有的内存保留状态,已释放的库不会因旧消息或延迟点击重新打开释放面板。窗口进入关闭流程后,不接受新的释放面板请求。
等待消息在覆盖当前目标前,同样检查消息所指库现在是否仍被保留。另一个库的普通排队消息、已过期的保留消息或 ICDB 开始事件,不会抹掉仍有效的等待入口。
独立复查补充发现:前台任务结束会清除另一个库正在等待的后台 ICDB 入口。现在前台结束时保留该后台任务的有效入口;后台取消或结束时,移除属于它的过期入口,同时保留同库前台仍需要的入口。路径比较使用已知别名,不查询磁盘、网络或 WMI。
没有新增库锁、占用扫描、轮询线程或强制结束逻辑。原生执行仍串行;未确认释放时的同库保护和自动 Repair 关闭状态不变。本次不调整整批调度。
## 实际执行的验证
| 验证 | 结果 |
| --- | --- |
| 既有队列入口基线 | 7 通过 |
| 原问题及交错操作,修复前 | 7 失败、2 通过;随后扩展的交错用例另有 3 项预期失败 |
| 前台结束/后台取消两种顺序,修复前 | 2 项均失败 |
| 全量 Python 测试 | 1,789 通过、0 失败、0 跳过;既有 STA COM 模式警告 1 项 |
| 加强并发屏障后的相关回归 | 57 通过 |
| Windows 原生 Qt 控件,独立私有桌面 | 21 项通过;直接启动的最终验证记录另存 JSON/XML |
| 负向对照 | 仅在测试进程内恢复旧的队列接纳方法,新用例稳定失败;不修改源文件 |
| 编译及差异检查 | `compileall` 和 `git diff --check` 通过 |
| 新冻结候选包 | 5 项冒烟通过;62 个包内模块与源码一致;包内图标及 EXE 图标资源一致 |
全量测试后仅加强了一个测试的时序屏障,业务源码没有再变化;加强后的测试文件已重新执行相关回归和 Windows 控件验证。全量时与最终测试文件的哈希分别保留。
新增的 14 项用例覆盖:执行尚未结束时隐藏旧入口;五次延迟点击;过期等待消息;独立库事件交错;反复打开、关闭和重开面板;重复点校验后停止再点释放;前后台任务先后结束;关闭过程中送达的点击。控件点击使用真实 PySide6 控件与事件循环;库数据和原生扫描采用隔离替身,不操作现场库。
一次较长启动链的 Windows 验证中,21 项断言均通过,但外层控制器返回 1,未计为完整通过。随后缩短验证启动链,直接启动 pytest,并加入超时堆栈记录;直接运行的正常退出结果单独保留。不能据此把那次控制器异常认定为已复现的业务故障,也不隐去该次未完成验证。具体记录见相邻验证 JSON。
## 候选包与边界
新包位置记录于 `2026-11-11-pr13-review381-validation.json`,构建和验证产物保存在本机 Codex visualizations 目录。EXE SHA-256:
`71812c1fdfa6522fad4468f4ca4c7d8f77aa9c6880b70f0d8b7ad58a31d55ec2`
冻结冒烟包括 GUI 创建及正常退出、启动前取消的 Verify worker、桌面不匹配时拒绝操作的管理 worker、空所有者查询,以及私有桌面上空 PID 范围的弹窗观察。后四项是受控边界验证,不代表真实原生操作成功。
本轮未执行真实 Mentor Save/Close/Add/Remove 或 ICDB 修复,未写入现场中心库;也未验证 Windows 10、跨机器共享库或现场故障一致副本。不能将本次界面和生命周期验证作为 ICDB 自动修复成功的依据。不合并发布、不关闭现场工单。
@@ -0,0 +1,235 @@
# 当前程序流程、阻塞点与错误节点
> 证据路径说明:本文的本机绝对路径、日志名、截图、测试 XML 和冻结包指向验收机器上的原始材料,未纳入 Git;仓库仅保留说明和验证摘要。
> 这是修改前 `42d0ceb` 的流程快照。后续修复及真实复测见 [占用范围验证](2026-11-12-scoped-release-validation.md)。
依据提交 `42d0cebc54191681b53dcecb120f53236c1963bb` 的代码及 2026-11-12 冻结程序原生验收记录。本文件只分析当前行为,没有实施下面建议的调整,也没有新增原生验收。
后续只读核查已用外部软件自身日志确认 E01 的两个未知阻塞进程来自 DxPartReplace 测试,创建时间、父进程和私有桌面交叉匹配;外部工程引用原库,本次 CBB 验收使用隔离库。详见 [进程来源核查](2026-11-12-process-origin-audit.md)。这进一步确认了并行软件触发本程序范围判定漏洞,不能将其解释成本程序自身客户端未退出。
红色 E 节点表示真实验收已经发生、仍未解决的问题;橙色 B 节点表示代码中确实存在的等待、拒绝或失败分支,但不表示每个条件都已复现;绿色 F 表示本轮修复并复测通过的位置。各入口在 UI 中独立,底层原生调用仍共享串行通道。
## 一、最需要处理的阻塞链
```mermaid
flowchart LR
A[单个 CBB 失败或被停止] --> B[PCB 或助手未确认退出]
B --> C[safe_to_continue 为 false]
C --> D[B02 同库后续项被跳过]
C --> E[B01 后续任务等待库可用]
E --> F[用户打开释放窗口]
F --> G[正常关闭本工具 PCB]
G --> H[E01 其他任务的未知会话也被纳入必须退出的集合]
H --> I[整库释放仍为 PARTIAL]
I --> E
classDef error fill:#fee2e2,stroke:#dc2626,color:#7f1d1d
classDef risk fill:#fff3cd,stroke:#d97706,color:#78350f
class H,I error
class D,E risk
```
这条链中的核心缺陷是:**当前项是否失败、当前项是否留下占用、整个库能否继续、其他工具是否正在运行,没有彻底分开。** 某项失败但 `safe_to_continue=true` 时,当前代码会继续处理下一项,真实多来源新增已验证这一点;`false` 则会阻止同库后续项,不能称为“所有失败都不会影响整批”。
## 二、阻塞与失败清单
| 标记 | 发生位置与触发条件 | 当前后果、影响范围 | 证据与处理方向 |
| --- | --- | --- | --- |
| **E01 已复现、未修复** | 释放集合包含 scope_confirmed=false 的其他任务进程;释放函数要求全部观察项 RELEASED;底层证明也把没有目标路径的进程作为可能占用 | 本工具 PCB 已退出,整库仍 PARTIAL;保留状态无法清除,同库后续任务继续等或被跳过 | R7 的 PCB 21856/21248 均退出;其他任务 viewdraw 29832 为 WAITING。必须区分观察项与本库目标,并用目标资源证明决定能否继续,不能直接结束其他任务 |
| **E02 已复现、未走通** | PCB 启动出现原生“确认无其他用户,以读写权限继续”提示 | 当前打开判为 DATABASE_SAFETY_STOP;普通校验未 READY,ICDB 检查无法确认,备份副本不安装。关闭成功时本项失败仍可安全继续 | 该正文不直接证明 ICDB 损坏。代码有受限的修复后授权分支,但不是普通打开时通用的用户确认入口 |
| **E03 已复现、未修复** | A/B 来源工程的原生 EDIF 导出器退出 0xC0000005 | 本项上传失败;退出及资源释放证明成功后下一项继续,不能跳过身份比对报告成功 | 相同批次 C 能新增成功。需进一步定位原生导出与输入兼容性 |
| **B01 可长期排队** | NativeQueue 没有排队总超时;等待 active=false 且本库 library_available=true | UI 线程保持运行,但普通校验、管理、刷新等任务可能一直显示等待;可取消排队。活动任务未返回时,即使不同库也不能使用同一个原生通道 | 属于后台任务等待,不等于界面卡死;不同库且条件满足的票据可以越过暂不可用票据。需要展示等待对象、阶段及下一步,而不是单一“忙” |
| **B02 失败传播到同库** | run_batch、run_management_batch、IcdbCheckTask 遇 safe_to_continue=false;未知异常也可能保守置 false | 同库后续项 SKIPPED;其他库不因这个集合被跳过;CANCELLED/QUEUE_STOP 或用户停止会停止余下队列 | 现有规则确实仍会让单项占用影响本库整批。后续需分辨仅此工程占用与共享库占用,不能盲目把 false 改 true |
| **B03 执行互斥与协调锁** | Windows 会话级 Local/CbbRefreshVerify795 被占用;库协调锁被持有;历史 blocked/pending 记录存在或身份不符 | 直接拒绝本次操作,或先复核旧记录;未确认时保留本库协调锁/记录 | 多个应用实例、独立上传工具、崩溃遗留均可能触发。历史记录本身不是仍在写库的证明 |
| **B04 地址、数据库、在途调用** | 存在 sAddress.adr;数据库/LMC/Layout 独占读失败;仍有本库原生写入或导出副本所有者 | 当前操作或释放证明失败;真实在途写入应继续保护。陈旧地址和不完整日志也可能导致无法确认 | 当前主要哈希排除地址文件,但占用判断仍会使用它们;不能用“哈希忽略了”推断“占用也已忽略” |
| **B05 原生调用未返回** | Add/Remove/COM 调用或原生进程超过监督时限 | 有些监督入口返回超时并保留真实 worker;真实写入继续观测;CleanWGLib 超时后仍等待其自然退出,不能把超时当释放 | 可能同时占着执行通道、库锁或保留状态。需要区分取消未开始项、请求原生取消、仍在写入、可以关闭的客户端 |
| **B06 原生版本窗口仍打开** | iCDB Project Backup 原生窗口尚未关闭,或关闭结果未确认 | 同库校验/恢复受到限制;同库 ICDB 后台检查会暂停,关闭并确认后恢复;不是自动修复失败 | 这是实际原生编辑会话的互斥。不能通过关闭应用或删除记录假装窗口已关闭 |
| **B07 格式、能力与来源校验** | 缺失/越界引用、不支持的版本或接口、目标分区不唯一、来源同名不匹配、未完成变更记录仍需恢复 | 本项拒绝;新增遇同名冲突明确失败;未完成写操作不会在重启后自动重放 | 这些多数属于输入/恢复边界;要给出具体原因和可纠正入口,不能混成 ICDB 问题 |
| **B08 来源/目标内容变化** | 暂存复制前后文件集合或主要哈希变化;基线后中心库更新;恢复前来源或目标变化 | 停止当前写入或恢复。更新已 Remove 但尚未 Add 成功时可能 RECOVERY_REQUIRED | 应保留阶段、原 Work 或人工备份路径;不能把部分成功视为整体成功 |
| **B09 权威状态与工程身份** | 保存/关闭失败;注册内容不符;中心库复读失败;校验后不是 GREEN/READY | 当前项不通过;是否影响后续取决于释放事实。上传成功可为 UPLOADED,但不等于 UPLOADED_VERIFIED | 不能仅以 PCB 窗口打开、COM 返回或文件目录存在判成功 |
| **B10 IO、路径与证据存储** | 网络盘不可见、路径映射失败、证据目录不可写、检查点变化、空间不足 | 读取/当前任务失败;关键写前检查点保存失败时不开始写入;某些辅助诊断写入失败只记录警告 | 网络/磁盘与原生数据库错误要分开报告;恢复需先复制验证库,因此有实际容量和耗时 |
| **B11 程序关闭等待** | 关闭时检查仍活动、当前任务仍忙、线程池或路径租约操作仍未结束 | 首次关闭可能只请求停止;进入关闭流程后等线程池/租约结束,代码没有统一关闭总超时 | R7 正常退出通过,不代表 COM 永久不返回场景也能退出;需要单独验收这一条件 |
| **F01 已修复复测** | 原生私有桌面 GetThreadDesktop 拒绝访问;重新 OpenDesktop 后 HWND 不可读 | 之前误报窗口归属变化,无法正常释放 | 已改为有界桌面枚举、映射窗口后再核对 PID,真实 PCB 正常取消退出通过 |
| **F02 已修复复测** | 状态刷新期间结果窗口的释放按钮仍可点击,函数因 busy 直接返回 | 之前按钮点击无窗口 | 已跟随 busy/排队目标更新,R7 先禁用后可用,点击能打开释放窗口;不等于 E01 已解决 |
另有 **P2 结果展示错误**:释放数量包含自然退出且未允许操作的观察项,因此出现“已释放 4 个,确认允许 2 个”。这是统计问题,不是解除占用的证明。
## 三、完整业务流程总图
下图按实际任务边界展开:所有菜单可以独立提交请求;排队串行的是 Mentor 原生操作。为保持图可读,把共享的 PCB 子流程和释放子流程放在后两张展开图中,节点 V、R 在各图中指向同一过程。
```mermaid
flowchart TD
START[启动桌面应用] --> INIT[配置日志、加载原生能力、异步发现未完成任务]
INIT --> LIB[选择中心库并建立路径映射 B10]
LIB --> READ[读取 CBB、工程文件及权威状态 B07 B09]
READ --> LIST[搜索、筛选、可见项勾选]
LIST --> OP{选择独立操作入口}
OP --> VERIFY[检验:只取可见非 READY 项]
OP --> CHECK[ICDB 批量检查:只取可见非 READY 项]
OP --> HISTORY[版本历史与备份:单个非 READY 项]
OP --> RECOVER[从完整工程备份恢复:非 READY 项]
OP --> MANAGE[中心库管理:READY 可更新或移除]
OP --> R[释放残留:进入释放子流程]
VERIFY --> CONFIRM[显示目标与确认;写操作提示人工备份]
CHECK --> CONFIRM
HISTORY --> ADMIT
RECOVER --> CONFIRM
MANAGE --> CONFIRM
CONFIRM --> ADMIT[原生队列准入 B01:可取消;按目标库保留状态判断]
ADMIT --> DISPATCH{任务类型}
DISPATCH -->|普通校验| V[共享 PCB 打开、弹窗、校验、关闭子流程 E02]
DISPATCH -->|检查| CITEM[每项重新判断 READY 和内容是否更新]
CITEM -->|应跳过| RESULT
CITEM -->|应检查| CV[调用 V;仅诊断打开,不保存校验、不 Repair]
CV --> CLASSIFY{本次错误正文含 ICDB 特征吗}
CLASSIFY -->|正常打开并关闭| CLEAN[CHECKED;不等于 READY]
CLASSIFY -->|含有特征| ICERR[报告 ICDB 问题;不执行自动修复]
CLASSIFY -->|其他弹窗或无法打开| UNCERTAIN[无法确认;保留原错误 E02]
CLEAN --> RESULT
ICERR --> RESULT
UNCERTAIN --> RESULT
DISPATCH -->|版本管理| HOPEN[绑定工程后打开原生备份窗口 B06]
HOPEN --> HUSER[用户查看、创建备份或选择版本回退]
HUSER --> HCLOSE[等待窗口正常关闭、释放结果及状态复读 B06 B09]
HCLOSE --> RESULT
DISPATCH -->|备份恢复| BVALID[来源身份、路径、文件、空间检查 B07 B10]
BVALID --> BCOPY[复制隔离验证库与备份工程 B08]
BCOPY --> BV[调用 V:在副本校验 E02]
BV --> BOK{PASS、GREEN 且安全释放}
BOK -->|否| BUNCHANGED[恢复失败;目标不安装]
BOK -->|是| BCOMPARE[复核来源和目标未变化 B08]
BCOMPARE --> INSTALL[旧 Work 改名保留,再安装副本 B08 B10]
INSTALL --> TV[调用 V:重新打开实际目标 B09]
TV --> RESULT
BUNCHANGED --> RESULT
DISPATCH -->|管理| MPRE[原生能力、来源、历史任务、库锁检查 B03 B07]
MPRE --> MCAT[原生目录复读并核对名称和分区 B09]
MCAT --> MBASE[记录内容基线;来源工程复制暂存 B08]
MBASE --> MOP{新增、更新或移除}
MOP -->|新增或更新| EXPORT[原生 ASCII、CSV、EDIF 身份导出 E03]
EXPORT --> MWRITE[新增 Add;更新 Remove 后 Add B05 B08]
MOP -->|移除| REMOVE[原生 Remove B05]
MWRITE --> MREAD[目录复读、内容身份及释放核对 B09]
REMOVE --> MREAD
MREAD --> AUTO{新增或更新启用自动校验吗}
AUTO -->|是| MV[调用 V 进行校验;上传成功与校验成功分别记录]
AUTO -->|否或移除| RESULT
MV --> RESULT
V --> RESULT[逐项记录结果、阶段、证据与释放事实]
RESULT --> SAFE{safe_to_continue}
SAFE -->|true| NEXT[继续下一项;ICDB 检查在每项之间让出原生队列]
SAFE -->|false| RETAIN[B02 保留本库会话状态;同库余项跳过]
RETAIN --> R
NEXT --> MORE{还有任务且未停止吗}
MORE -->|是| ADMIT
MORE -->|否| REFRESH[刷新受影响条目的权威状态 B09;生成报告 B10]
REFRESH --> LIST
R -->|证明成功| RESUME[清理本库保留状态;必要时用户继续未执行项]
R -->|未确认 E01| RETAIN
RESUME --> LIST
LIST -->|关闭应用| EXIT[停止当前任务并等待线程池、租约结束 B11]
EXIT --> END[程序退出]
classDef error fill:#fee2e2,stroke:#dc2626,color:#7f1d1d
classDef risk fill:#fff3cd,stroke:#d97706,color:#78350f
classDef good fill:#dcfce7,stroke:#16a34a,color:#14532d
class V,CV,UNCERTAIN,BV,EXPORT,R error
class LIB,READ,ADMIT,HOPEN,HCLOSE,BVALID,BCOPY,BCOMPARE,INSTALL,TV,MPRE,MCAT,MBASE,MWRITE,REMOVE,MREAD,RETAIN,REFRESH,EXIT risk
class CLEAN,RESUME good
```
图中的“回到队列”是下一次任务准入的概览:当前普通校验和管理批次在批次外围占用队列;ICDB 检查每完成一个 CBB 才释放并重新排队,不能理解为三种批次都已经逐项让出通道。读取/状态刷新也会使用原生队列,纯内存搜索与筛选不启动 Mentor。
## 四、共享 PCB 任务展开图 V
```mermaid
flowchart TD
V[开始:普通校验、检查或恢复副本校验] --> GATE[取得会话互斥、复核历史释放、取得库锁 B03 B04]
GATE --> PRE[检查 PRJ、PCB、依赖和证据目录 B07 B10]
PRE --> CONTENT{普通校验且配置了额外内容检查吗}
CONTENT -->|是| EXTRA[Importer 内容检查;失败结束本项]
CONTENT -->|否| LAUNCH
EXTRA -->|通过| LAUNCH[创建私有桌面与监督 worker;启动 PCB B05]
EXTRA -->|失败| FAILURE
LAUNCH --> OPEN{原生打开结果}
OPEN -->|读写确认提示| PERMISSION[E02 普通入口缺少通用授权交互;当前安全停止]
OPEN -->|可识别版本选择| VERSION[用户选最后手动保存、自动保存或取消;适用入口才提供]
OPEN -->|其他错误或未知弹窗| FAILURE[本项失败,记录实际文本或超时 B05]
OPEN -->|打开完成| MODE{仅诊断打开吗}
VERSION -->|可继续| OPEN
VERSION -->|取消| FAILURE
MODE -->|是| DIAG[证明目标已打开;请求关闭,不执行保存校验]
MODE -->|否| VERIFY[实际原生校验、保存、关闭 B05 B09]
VERIFY --> READY[中心库权威复读:必须 GREEN/READY B09]
PERMISSION --> CLOSE
FAILURE --> CLOSE
DIAG --> CLOSE[正常关闭、取消已知启动弹窗、释放 COM 和自动化助手]
READY --> CLOSE
CLOSE --> OWNERS[确认 worker、PCB、启动器和目标资源已释放 B04;桌面访问 F01 已修复]
OWNERS -->|成功| OUT[返回本项业务结果,safe_to_continue=true]
OWNERS -->|失败或无法确认| RETAIN[保留证据与会话,safe_to_continue=false,触发 B02]
classDef error fill:#fee2e2,stroke:#dc2626,color:#7f1d1d
classDef risk fill:#fff3cd,stroke:#d97706,color:#78350f
class PERMISSION error
class GATE,PRE,LAUNCH,FAILURE,VERIFY,READY,OWNERS,RETAIN risk
```
普通校验和检查都不默认启动自动 Repair。仓库中仍保留修复与修复后授权代码,但当前桌面批量检查路径明确关闭自动修复,不能把存在函数等同于产品入口已接通。
## 五、释放任务展开图 R
```mermaid
flowchart TD
R[主窗口或结果窗口点击释放] --> UI[校验当前目标与按钮可用状态 F02 已修复]
UI --> SCAN[异步枚举进程、线程桌面、窗口、历史会话]
SCAN --> GROUP{能证明属于本库吗}
GROUP -->|是| TARGET[可确认的目标客户端]
GROUP -->|否| OBSERVE[归属未知,只允许观察]
TARGET --> CONSENT[用户确认正常关闭;可查看原生提示]
CONSENT --> ACTION[重新核对 PID、创建时间、映像、窗口后正常关闭 F01 已修复]
ACTION --> PROMPT{还有原生提示吗}
PROMPT -->|保存提示| SAVE[用户保存、放弃或保留;真实分支尚未完整验收]
PROMPT -->|支持取消的活动窗口| CANCEL[确认后请求原生 Cancel;不当成取消已完成]
PROMPT -->|无| EXITCHECK
SAVE --> EXITCHECK[观察真实进程是否退出]
CANCEL --> EXITCHECK
EXITCHECK --> ALL{当前代码:所有观察项都 RELEASED 吗}
OBSERVE --> ALL
ALL -->|否| ERROR[E01 未归属本库的其他任务也会让整体 PARTIAL]
ALL -->|是| PROOF[持有证明句柄检查本库、共享库、地址、导出副本和在途调用 B04]
PROOF --> UNKNOWN[E01 底层仍可能把路径未知进程当候选占用]
UNKNOWN -->|不能证明| ERROR
UNKNOWN -->|证明成功| RECORD[归档匹配的历史记录,清理对应保留状态及映射]
RECORD --> RELEASED[RELEASE_CONFIRMED;允许重新尝试]
ERROR --> WAIT[B01 队列仍等或 B02 本库余项已跳过;显示残留原因]
WAIT -->|用户重新检查| SCAN
classDef error fill:#fee2e2,stroke:#dc2626,color:#7f1d1d
classDef risk fill:#fff3cd,stroke:#d97706,color:#78350f
classDef fixed fill:#dcfce7,stroke:#16a34a,color:#14532d
class ALL,ERROR,UNKNOWN error
class PROOF,WAIT,SAVE,CANCEL risk
class UI,ACTION,RELEASED fixed
```
强制结束是另一个需要确认的分支:只针对已确认归属、符合条件的客户端,不能结束数据库服务器或归属未知进程;本轮真实操作只验收了打开确认后取消,没有执行强制终止。它不会消除 E01,不能作为释放判断错误的替代方案。
## 六、代码位置与四层协调机制
| 机制 | 位置 | 控制范围 |
| --- | --- | --- |
| 原生队列 | `operation_queue.py:18`、`ui/main_window.py:844`、`ui/management_dialog.py:174` | 应用内一次执行一个原生任务;按库保留状态允许其他可运行票据越过等待项 |
| Windows 会话互斥 | `verify_runner.py:547`、`management.py:172`、`backup_recovery.py:145` | 同一 Windows 会话中的本工具原生调用;不是每工程独立互斥 |
| 库协调锁与历史阻止记录 | `library_lock.py:129`、`:443`、`:523` | 按中心库控制写入与释放;历史记录需复核,不能直接删除当成功 |
| 运行时保留状态 | `session_record.py:25`、`:236`、`:354`、`:443` | 原生队列准入、助手与客户端身份观测;日志阶段不等于真实进程状态 |
| 导出副本所有权 | `native_export_guard.py:10`、`native_engineering_identity.py:288` | 防止原生导出器仍持有分析副本时释放相关操作 |
关键传播代码:`verify_runner.py:1294`、`management.py:430`、`ui/icdb_check_task.py:55`。E01 的双重判定:`residual_sessions.py` 中 `_release_residual_sessions` 的全部行 RELEASED 条件,以及 `library_lock.py:443` 的 `_prove_library_released`。E02:`dialog_rules.py:222`、`verify_runner.py:895`。E03:`native_engineering_identity.py:288`。退出等待:`ui/main_window.py:1709`。
建议处理顺序:**E01 的本库释放判定 → B02 的工程/共享库影响范围 → E02 的明确用户确认路径 → B01/B05/B11 的等待及停止结果 → E03 输入兼容性与剩余条件按钮原生验收**。这些是后续工作建议,未在本次流程整理中实施。
@@ -0,0 +1,73 @@
# 误判复核与优化顺序
> 证据路径说明:本文的本机绝对路径、日志名、截图、测试 XML 和冻结包指向验收机器上的原始材料,未纳入 Git;仓库仅保留说明和验证摘要。
> 这是当天早期的复核与建议;后续实施结果见 [占用范围验证](2026-11-12-scoped-release-validation.md),不将早期未通过项追记为成功。
依据 `42d0ceb` 的六处修复记录、真实冻结程序验收,以及 13:44–13:46 的外部进程归属交叉核对。本轮只做分类和优化建议,没有修改业务代码。统计单位是已知问题根因,不能据此估算整个程序的缺陷率。
## 已修复的六处问题
| 问题 | 复核分类 | 依据 |
| --- | --- | --- |
| 导出器退出期间 OpenProcess 返回 5,立即认为释放失败 | 误判/过早得出失败结论 | 原生退出竞争;需要继续观察新快照与目标文件,而不是只看一次权限错误。已修复并有真实导入/更新复测 |
| 无工程参数的外部 viewdraw 阻止备份入口 | 跨工程占用误判 | 未证明目标被占用;恢复入口已改为结合路径、地址与目标文件实际探测 |
| 历史 sAddress_.adr 不一致导致报告工程内容不同 | 内容校验范围误判 | 地址元数据不等于工程内容;完整复制保留,但主要内容哈希排除 |
| 旧工程没有辅助 PCB/pcb.prj 被报告备份不完整 | 工程格式误判 | PCB 唯一关联主 PRJ 的旧格式本身可用;已复用同一身份检查并真实进入副本校验 |
| 私有桌面窗口不可读,不能正常取消 PCB | 真实 Windows 接入缺陷 | 正常进程查询权限不等于可读外部桌面 HWND;已修复并真实关闭客户端 |
| 释放按钮可点击但 busy 导致处理函数直接返回 | 真实 UI 状态缺陷 | R6 原包实际点击无窗口,R7 新包先禁用再可用,实际打开窗口 |
六处中四类属于误判/过严判断,两类属于真实功能缺陷。这里“误判”仍然是程序需要修复的缺陷,不表示可以撤销失败记录或补记验收成功。
## 当前剩余五类问题
| 类别 | 判定 | 是否仍需修改 |
| --- | --- | --- |
| 外部未知会话导致整库释放 PARTIAL | **已确认占用范围误判,1 个根因至少影响两道释放门禁** | 是,最高优先级;操作集合全退出判断与底层资源证明必须一起修正 |
| 显示释放 4 个,但确认允许操作只有 2 个 | **统计归因错误** | 是;区分本次请求关闭、实际退出、原已退出、外部只读观察,不混成主动释放数量 |
| PCB 读写确认挡住校验与备份恢复 | **原生提示真实发生,普通流程缺少相应确认路径;当前停止不等于检测错** | 是;明确用户授权、目标身份和实际占用核查;不能当 ICDB 损坏,也不能直接自动同意 |
| A/B 的原生 EDIF 导出崩溃 | **真实导出失败,输入/原生兼容性根因未定** | 是;保留错误码、输入及导出证据,独立复现,不绕过内容身份校验 |
| 条件按钮、现场失联副本、Win10 等未完成 | **验收缺口,不能记为已发生漏洞或误判** | 补真实覆盖;测试后再决定需要何种修改 |
此前“本程序遗留会话导致释放不了”的描述需细化:本程序在停止启动过程中确实留下过两个 PCB 客户端,属于真实清理问题;正常关闭后,继续挡住整库确认的是外部软件。不能用后半段的误判抹去前半段实际残留,也不能把两者当成同一种“锁”。
## 优化顺序与验收条件
### P1:统一占用归属和资源范围
当前备份入口与整库释放入口使用不同的判断强度;一处已经按目标资源探测,另一处仍能因未知进程直接停止。应复用统一的“来源、目标范围、活动阶段、资源证据”判断,避免每个入口再次发明一套锁规则。
- 来源记录:创建时保存 PID、创建时间、映像、任务号、私有桌面、目标工程/库和父子关系;重启后按身份核对,不能仅凭 PID。
- 范围判定:区分同工程、同库共享依赖、明确其他库、未知。由谁拉起与是否占用本库分别保存。
- 观察项不自动变成阻止项。明确属于其他库的进程不要求退出;未知项展示不确定原因,继续取本库在途调用、目标文件/数据库、地址、必要原生读取等证据。证据不足不能猜测安全。
- 不能只把 `scope is not False` 改成 `scope is True`,也不能只改“所有行退出”条件;否则可能放过真实打开同库的外部会话,或仍被底层第二道门禁挡住。
- 目标文件独占打开只是证据之一,不单独保证原生接口、共享依赖和检查期间的状态均安全。后续写入前还需重新核对。
验收至少包括:其他软件开其他库且持续运行时,本库能正常完成操作;其他软件确实持有本库数据库时,本库拒绝相关写入;外部进程来源不明但目标文件无占用时,不能单凭进程名永久阻止;PID 被复用时不得操作新进程。
### P1:把单项失败与批次停止分开
现有一个 `safe_to_continue` 同时承载多种情况:当前项失败、当前工程占用、共享库写入未结束、释放证据不足。因此某个工程的问题容易扩大为整库不可用。
逐项保留业务结果,再明确能否继续及影响范围:仅当前工程、整个共享库、原生执行通道、用户主动停止。安全释放的失败项继续下一项;仅当前工程问题是否可绕过,需要以共享资源证据决定;共享写入未返回则继续保护。不要全局放开 false 或直接删历史记录。
验收要使用同一批次“坏工程 → 好工程”以及“共享库真实占用 → 后续写入不得开始”两个方向。现有真实多来源新增已通过前一方向的一种安全失败场景,不代表所有占用场景都通过。
### P1/P2:接通原生读写提示与非阻塞操作
普通校验、检查、恢复副本按操作目的决定是否允许读写授权,不复用只适用于修复后的隐式授权。识别精确提示后展示目标和选项,用户可继续、跳过、取消;等待用户决定时保持 UI 可用,并明确后续任务是否仍需等当前原生会话。检查流程不能因用户点击而悄悄变成自动修复。
队列、原生执行和关闭分别展示等待对象与阶段。取消排队立即生效;已进入原生写入时说明是否正在等待返回或支持原生取消。普通校验/管理当前在批次外层持有队列,评估能否在完整单项事务结束后让出通道,不在 Remove→Add 中间插入另一写任务。
### P2:结果和性能
- UI 使用“检测结果、操作结果、释放状态”分别表达,避免将“待用户确认”写成工程损坏,或将“上传成功”写成 READY。
- 修正释放数字归因,报告观察项不增加本次主动释放数量。
- 来源、版本、依赖等便宜的输入检查先做;只复查本次受影响工程和实际依赖。避免每个按钮都全局扫描并读取无关窗口。
- 原生 EDIF 单独诊断;保存失败输入和退出码,让独立 CBB 继续,不能降低身份校验来消除报错。
### P2:真实并行与异常操作回归
保留其他软件测试作为并行验收条件,覆盖:外部独立工程持续打开、外部共享同库、检查中排队再取消、停止后立即释放、刷新时连续点击、版本窗口开关、断网、源文件变化、主程序关闭与重启后的 PID 复用。分别核查目标变更、真实进程退出、权威状态和剩余任务,而不是只看按钮恢复可用或 Python 测试通过。
当前最有价值的改动是修 E01 的作用范围与证据链,而不是继续增加更多全局拒绝条件。最终验收仍要求实际完整校验与恢复到 READY;本次分类没有补齐这一缺口。
@@ -0,0 +1,337 @@
{
"schema_version": 1,
"created": "2026-11-12T11:32:48",
"overall_acceptance": "NOT_PASSED",
"automatic_icdb_repair": false,
"merged": false,
"issue_closed": false,
"environment": {
"windows": [
"11",
"10.0.26200",
"SP0",
"Multiprocessor Free"
],
"python": "3.12.10",
"mentor": "EE7.9.5"
},
"baseline_commit": "b86e4e315a2ac7d8a3d05587389d0237b6576881",
"candidate": {
"executable": "F:\\CbbAccept-20261112\\candidate-release4\\dist\\CbbRefreshVerify-Background\\CbbRefreshVerify.exe",
"module_count": 62,
"source_and_frozen_code_equal": true,
"source_sha256_checked": true,
"packaged_assets_equal": true,
"exe_icon_images_equal": true,
"executable_sha256": "2ad132e434a8e1de0b61f271aa87215749b0f3174ea0682eb578bc31271e8a4d"
},
"tests": {
"name": "pytest",
"errors": "0",
"failures": "0",
"skipped": "0",
"tests": "1851",
"time": "382.059",
"timestamp": "2026-11-12T11:20:58.800467+08:00",
"hostname": "Abs"
},
"compileall_passed": true,
"git_diff_check_passed": true,
"original_integrity": {
"time": "2026-11-12T11:29:23",
"results": [
{
"scope": "A",
"source": "F:\\Sync Tool Skyworth\\Work Maps\\Skyworth_CBB_Lib\\CBB\\1GPHY_RTL8211F",
"checked": 187,
"changes": []
},
{
"scope": "B",
"source": "F:\\Sync Tool Skyworth\\Work Maps\\Skyworth_CBB_Lib\\CBB\\1GPHY_RTL8211F_A",
"checked": 185,
"changes": []
},
{
"scope": "field-library-originals",
"source": "F:\\Sync Tool Skyworth\\Work Maps\\Skyworth_CBB_Lib\\Mentor_lib_EE7.9",
"checked": 26887,
"changes": []
}
],
"originals_unchanged": true
},
"native_crud": {
"same_name_update": {
"operation": "update",
"name": "1GPHY_8212",
"stage": "complete",
"status": "UPLOADED",
"safe_to_continue": true,
"detail": "上传已确认;本批次未启用自动校验。"
},
"remove": {
"operation": "remove",
"name": "1GPHY_8212",
"stage": "complete",
"status": "REMOVED",
"safe_to_continue": true,
"detail": "已由原生操作移除,重新读取目录确认不存在;应用未创建备份。"
},
"batch_first_safe_failure": {
"operation": "add",
"name": "1GPHY_RTL8211F",
"stage": "complete",
"status": "UPLOAD_FAILED",
"safe_to_continue": true,
"detail": "Native identity exporter failed or reported an error."
},
"batch_next_added": {
"operation": "add",
"name": "1GPHY_8212",
"stage": "complete",
"status": "UPLOADED",
"safe_to_continue": true,
"detail": "上传已确认;本批次未启用自动校验。"
}
},
"final_ui_observations": [
{
"time": "2026-11-12T11:26:17",
"event": "legacy-backup-recovery-result",
"bundle_sha256": "2ad132e434a8e1de0b61f271aa87215749b0f3174ea0682eb578bc31271e8a4d",
"text": [
"目标 CBB",
"备份工程",
"恢复未通过:备份校验未通过,目标未更改。\nDATABASE_RECOVERY_REQUIRED: 是(&Y) | 否(&N) | If an XDS or ExpeditionPCB is running on this design, it's likely that\nyour file system and/or network system was too slow to determine\nwhich product the design is open in within the 15 second timeout\nperiod.\n\nOtherwise, the design may not have been closed properly due to\nthe XDS or ExpeditionPCB crashing in the last session.\n\nBefore continuing, please confirm that no other users have the\ndesign open.\n\nContinue to open the design with read/write access permission?\n记录:F:\\CbbAccept-20261112\\final-data\\evidence\\backup-restore-90b843115d07\\recovery-result.json"
],
"target_primary_unchanged": true
},
{
"time": "2026-11-12T11:27:59",
"event": "final-backup-queue-cancelled",
"bundle_sha256": "2ad132e434a8e1de0b61f271aa87215749b0f3174ea0682eb578bc31271e8a4d",
"text": [
"目标 CBB",
"备份工程",
"恢复未通过:已取消排队,未启动原生操作。"
]
},
{
"time": "2026-11-12T11:28:27",
"event": "summary-release-initial-state",
"bundle_sha256": "2ad132e434a8e1de0b61f271aa87215749b0f3174ea0682eb578bc31271e8a4d",
"enabled": false,
"text": [
"ICDB 检查结束:未发现异常 0,发现问题 0,占用 0,无法确认 0,取消 1,未执行 0",
"中心库",
"可见 1 / 1 · READY 0 · 异常/待确认 1",
"ICDB 检查结束:未发现异常 0,发现问题 0,占用 0,无法确认 0,取消 1,未执行 0",
"正在执行当前任务…"
]
},
{
"time": "2026-11-12T11:28:28",
"event": "summary-release-opened",
"bundle_sha256": "2ad132e434a8e1de0b61f271aa87215749b0f3174ea0682eb578bc31271e8a4d",
"text": [
"正在检查当前库关联的会话…",
""
]
},
{
"time": "2026-11-12T11:28:34",
"event": "force-warning-cancel",
"bundle_sha256": "2ad132e434a8e1de0b61f271aa87215749b0f3174ea0682eb578bc31271e8a4d",
"text": [
"仅强制结束以下客户端。客户端中其他已打开工程也会退出;未保存的更改会丢失,正在写入的工程可能需要恢复。\n数据库服务器和归属未知的进程不会被强制结束。\n\n1GPHY_8212 — ExpeditionPCB.exe (PID 21856)\n1GPHY_8212 — ExpeditionPCB.exe (PID 21248)\n\n仍要结束这些进程吗?"
]
},
{
"time": "2026-11-12T11:28:59",
"event": "r7-normal-exit",
"bundle_sha256": "2ad132e434a8e1de0b61f271aa87215749b0f3174ea0682eb578bc31271e8a4d",
"exit_code": 0
}
],
"backup_recovery": {
"installed": false,
"status": "FAIL",
"detail": "备份校验未通过,目标未更改。\nDATABASE_RECOVERY_REQUIRED: 是(&Y) | 否(&N) | If an XDS or ExpeditionPCB is running on this design, it's likely that\nyour file system and/or network system was too slow to determine\nwhich product the design is open in within the 15 second timeout\nperiod.\n\nOtherwise, the design may not have been closed properly due to\nthe XDS or ExpeditionPCB crashing in the last session.\n\nBefore continuing, please confirm that no other users have the\ndesign open.\n\nContinue to open the design with read/write access permission?",
"target_primary_unchanged": true
},
"residual": {
"library": "F:\\CbbAccept-20261112\\library\\native-add\\fixture\\mentor_lib.lmc",
"status": "PARTIAL",
"confirmed_scope": [
{
"name": "ExpeditionPCB.exe",
"pid": 21856,
"release_status": "RELEASED",
"consent_allowed": true
},
{
"name": "ExpeditionPCB.exe",
"pid": 21248,
"release_status": "RELEASED",
"consent_allowed": true
}
],
"final_target_clients": 0,
"observed_unconfirmed": [
{
"name": "viewdraw.exe",
"pid": 29832,
"scope_confirmed": false
},
{
"name": "cscript.exe",
"pid": 8652,
"scope_confirmed": false
}
],
"foreign_task_processes_terminated": false,
"lock_or_address_files_deleted": false
},
"remaining": [
{
"priority": "P1",
"problem": "Unconfirmed processes from another task still prevent whole-library release proof"
},
{
"priority": "P2",
"problem": "Release count includes unconsented observation rows that exited independently"
},
{
"problem": "Native PCB read/write permission prompt prevents successful verify/READY and backup installation"
},
{
"problem": "Native EDIF exporter crashes for isolated A/B inputs, exit 0xC0000005"
},
{
"problem": "Some conditional native buttons, field lost-session recovery, Windows 10 remain unaccepted"
}
],
"evidence": [
{
"path": "F:\\CbbAccept-20261112\\candidate-release4\\bundle-verification.json",
"bytes": 261,
"sha256": "8dbefef3d8cbd65f3531a1919b11a8bea1b494f01fe63a65ac277e05b572b7de"
},
{
"path": "F:\\CbbAccept-20261112\\candidate-release4\\source-sha256.json",
"bytes": 8168,
"sha256": "35d9904b515297ab033e3290ae69785fa827e7f3440d1a670e665016c18b28f8"
},
{
"path": "F:\\CbbAccept-20261112\\full-suite-delivery.xml",
"bytes": 265893,
"sha256": "cb3298e2aa72c5f4899e1b4dc21c88b35b58af6343de5f3d148e6a5acfcba7de"
},
{
"path": "F:\\CbbAccept-20261112\\desktop-map-red.xml",
"bytes": 11144,
"sha256": "e209aa9aadefec21013ffaa1364c42cf210b3bfb738fb7cc817bd9f39f1aaa5f"
},
{
"path": "F:\\CbbAccept-20261112\\desktop-map-green.xml",
"bytes": 4929,
"sha256": "1a5b5814bc21d3f739573700f4c93da3b241161124f80ed44d2ee0bcf5f01c39"
},
{
"path": "F:\\CbbAccept-20261112\\legacy-backup-red.xml",
"bytes": 7828,
"sha256": "6c05be80efcd1430de99a905b9a32aaa29079aedb7cef606a673896bebf200dc"
},
{
"path": "F:\\CbbAccept-20261112\\legacy-backup-green.xml",
"bytes": 13877,
"sha256": "2a0fc3ddb5b743f2cb2db00a81dc1bc025f76a75a5c6099f667e3461a555d49b"
},
{
"path": "F:\\CbbAccept-20261112\\summary-release-red.xml",
"bytes": 3116,
"sha256": "3dcec893e4d3dca47d8d48bd8e83752f8a3860da09195c40429d1841bf6b9a6b"
},
{
"path": "F:\\CbbAccept-20261112\\summary-release-green.xml",
"bytes": 10321,
"sha256": "f8beefdea1b616d6459f4e12becdbdac633b4650f487332627d5e54c85518ce6"
},
{
"path": "F:\\CbbAccept-20261112\\originals-rechecked.json",
"bytes": 632,
"sha256": "693155ed50d18af7d9ae1bda41a696d25687b486541fc0e9df9956e7dbfe7653"
},
{
"path": "F:\\CbbAccept-20261112\\final-native-inspection.json",
"bytes": 22288,
"sha256": "038a635a8bd3e5d5de5eb4ad73125b12bbd1416d53a8f27ebe014e2a6bb84536"
},
{
"path": "F:\\CbbAccept-20261112\\r7-accept.log",
"bytes": 25203,
"sha256": "ceca1412c2df483f4290e7131c587af5de16d30874f23a588f53d32db432f9e7"
},
{
"path": "F:\\CbbAccept-20261112\\final-data\\management\\management-741e463dc75d4f35acd992ad42200fbf\\checkpoint.json",
"bytes": 3547,
"sha256": "7fe68525e01d8e35bacb2bc84cdc4fa1469bb599f40bd6a77dcad7a6725bce24"
},
{
"path": "F:\\CbbAccept-20261112\\final-data\\management\\management-d83e404f22864885ab57943783897ebe\\checkpoint.json",
"bytes": 11411,
"sha256": "6928ef7a8d9664da7ee209afeb280c61fd081f5d207ef8686ebf09ad57d8140a"
},
{
"path": "F:\\CbbAccept-20261112\\final-data\\management\\management-a900d2b62a544623b499ce55d099dfd2\\checkpoint.json",
"bytes": 1140,
"sha256": "dc0725ff0db339f81be5af021925396cba50f303cfea5c0aa2370a8deff18486"
},
{
"path": "F:\\CbbAccept-20261112\\final-data\\management\\management-c8a18e3bb8bc4aa19f7dbc982b3602d2\\checkpoint.json",
"bytes": 11652,
"sha256": "d09ad8d7b88e7ba607ddb34cac523ef8074f5d3f43fe2f033c35125629bc251c"
},
{
"path": "F:\\CbbAccept-20261112\\final-data\\evidence\\backup-restore-90b843115d07\\recovery-result.json",
"bytes": 3901,
"sha256": "df44ea3f2d01ace1189c2c5f93af8c67abc6c196cf2ba8e99318d222a582c58a"
},
{
"path": "F:\\CbbAccept-20261112\\final-data\\release-checks\\sessions-346da47efa8d\\result.json",
"bytes": 8947,
"sha256": "743fc764021ac8fc22d6f799f4f9663d704d0fd6c41aafd14409666e85747900"
},
{
"path": "F:\\CbbAccept-20261112\\step-restore_execute.py.log",
"bytes": 4134,
"sha256": "a6da200bd6a3895d60ae85eac419b440696a5b9b5503856990782ed04f1f9795"
},
{
"path": "F:\\CbbAccept-20261112\\r5-batch-add-result.png",
"bytes": 75306,
"sha256": "a812edf98f2a96ce4c1f5cd218c9ec0b045e94538af11dbd65c35db799e03e2d"
},
{
"path": "F:\\CbbAccept-20261112\\r5-release-result-1.png",
"bytes": 44151,
"sha256": "a916a2080504201902c5d873fc3d7913e47804b7c5e1aa410a8ef8f778e759b5"
},
{
"path": "F:\\CbbAccept-20261112\\r7-summary-release-initial-state.png",
"bytes": 11891,
"sha256": "d13236159a38adf8030d84a66f59db4269cba471f305e835918cf1e4f86b7397"
},
{
"path": "F:\\CbbAccept-20261112\\r7-summary-release-opened.png",
"bytes": 87863,
"sha256": "a77e0169e7dce2f25f5f3343afa73af75418837987bb7561c24dca18485e3627"
},
{
"path": "F:\\CbbAccept-20261112\\r7-final-release-result.png",
"bytes": 88350,
"sha256": "1b504c4b0c42c18cdbc73bc58b5aa716cde43cb0330106b07fc7cb5e40c7a514"
}
]
}
@@ -0,0 +1,74 @@
# 2026-11-12 冻结程序与 Mentor 原生验收
> 证据路径说明:本文的本机绝对路径、日志名、截图、测试 XML 和冻结包指向验收机器上的原始材料,未纳入 Git;仓库仅保留说明和验证摘要。
**结论:未通过整体验收,不合并发布、不关闭工单。**
本轮实际运行冻结 EXE、Mentor EE7.9.5、原生 COM/命令和私有桌面窗口;点击使用真实 UIA/Win32 控件,没有替换业务服务、伪造原生结果或抢占前台鼠标。此前 48 按钮 Qt 模拟结果不计为本次原生验收。
最终候选包:`F:\CbbAccept-20261112\candidate-release4\dist\CbbRefreshVerify-Background\CbbRefreshVerify.exe`。
SHA-256:`2ad132e434a8e1de0b61f271aa87215749b0f3174ea0682eb578bc31271e8a4d`。
冻结档案中的 62 个模块逐一与源代码编译结果相符;打包资产及 EXE 图标与当前源图标相符。
原始记录、截图、失败尝试和各轮候选包保存在 `F:\CbbAccept-20261112`,便于核对本报告中的具体测试轮次。
## 隔离范围与原件核对
- 原生创建独立空库 `library\native-add\fixture\mentor_lib.lmc`,随后实际新增、更新和移除测试 CBB。
- 另复制现场库的共享依赖及四个工程到 `field-lib`:26,887 个文件,815,607,385 字节。该库实际读取到四个 READY 项。
- 独立来源 A/B 的库引用原指向不可用的 S 盘,仅对隔离副本改为测试库;原工程未改。
- 验收结束重新核对原件:A 的 187 个主要文件、B 的 185 个主要文件、复制现场库所对应的 26,887 个原文件,大小及 SHA-256 均未改变。见 `originals-rechecked.json`。
- 完整旧 Work 已人工保留到 `manual-old-work`;应用管理操作继续采用手动备份提示,没有新增自动整库备份。
- 原生进程只使用正常关闭/取消。强制结束的确认窗口实际打开后选择了取消;没有强制结束其他工具的进程,也没有删除地址或锁文件。
## 实际发现并修改的六处问题
| 问题 | 修改 | 实际复测及边界 |
| --- | --- | --- |
| 导出器退出期间 OpenProcess 返回 5,立即被当成释放失败 | 5/87 均继续观测;只有新快照无目标、已取得句柄退出且数据库探测通过才放行 | R3 同名更新、R5 新增及同批失败续跑;真实活动进程和数据库占用仍拒绝放行 |
| 另一个工程无 PRJ 参数的 viewdraw 阻止备份入口 | 备份空闲检查结合目标路径、地址及实际独占读探测,不再单凭未知命令阻止 | R7 中其他 DxDesigner 仍运行,备份进入隔离库及原生校验;这不等于下面的整库释放问题已解决 |
| 原生导入不保留历史 `sAddress_.adr`,被误报内容不一致 | 将其与 `sAddress.adr` 一起排除在主要内容哈希之外,规则版本升为 2,完整复制仍保留 | R3 更新 `UPLOADED`,原理图 ASCII/CSV/EDIF、布局与注册内容一致;R5 再新增成功 |
| 外部私有桌面的 GetThreadDesktop 返回拒绝访问,重新 OpenDesktop 后 HWND 还不可读 | 对失败 PID 枚举当前窗口站的桌面;打开桌面后先枚举窗口,再核对 PID 并发送关闭/取消 | R5 原先残留的 PCB 启动器/客户端两个真实进程正常退出,获得 RELEASE_CONFIRMED;R6/R7 同类客户端也退出 |
| 恢复入口强制要求旧格式工程没有的辅助 `PCB/pcb.prj` | 复用既有工程身份检查;PCB 必须唯一关联所选主 PRJ,数据库与布局仍必需 | R5 原包拒绝;R7 新包进入隔离复制和真实 PCB 校验。原生读写确认仍导致校验失败,目标主要文件未改,未安装备份 |
| ICDB 结果窗口在前台任务忙时仍允许点击释放,处理函数却直接返回 | 结果按钮跟随执行状态;目标库因保留会话而排队时使用排队释放入口 | R6 实际点击无窗口;R7 同样操作时按钮先禁用,随后恢复,点击成功打开释放窗口。排队入口另有回归覆盖 |
## 实际按钮与流程结果
| 入口/动作 | 观察结果 | 验收判断 |
| --- | --- | --- |
| 选择库、读取、三个菜单、搜索/筛选、选择可见项/清空选择 | 空库、一个 PENDING、四个 READY 的真实数据可读取和操作;READY 不进入维护入口 | 已覆盖这些真实数据场景;381/3,000 项响应测试仍仅是此前模拟 |
| 高级设置及原生 Verify 开关 | 展开、收起、切换和还原成功 | 仅验证控件;Win10 原生 Verify 运行未验收 |
| 查看运行日志、打开日志目录 | 实际点击可用 | 已覆盖 |
| 已选校验、可见非 READY 校验 | 前者实际运行;后者确认后取消,没有执行写入 | 已选校验未通过,不能报告 READY |
| 检查中提交普通校验 | UI 可用、确认成功、显示排队;检查结束后实际启动普通 PCB 校验 | 排队行为通过;校验本身失败 |
| ICDB 批量检查(取消确认/完整执行/停止/再次使用) | 完整检查约 38 秒;读写确认不含 ICDB 错误,归类为无法确认;停止后可重新使用,可能留下客户端供确认释放 | 未执行自动 Repair;不能视为现场修复通过 |
| 版本历史与备份 | 从应用打开真实原生窗口,读取五个备份;创建 User 备份,再从原生 Restore 对话框恢复所选备份 | 原生输出 `Backup restored successfully.`;关闭后 CLOSED、safe=true,应用按钮恢复。证据为 R3 原生工具操作 |
| 从备份恢复(无效来源、取消确认、重复点击) | 无效输入拒绝;取消不启动;执行后按钮禁用,重复点击未重复启动 | 已覆盖 |
| 恢复排队、取消排队 | R6/R7 在检查期间提交,显示排队;取消后明确未启动原生操作 | 已覆盖 |
| 旧格式备份完整恢复 | R7 通过格式/复制检查,在隔离库打开 PCB 时被读写确认拦下 | **未完成恢复**;installed=false、目标主要文件不变,保留证据 |
| 用工程替换已选 CBB | R3 实际 Remove → 原生缺项复读 → Add → 内容比对 → 目录复读 | UPLOADED、safe=true;关闭自动校验,不等于 READY |
| 移除 | R5 原生 CleanWGLib 返回成功;独立目录读取为空,应用读取 0 项 | REMOVED、safe=true |
| 多来源新增 | 同批 A 的 EDIF 导出失败、safe=true;随后 C 实际新增成功;独立复读只有 C | 未因首项安全失败而停止整批 |
| 重复名称新增 | 实际拒绝;独立复读的条目、工程身份不变 | 已覆盖 |
| 管理执行确认取消、再次确认、连续点击执行 | 取消后可重新确认;运行时禁用执行,重复点击不创建第二项 | 新增、更新、移除均有真实操作记录 |
| 查看原生提示、重新检查、确认释放、关闭 | 窗口和提示可读;本工具两个 PCB 进程正常退出 | R5 完整释放通过;R6/R7 仍被下面的跨任务观察项阻止整库确认 |
| 强制结束客户端 | 确认名单只列两个本库 PCB,点击取消后继续正常释放 | 仅确认/取消路径通过,实际强制终止未执行 |
| 主程序退出 | R3/R4/R5/R6/R7 正常退出,末次退出码 0 | 不将退出应用造成的内存释放算作整库释放功能通过 |
**仍未实际触发的条件按钮**:处理 PCB 未保存更改的保存/放弃选择、活动原生备份的“请求原生取消”、管理批次“当前项结束后停止”、实际强制终止、ICDB 特征错误后的工程副本重导入、成功安装备份后的再次打开并 READY。它们的 Qt/回归测试不能替代原生验收;不宣称每一个按钮的所有原生分支都已通过。
## 未解决的验收卡点
1. **P1:未确认归属的其他任务仍能阻止整库释放。** R7 `sessions-346da47efa8d/result.json` 中,确认允许的 PCB PID 21856/21248 均 RELEASED,但另一个私有桌面上的 viewdraw PID 29832 为 scope_confirmed=false、consent_allowed=false、WAITING,整体仍 PARTIAL。末次检查还观察到其他工具的 cscript;没有本库确认所属的客户端残留。原因在于 `residual_sessions._release_residual_sessions` 要求全部观察项 RELEASED,随后 `library_lock._prove_library_released` 也把无法确定路径的进程当成候选占用。不能要求关闭其他任务来通过本库验收。后续需要将观察项与可操作目标分开,并用本库数据库/地址/独占句柄及在途调用证明决定能否继续;同时覆盖真实持有本库数据库时仍必须拒绝的场景。本轮未修改此判定。
2. **P2:释放数量混入自然退出的观察项。** R7 显示“已释放 4 个会话,确认允许 2 项”;另两个未允许项在操作期间自行退出,不能让用户误以为工具主动关闭了四项。
3. **原生 PCB 读写确认尚未走通。** `DATABASE_RECOVERY_REQUIRED` / `DATABASE_SAFETY_STOP` 的正文要求确认没有其他用户后以读写权限继续,未直接提供 ICDB 损坏证据。检查、普通校验和备份副本校验均未获得 READY;不能自动同意写入后宣称通过。
4. **输入兼容性:A/B 的原生 EDIF 导出崩溃。** 退出码 3221225477(0xC0000005),本机 WER 指向 MSVCR90.dll;相同环境下 C 的三种导出均成功。A/B 继续报告上传失败,不跳过身份校验伪造成功。
5. 现场服务器会话失联的一致故障副本完整入口、Windows 10、自动修复仍未验收。ICDB 自动 Repair 保持关闭。
## 复核依据
- 最终源代码:1,851 项 Python 测试通过,1 条 pywinauto COM 警告;`full-suite-delivery.xml`。编译检查及 `git diff --check` 通过。
- 回归保留修复前失败与修复后通过记录:`export-race-*`、`pathless-red.xml`、`address-manifest-*`、`desktop-map-*`、`legacy-backup-*`、`summary-release-*`。
- R3 同名更新:`final-data/management/management-741e463dc75d4f35acd992ad42200fbf/checkpoint.json`。
- R5 移除:`management-d83e404f22864885ab57943783897ebe`;同批失败:`management-a900d2b62a544623b499ce55d099dfd2`;同批继续新增:`management-c8a18e3bb8bc4aa19f7dbc982b3602d2`。
- R7 恢复:`final-data/evidence/backup-restore-90b843115d07/recovery-result.json`。R7 释放按钮状态、原生提示、取消强制结束、最终释放及退出见 `r7-accept.log` 和对应截图。
- 机器可读索引和指定证据摘要哈希见同目录 `2026-11-12-native-acceptance-validation.json`。大体积原始工程和运行时控制记录保留在本机,不纳入提交。
@@ -0,0 +1,41 @@
# 阻塞进程来源核查
> 证据路径说明:本文的本机绝对路径、日志名、截图、测试 XML 和冻结包指向验收机器上的原始材料,未纳入 Git;仓库仅保留说明和验证摘要。
> 本文保留当时的只读归属结论;后续代码修改及试验见 [占用范围验证](2026-11-12-scoped-release-validation.md)。
2026-11-12 13:44–13:46 进行了只读核查。结论:此前阻止整库释放的两个未知会话来自 `dxdesigner-part-replacement` 测试,不是本 CBB 程序拉起;其他软件并行测试暴露了本程序的归属和释放判定漏洞。本次没有修改业务代码或关闭进程。
| 历史进程 | 归属证据 | 结论 |
| --- | --- | --- |
| viewdraw.exe,PID 29832,父进程 5940 | 外部工具 lifecycle.jsonl 记录 worker_started PID 5940;其 private-desktop.json 记录 cad_pid 29832、创建时间 2026-11-12 03:17:21.099 UTC、桌面 dx-f2b2947f8ab144e19760e4599fda3249,均与 CBB 残留快照一致;窗口工程位于外部工具 outputs 目录 | 外部工具创建的 DxDesigner |
| cscript.exe,PID 8652,父进程 5940 | 命令行运行外部工具 QuickPartReplace.vbs;对应任务 project-checkpoint.json 的工程和 cad_pid 29832 与窗口匹配 | 同一外部测试批次的脚本 |
| ExpeditionPCB.exe,PID 21856 / 21248 | 命令行明确为 CBB 隔离库 1GPHY_8212 的 PCB;父子链 30552 → 21856 → 21248;CBB 验收释放记录确认两项均 RELEASED | 本程序的客户端,已正常退出 |
外部工程 CentralLibrary 指向 `F:/Sync Tool Skyworth/Work Maps/Skyworth_CBB_Lib/Mentor_lib_EE7.9/mentor_lib.lmc`。CBB 验收目标为 `F:/CbbAccept-20261112/library/native-add/fixture/mentor_lib.lmc`,二者不同。
13:44 的实时快照中,历史 PID 29832、8652 均已不存在,没有把新进程当成旧残留。当前另有真实父子链:DxPartReplace.exe 29436 → DxPartReplace.exe 16540 → cscript.exe 21364;父进程创建时间均早于子进程,映像和脚本均位于外部工具目录。此快照仅代表当时状态,PID 不能作为长期身份。
13:46 对本次隔离库的 LMC、icdb.dat、LayoutDB.lyt 共 12 个实际文件同时执行 GENERIC_READ、共享模式 0、OPEN_EXISTING 的只读独占打开,全部成功并关闭句柄;库内未发现 sAddress.adr。这证明当时这些目标文件没有阻止独占读的占用,不能倒推 11:28 的全部句柄状态。
## 逻辑漏洞所在
1. `library_lock._owner_matches_scope` 主要读取命令行中的 PRJ/PCB/LMC/数据库引用。裸 viewdraw.exe 和外部 QuickPartReplace.vbs 命令没有这些路径,均返回 None(UNKNOWN)。它没有在这里利用父进程批次记录、私有桌面或窗口工程身份进一步归属。
2. `residual_sessions.inspect_residual_sessions` 的 `scope is not False` 将这两项 UNKNOWN 加入当前库残留观察列表;`scope_confirmed=false` 正确阻止关闭它们。
3. `_release_residual_sessions` 又要求列表中全部行均为 RELEASED。这使“不能确认归属、不能操作的观察项”成为“当前库必须退出的对象”,结果一直 PARTIAL。
4. 即使绕过上一层,`library_lock._prove_library_released` 仍将范围为 UNKNOWN 的候选进程加入 owners,可能在目标文件独占打开之前直接拒绝证明。
以历史真实命令行回放当前归属函数,两项均得到 UNKNOWN,和真实残留列表一致。不是用户并行测试不合规,而是程序把“未知来源”升级成了“目标库占用”。
修复需要分别保存**进程来源**和**资源范围**:本程序创建的进程未必仍占目标;其他软件创建的进程也可能实际打开同一个库。因此不能只按进程名、父 PID 或软件目录粗暴排除外部进程,最终应结合创建时间绑定的身份、实际工程/库引用、本库在途调用、目标资源与地址探测决定是否阻止。外部归属只读展示,不自动结束;真实共享目标占用仍需保护。
## 证据
- 本机完整核查:`F:\CbbAccept-20261112\process-origin-audit.json`。
- 身份交叉核对、外部证据 SHA-256、分类回放及只读资源探测:`F:\CbbAccept-20261112\process-origin-confirmed.json`。
- 外部工具自身记录根目录:`F:\Sync Tool Skyworth\Work Maps\dxdesigner-part-replacement\outputs\frozen-random-final-20261112-r4`。
- 父 worker 记录:`data/batch-resources/28182d59cb9d4e0087fa312baf11d804/lifecycle.jsonl` 第 2 行。
- 原生进程创建身份:该批次 `generation-1/private-desktop.json`。
- 对应工程:`data/tasks/4f49b13e-3603-4578-b7f2-c9fe88bca864/quick-run/project-checkpoint.json`。
这里只完成核查和归因,没有声称跨软件阻塞已修复。
@@ -0,0 +1,19 @@
# PR #13 评论 391:合并前补充修复
> 证据路径说明:本文的本机绝对路径、日志、截图、测试 XML 和冻结包位于验收机器,未纳入 Git;仓库仅保留说明和验证摘要。
针对 [评论 391](https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/13#issuecomment-391),以 `466f400` 为修改前基线。用户已明确授权补齐后依次合并 #10、#11、#12、#13;代码合并不等同于现场验收或发布。
- 导出进程证据缺失、损坏或不是 JSON 对象时,日志及占用错误保留 `exclusion_unavailable`、具体文件或目录、PID 和原因。仍拒绝把无法核实的活跃句柄认定为已释放。
- 批次结果在原生操作及报告完成后立即显示,权威状态刷新继续通过原生队列串行执行。刷新前不以 PASS 将清单改成 READY;等待/执行刷新有明确状态。刷新期间关闭结果窗口不重建窗口;取消排队刷新保留已完成结果,并将未复核清单状态记为 UNKNOWN。
- 11 月 12 日验收文档统一标明本机证据未入库,README/SOP 保留 EDIF 原生崩溃、SMB/跨机、Windows 10、现场失联副本及自动 Repair 的限制。
验证记录位于 `F:\CbbAccept-20261112`。新增用例先复现失败,再验证修复;私有桌面 `CbbReview391NativeUi` 下实际 Windows Qt 控件和进程句柄的 6 项回归通过。覆盖正常刷新、提前关闭结果、取消排队刷新,以及缺失/坏 JSON/错误结构的诊断与保守保护。独立合并前代码复核未发现新增阻断项。
最终完整回归 **1,886 passed**,1 条既有 pywinauto STA 警告,记录 `review391-final.xml`。编译及 diff-check 通过。未完成的早期全量运行已停止,不计作通过;坏 JSON、缺失证据、错误结构及结果延迟的失败复现记录分别保留为 `review391-red.xml`、`review391-ui-red.xml`、`review391-shape-red.xml`。
新冻结包为 `candidate-review391/dist/CbbRefreshVerify-Background/CbbRefreshVerify.exe`,SHA-256 为 `eb9200c06cf8b054dbbfe678fad14c5f5a16f634526e38d2487f5960e07760ab`。64 个冻结模块逐一与当前源码编译结果一致,资源及 PE 图标一致。私有桌面启动、三个实际菜单切换及正常退出通过,详见 `review391-frozen-smoke.json`;机器摘要见 [验证 JSON](2026-11-12-review391-validation.json)。
本轮没有在生产库执行操作,也没有重新验收现场 ICDB 故障;此前 Mentor 隔离库完整恢复至 READY 的证据见 [占用范围验证](2026-11-12-scoped-release-validation.md),该历史包与本轮新包分别记录,不混用源码与二进制证据。
依赖链合并时,#10 先进入 main,再将 #11、#12、#13 逐项改为以 main 为目标并合并,保留分支。逐项核对 head SHA,最后比较 main 与被测最终 head 的文件树;不把只合进中间分支当成全部进入 main。
@@ -0,0 +1,63 @@
{
"created": "2026-11-12T17:35:46.043003+08:00",
"base_commit": "466f400514acc8531a219db8e75641be35304a3f",
"review_comment": "https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/13#issuecomment-391",
"full_suite": {
"path": "F:\\CbbAccept-20261112\\review391-final.xml",
"tests": 1886,
"failures": 0,
"errors": 0,
"skipped": 0
},
"windows_private_desktop_suite": {
"path": "F:\\CbbAccept-20261112\\review391-native-ui.xml",
"tests": 6,
"failures": 0,
"errors": 0,
"skipped": 0
},
"bundle": {
"module_count": 64,
"source_and_frozen_code_equal": true,
"source_sha256_checked": true,
"packaged_assets_equal": true,
"exe_icon_images_equal": true,
"executable_sha256": "eb9200c06cf8b054dbbfe678fad14c5f5a16f634526e38d2487f5960e07760ab"
},
"frozen_smoke": {
"executable": "F:\\CbbAccept-20261112\\candidate-review391\\dist\\CbbRefreshVerify-Background\\CbbRefreshVerify.exe",
"sha256": "eb9200c06cf8b054dbbfe678fad14c5f5a16f634526e38d2487f5960e07760ab",
"tabs": [
"检验",
"ICDB 维护",
"中心库管理"
],
"exit_code": 0,
"private_desktop": "CbbReview391Frozen"
},
"source_sha256": {
"src/cbb_refresh_verify/native_export_guard.py": "c3eeadd557d2f89bfce855c0736c953043ebe97bc81aa74e40bd370780a54741",
"src/cbb_refresh_verify/ui/main_window.py": "af65fcd8e41f35f412e8c23f525821c92b2cfb78b9d0fd4f88c515249dbd24e1",
"tests/test_review391.py": "304d46d63af87ea4a9bca83834b77cc44c5c9addf7b0f3752431304ee612568c",
"tests/test_batch_item_yield.py": "ed4be6d1e854b18d054a75c92171cd46b6b64411c2f74e9a3caf31fba3c7366b"
},
"independent_code_review": "No actionable critical, important or minor issues; read-only review, no additional tests run.",
"merge_plan": {
"pull_requests": [
10,
11,
12,
13
],
"target": "main",
"retarget_stacked_pull_requests": true,
"preserve_branches": true,
"force_merge": false
},
"evidence_location": "Raw logs/XML/packages are local-only; repository contains this summary.",
"native_mentor_retested": false,
"known_limits": [
"Field ICDB loss-of-session recovery, Windows 10 and cross-machine SMB remain unverified.",
"Automatic ICDB Repair remains disabled; EDIF 0xC0000005 on A/B remains unresolved."
]
}
@@ -0,0 +1,61 @@
{
"schema_version": 1,
"time": "2026-11-12T16:00:50.172809+08:00",
"bundle": {
"module_count": 64,
"source_and_frozen_code_equal": true,
"source_sha256_checked": true,
"packaged_assets_equal": true,
"exe_icon_images_equal": true,
"executable_sha256": "85ec649c18fefcf50d8b4caf0a66005f108713e7d5f170d1aecc82e3c5fcfe33"
},
"tests": {
"passed": 1880,
"failures": 0,
"errors": 0,
"xml": "F:\\CbbAccept-20261112\\full-scoped-final10.xml"
},
"recovery": {
"path": "F:\\CbbAccept-20261112\\scoped10-data\\evidence\\backup-restore-5f9b134ab83c\\recovery-result.json",
"installed": true,
"source_status": "PASS",
"target_status": "PASS",
"library_flag": "GREEN",
"safe_to_continue": true,
"retained_work": "F:\\CbbAccept-20261112\\field-lib-recovery\\ReusableBlocksLP\\1GPHY_8212\\Work.before-restore-5f9b134ab83c"
},
"release": {
"path": "F:\\CbbAccept-20261112\\scoped10-data\\release-checks\\sessions-fc23a2a2024f\\result.json",
"status": "RELEASE_CONFIRMED"
},
"original_integrity": {
"time": "2026-11-12T16:00:19",
"results": [
{
"scope": "A",
"source": "F:\\Sync Tool Skyworth\\Work Maps\\Skyworth_CBB_Lib\\CBB\\1GPHY_RTL8211F",
"checked": 187,
"changes": []
},
{
"scope": "B",
"source": "F:\\Sync Tool Skyworth\\Work Maps\\Skyworth_CBB_Lib\\CBB\\1GPHY_RTL8211F_A",
"checked": 185,
"changes": []
},
{
"scope": "field-library-originals",
"source": "F:\\Sync Tool Skyworth\\Work Maps\\Skyworth_CBB_Lib\\Mentor_lib_EE7.9",
"checked": 26887,
"changes": []
}
],
"originals_unchanged": true
},
"interaction_log": "F:\\CbbAccept-20261112\\scoped10-accept.log",
"known_limits": [
"Native shared-symbol server sometimes cannot remove its own address file after STOP; not bypassed.",
"Actual field server-session-loss recovery and Windows 10 acceptance remain unverified.",
"Automatic ICDB Repair remains disabled. No merge, release or issue closure."
]
}
@@ -0,0 +1,54 @@
# 占用范围、单项队列及原生确认验证
> 证据路径说明:本文的本机绝对路径、日志名、截图、测试 XML 和冻结包指向验收机器上的原始材料,未纳入 Git;仓库仅保留说明和验证摘要。
本次基于 `42d0ceb`,沿用 `codex/runtime-lifecycle-safety`。当天较早的进程归属、误判分析和完整流程图是修改前快照;本记录描述后续实现及验收,不追溯改写先前失败。
## 修改
- 未确认归属的外部客户端只作观察,不要求其退出、不发送关闭,也不增加主动释放数量。释放仍检查本库在途调用、目标数据库地址和实际独占只读文件句柄;自身请求身份缺失、共享写入未返回时继续保护。
- 旧 NATIVE_CALL/CLOSING/UNCONFIRMED 日志结合进程创建身份核实;已退出或 PID 被复用不会仅因日志阶段被认定仍在写入,读取失败保留未确认。
- 单项失败后只对绑定任务取得新的释放证明;证明成功才允许后续项,原业务失败和原证据保留。校验和管理在完整单项事务结束后让出原生队列,不在移除与重新导入之间穿插其他任务。排队取消保留已完成项和汇总。
- 普通校验、上传后的校验、备份副本/目标校验接通精确 PCB 读写提示和历史版本选择。用户可以继续、跳过、取消;绑定本任务 PCB 的创建身份、启动链、工程和私有桌面,临近点击重新检查。ICDB 诊断检查不借此执行 Repair 或隐式取得写入授权。
- 实测新增修复:同名更新自动校验不再把自己当前的事务当成历史未释放事务。仅允许同线程、同 lease、同 PID、同目标及无待处理导出/检查点的嵌套调用。取消批次后不再把刷新排进已经取消的队列而丢失结果。
- 第二轮真实试验暴露导出器进程链误判:CSV 导出器 PID 16068 创建于 14:56:19,却收录了 12:57:45 已创建的 OneDrive 服务 PID 23804。运行时追踪改为排除原有进程、绑定原生进程句柄创建时间、拒绝早于父进程的子进程及 PID 复用后的新进程链。旧记录复核也排除出生时间证明不可能属于导出任务的条目;仍需原生范围及分析副本数据库的全套释放证明,未发送退出请求给 OneDrive。
- 失败释放复核包含在单项原生队列内部,完成后才让出通道,避免下一项开始时前一项又来争取管理互斥。新增两项回归先失败后通过,相关 46 项通过。
- 普通读写确认比较真实控件的句柄、ID、类和正文,允许 UIA 后续增加已识别的标题栏辅助节点、重排控件及省略按钮快捷键标记;替换真实按钮仍拒绝。实际快照显示 Win32 的“是(&Y)”变成 UIA 的“是(Y)”,进程及窗口/按钮句柄均未变化;不能据此认定工程或窗口被替换。来自恢复副本真实失败的回归先失败后通过,相关 52 项通过。
## 冻结包和真实试验
当前候选程序:`F:\CbbAccept-20261112\candidate-scoped10\dist\CbbRefreshVerify-Background\CbbRefreshVerify.exe`。
SHA-256:`85ec649c18fefcf50d8b4caf0a66005f108713e7d5f170d1aecc82e3c5fcfe33`。64 个冻结模块逐个与当前源码编译结果一致,资源及 PE 图标一致;检查记录 `candidate-scoped10/bundle-verification.json`。较早的 scoped5 为 `4cf682130923c4534d3193a8b967ffc83d0cc519e580f6739a4861daed4f36c3`,scoped7 为 `e37b2601bb3851e9170ab8d552800547004b1a5b9882211f1180693dde76957c`,scoped9 为 `a59488dcbad72bdd7159772aabdb33b6411dea972babec33e64412a3406f810f`;各次动作日志绑定实际运行包哈希,不能混为同一版本。
原生写操作仅在 `F:\CbbAccept-20261112` 隔离库及工具生成的 `C:\Users\ABS_Windows\.cbb-restore` 校验副本中进行,环境为 Windows 11 build 26200、Mentor EE7.9.5 及私有桌面。操作驱动使用真实程序按钮和后台窗口消息,没有前台鼠标抢占、强杀原生客户端或手删地址文件。
| 试验 | 事实与证据 |
| --- | --- |
| 外部脚本持续存在,目标真实句柄被占用 | `scoped3-accept.log`:持有 LMC 的 Windows 独占句柄时拒绝释放,错误 32;外部脚本仍存活 |
| 目标句柄关闭,外部脚本继续运行 | 同一日志:RELEASE_CONFIRMED,主动释放 0、授权 0;外部行为保持 OBSERVED,脚本最后自行退出 |
| 同名更新并自动校验 | `management-38fdf8365d834a13ba7dca67111861ff/checkpoint.json`:UPLOADED_VERIFIED、safe=true;真实选择继续打开和最后手动保存版本;校验 InSynch、Save/IsSaved/Close、PCB EXITED,IsVerifiedReuseBlock=1/READY。校验约 91.6 秒 |
| 重复更新发现地址残留 | `management-0cee6b12e93b45918caf18101c603dc5/checkpoint.json`:RECOVERY_REQUIRED;原生移除已发生,重新导入未开始,不能记为成功 |
| 地址残留的原生证据 | 共享符号库服务器 PID 17344 日志记录 Database closed、defragmentation completed、Could not remove server address file、S: STOP;之后没有该服务器进程。`address-native-probe/result.json`:原生目录可复读,地址仍在、端点不可达;未将不可达视作无占用证明 |
| 早期空白库试验 | 读写和手动版本选择成功,但原生 PartPkg 报缺 PartsDB 分区;不作为校验成功 |
| 早期自动校验试验 | `management-cb85cb2b564d4778912df7accf85c109` 上传完成,校验被自身事务锁拒绝;已补回归后由上述真实完整成功试验复核 |
| 导出器错误收录 OneDrive | `scoped6-data/management/management-498258fe1e1541e9a135ca4117cb7344/source-identity/csv-process.json`:CSV 日志已完成、分析副本无地址,但历史父 PID 关联扩大至外部进程;后来进程查询超时,本次无原生移除/新增,不计上传成功。`scoped6-active-handle-attribution.json` 为只读 Windows 句柄归属诊断;新增失败回归 `export-lineage-red.xml`、`export-history-red.xml`,修复后两组 73 项通过 |
| 旧误判记录解除及更新重试 | scoped7 真实点击释放:RELEASE_CONFIRMED、主动关闭 0;保留旧证据。随后 `scoped7-data/management/management-d4017aaae47041b08dae6445858a7419/checkpoint.json` 完成原生更新及工程身份复读,UPLOADED、safe=true,关闭了本批次自动校验 |
| 恢复确认窗口误判 | scoped7 的 `evidence/backup-restore-f52ee9adc188/recovery-result.json` 和 scoped9 的 `evidence/backup-restore-22512e81160b/recovery-result.json` 都在副本阶段失败,installed=false。先处理控件辅助节点后,第二次实际差分又明确了快捷键标记差异;未将两次失败记为通过 |
| 快捷键表示差异的原生定位 | `open-identity-real-diff.json` 捕获同一 PID/创建时间、HWND、真实按钮句柄下的 Win32/UIA 差异。修正后 `identity-debug/verify-f34d9f72d4aa/result.json` 实际 PASS、GREEN/READY=1、Save COMMITTED、Close CLOSED、PCB EXITED、资源 RELEASED。这是定位与复测,不能替代完整恢复按钮验收 |
| 立即停止后的残留 | scoped7 的 `evidence/icdb-check-73cdeda6d163/pcb-open/verify-bb3bdc393b0b/result.json`:CANCELLED、safe=false;本任务启动器 12408、PCB 344 在扫描时尚无可关闭窗口。未以队列取消成功冒充原生退出成功 |
| 残留后的再次释放 | 后续读取发现原生 Tip of the Day;在精确绑定的私有桌面正常关闭此帮助窗口后,PCB 完成打开。scoped9 实际“确认释放”正常关闭 3 项并 RELEASE_CONFIRMED,应用正常退出。`cleanup-scoped7-stop*/startup-close.json` 保留早期未确认记录;未强杀进程、未把帮助窗口关闭本身记成数据库释放 |
| 最终包验收 | 事实与证据 |
| --- | --- |
| 最终包完整恢复按钮 | scoped10 真实执行空来源校验、取消确认、重新确认、运行期间重复点击,随后完成副本校验、目标变更检查、保留旧 Work、安装、目标再次原生校验。`scoped10-data/evidence/backup-restore-5f9b134ab83c/recovery-result.json`:installed=true,source_validation 和 result 都为 PASS、GREEN/READY=1、safe=true,Save COMMITTED、Close CLOSED、PCB/助手 EXITED、资源 RELEASED |
| 最终状态及退出 | scoped10 原生状态刷新后可见 READY 1;READY 项检查及恢复入口禁用。最终释放为 RELEASE_CONFIRMED,主动关闭 0/允许 0,主程序正常退出 0;未修改旧失败记录 |
| 原始文件复核 | 16:00:19 对 A/B 来源及原始中心库共 27,259 个文件重新检查大小、SHA-256,无变化;`originals-rechecked.json`。先前失败校验副本归档到 F 盘以保留证据并腾出系统盘空间,映射见 `archived-validation-copies/map.json` |
最终源码全量测试 **1,880 passed**(1 条既有 pywinauto STA 警告),记录 `F:\CbbAccept-20261112\full-scoped-final10.xml`;compileall、diff-check 通过。冻结代码和图标一致性、完整恢复、释放及原始文件复核的机器摘要见同目录 [验证 JSON](2026-11-12-scoped-release-validation.json)。历史版本是按此前选择偏好使用最后手动保存版本,READY 不表示已经人工核对版本内容或无数据回退。
## 保留限制
自动 ICDB Repair 继续关闭。现场服务器会话失联的一致故障副本、Windows 10、跨机共享库掉线尚未验收;不会自动合并、发布或关闭工单。共享符号数据库地址若原生工具自身无法删除,不能通过本程序删除文件或仅凭无进程放行。此前 A/B 输入 EDIF 导出崩溃与空白依赖库有关与否仍需独立复现,不绕过工程身份校验。
立即停止时的晚到窗口仍需要正常处理:本次 Tip of the Day 是在精确绑定的私有桌面关闭后才由应用完成会话释放,不能将此试验写成自动退出无缺口。未知/损坏检查点、尚未返回的真实共享写入继续保留未确认状态。
+39
View File
@@ -0,0 +1,39 @@
# v0.2.4-beta.1(0.2.4b1)开发候选说明
状态:待评审、待现场验收。不是自动发布指令,工单保持开放。
## 用户可见变化
- 检验、ICDB 维护、中心库管理为独立入口。ICDB 检查在后台逐项运行,其他原生任务可排队,搜索/筛选保持可用。
- 检查以当前 PCB 打开情况和本次原生错误为依据,只有明确 ICDB 错误才标记问题;不自动 Repair、Restore 或选择版本,CHECKED 不等于 READY。
- 停止队列同时停止后续后台检查;执行前再次排除 READY,校验排队不隐藏检查进度;释放失败显示实际原因。
- 原生新增、同名替换、移除和备份恢复继续要求确认及人工备份,失败结果和未完成任务逐项保留。
- 默认残留释放为正常关闭。单独确认的强制结束仅允许属于当前 Windows 用户和会话、已确认作用域和精确身份的受支持文档客户端;服务器、后台导入器和未知进程不可强制结束。
## 强制客户端操作的契约
支持 Expedition PCB、Library Manager、viewdraw 和 iCDB Project Backup。操作员须知晓客户端中其他工程可能一并退出、未保存更改可能丢失、写入中的工程可能需要恢复。确认名单先持久记录为库外 `consent.json`,失败则不执行;派发处复核身份和预览范围,句柄处复核创建时间、完整映像、用户 SID、Windows 会话与原生活动;变化时拒绝。
数据库服务器可能服务同库其他用户,命令行归属不构成独占授权。本工具不会强制结束任何 ICDB 服务器;它仍活跃或库状态未知时保留 PARTIAL,不能通过结束客户端伪造释放或修复成功。
## 安全状态和回退
单项失败且本任务确已释放时可继续独立项;未释放时保留现场、检查点、映射和证据。客户端强制结束不可撤销,恢复使用人工一致备份或原生历史版本;不会自动重新导入、回退共享库或重放未完成写操作。
程序回退:先关闭正在使用的客户端,再使用此前候选包;程序回退不恢复原生事务或丢失更改。工程恢复:保留故障现场、选择确认过的备份,在隔离副本验证后恢复并重新校验。
## 验证与限制
PR #3 comment 249 的修复通过服务器排除、确认记录失败、PID 复用、客户端/服务器混合名单等 fake 边界回归;没有为了验证本功能终止真实 Mentor 客户端或服务器。
此前 r10 的原生隔离检查与界面排队证据只证明该候选包已记录的打开/释放行为,不能当成新强制客户端功能的现场验收。现场服务器失联恢复、ICDB 自动修复、Windows 10 和共享库跨机器并发仍未验收。
规范:[SOP](../../SOP.md)、[ICDB 手动修复手册](../ICDB-MANUAL-REPAIR-AND-UPDATE.md)。依赖和拆分顺序见 PR #3 的独立审查链。
## 2026-11-06 审查修正
- 校验和管理统一复核历史占用,释放经过先留证;PASS 不再重复计入无法确认。
- 管理增加目标工程及共享依赖的写后差分,明确不覆盖其他 CBB 内容;报告失败不制造未完成原生事务。
- 释放逐项持久保存,快照异常保留已完成/未执行项;报告保存失败与独立释放证明分别展示。
- 预览与强制执行前读取原生活动,正在 Backup/Restore/Repair 或无法确认的备份客户端不可强制结束;显示后台提示、映像路径与桌面,并对唯一已识别的 Cancel/取消按钮提供确认请求。取消后的实际状态仍需复核。
- 本轮自动测试不包含真实 Mentor 的强制结束或原生取消验收;ICDB 自动修复继续暂停。
+13
View File
@@ -0,0 +1,13 @@
# 独立审查 1:只读进程探测与原生任务准入队列
基于 main `a56fd17`,不叠加 PR #3 的历史提交。
**行为变化:** 增加有界、版本化、只读 owner-probe helper 和可取消的 FIFO 原生任务队列。应用仅新增隐藏的只读 helper 分派;现有 UI、验证和原生写操作保持原来的入口。
**失败安全:** 查询超时或协议/身份异常返回未确认;暂停或取消的队列票据不会运行其动作,也不阻止后面的可执行任务。不启动或终止 Mentor 进程。
**回滚方式:** 尚未合入下游时撤销本提交即可;下游接入后须按依赖从上层向下回退。
完整回归:491 passed,1 条既有 pywinauto STA 提示。编译和 diff 检查通过。
与原评论建议的差别:`management_transaction` 直接导入管理模型和库锁;`scoped_manifest` 的协调记录与独占快照依赖库锁协议。这两者在后端审查中一起交付,避免构造导入不完整的“纯工具”基线。
+20
View File
@@ -0,0 +1,20 @@
# Native backend review unit
Base: codex/review-infrastructure. This unit integrates ICDB open diagnostics, scoped content manifests, native CBB transactions and the library-release protocol; it does not expose the new UI or residual termination action. Verification and ICDB checks remain separate, READY items are excluded from maintenance, and a released single-item failure can continue independent targets.
Failure safety: no automatic Repair/version choice, no server termination, no direct LMC edits or native lock-file deletion. Native writes are serialized, uncertain ownership retains evidence and mappings, and unfinished writes are shown without replay.
Dependency rationale: library release imports native engineering identity and retained management workers; engineering identity imports management source models and scoped manifests. Separating these into nominally independent utilities would leave runtime imports unresolved. This PR therefore keeps the backend dependency closure together; UI and user-confirmed termination remain separate later PRs.
Rollback: before dependents are merged, revert this commit. After merging dependents, revert in reverse order or use the previous candidate package after closing the current app. Code rollback does not undo a native transaction; use a consistent manual backup or explicitly selected native history on an isolated copy.
Validation: full Python suite 1,376 passed (one existing pywinauto COM-threading warning); compileall and git diff --check passed. Native field recovery, Windows 10 and cross-machine shared-library acceptance remain pending.
Review comment 344 follow-up (2026-11-09):
- Production status scans continue to query fresh bridge results. The low-level session now also defaults to fresh scans; explicit checkpoint reuse is documented as a controlled read-only option with no production resume promise.
- `library-changes.json` declares `rule_version`, the selected `scope`, excluded directory categories and suffixes, positional runtime rules, runtime filenames/prefixes, and recognized coordination checkpoints. Names, suffixes, prefixes and native database locations come from the same constants used by semantic classification. These are differences in participating engineering files; excluded content is not proven unchanged, and unrelated CBB content is not checked.
- Removed the tests-only `pcb_bridge.execution_outcome` helper. Production release safety continues to come from observed release state and remaining clients.
- Unlocked primary snapshots hash the copy stream, then independently hash the final source and destination. Source reads decrease from three to two. Rejected the suggestion to remove/threshold the final source hash: size and mtime cannot detect same-size, same-timestamp source edits. Locked LMC/icdb streams and owned lock snapshots retain their existing behavior; no source-rehash skip flag is added.
Validation for this follow-up: 287 targeted cases passed through the isolated mutex/PYTHONPATH wrapper (`test_batch_status`, `test_batch_status_faults`, `test_review_backend_followup`, `test_scoped_manifest`, `test_pcb_bridge`, `test_management_retry_blockers`, `test_backup_recovery`, `test_library_lock`, `test_management`). AST parsing, `git diff --check`, and original newline-style checks passed. No real Mentor, COM automation, shared library or process control was exercised; no full-suite run was requested.
@@ -0,0 +1,9 @@
# Residual release contract review unit
Base: codex/review-ui-integration. Behavior: expose the residual-session preview and cooperative close entry, plus a separately confirmed force action limited to recognized document clients of the current Windows user and session. Database servers, importers, unknown scope and unverified identities cannot be force ended. Version becomes 0.2.4b1, with README, SOP, manual and rollback notes aligned.
Failure safety: persist the exact allowed/excluded confirmation list before any native action, recheck scope/creation/executable/user/session at dispatch and the held handle, and require independent library-release proof after exit. Unreadable owner metadata disables force only. A terminated client is not a repaired/READY project; partial release retains evidence and mappings.
Rollback: reverse the dependent commits before reverting the backend, or close the current app and use the earlier candidate. Process termination cannot be undone, and code rollback does not restore native transactions or unsaved data. Use a consistent manual backup or an explicitly chosen native history version on a copy.
Validation: 1,601 passed for this final stack, 1,601 passed for the umbrella, and 43 focused release tests passed after the final UI wording change. Compileall, diff checks and byte-level final implementation equality passed. Detailed XML counts/hashes and content digest are recorded in ../releases/2026-11-05-pr249-review-validation.json. No real Mentor client/server was terminated for these checks. Field recovery, Windows 10, shared-library concurrency and the new frozen package remain unverified.
@@ -0,0 +1,9 @@
# Application integration review unit
Base: codex/review-native-backend. Behavior: verification, ICDB batch checking and CBB management have separate tabs; background checks yield between items, keep browsing responsive and let other native tasks queue. READY targets are excluded at execution, status refresh is scoped, and manual version/copy decisions remain explicit.
Failure safety: a released item failure can continue independent targets; uncertain native ownership retains evidence and mappings. There is no residual-session release backend in this review stage: its button and result actions are hidden, and its callback is inert. The next PR restores the complete release entry together with its tested backend, rather than exposing a button with a missing import. No automatic Repair or native version selection is added.
Rollback: revert this UI commit before its dependent release PR; after dependents, revert in reverse order or close the current program and use the prior candidate. This does not roll back library writes.
Validation: 1,545 passed (one existing pywinauto COM-threading warning), compileall and diff checks passed. Release-specific tests are introduced in the dependent PR, never discarded from the final tree. Windows 10, real shared-library concurrency and field ICDB recovery remain pending.
@@ -0,0 +1,9 @@
# 评论 275:占用复核与变更报告
校验和管理共用历史释放记录复核:旧 JSON/内存租约触发实际进程、地址、独占文件和导出副本证明,不直接视为真实占用,不结束进程或执行 Repair。确认释放后归档旧记录并继续;失败且本项尚未启动时保留原记录,结果不会把本项标成未释放。校验结果兼容新增 release_proof,管理首次原生调用前保存释放经过;进程释放不等于历史数据已恢复。
写后仅对原先的目标 Work/Cert、中心库配置和共享依赖范围生成 SHA-256 差分,报告新增、删除、修改路径,并保存写后清单。范围沿用原有 ManifestScope.library,不增加整库备份或其他 CBB 内容哈希,不声称已证明所有其他 CBB 未被波及。已释放的部分失败也保留差分;仍有原生所有者时不为报告读取数据库。报告失败单独标为 UNCONFIRMED 并返回原因,不伪造未完成事务、不制造永久占用。
PASS 不再重复计入 unconfirmed。未改变实际原生写操作、人工备份政策或服务器保护,也没有恢复自动修复。
回归覆盖历史记录无占用/真实地址占用、首次原生调用前留证、目标及共享依赖差分、失败后差分、报告保存失败后可继续。完整回归和限制以最终验证记录及 PR 描述为准;真实 Mentor、Win10、跨机器共享库尚未因此取得新验收结论。
@@ -0,0 +1,11 @@
# 评论 286:释放留证与原生活动
释放前持久保存 NOT_STARTED 检查点,动作前保存 REQUESTING,动作后逐项保存实际状态,快照超时/中途异常也保留已执行和未执行项。保存失败停止继续发出动作;最终保存失败明确标为 PARTIAL/evidence_error,但独立且匹配当前库的释放证明仍被界面接受,不伪造占用。确认意图 consent_allowed 与实际 release_status 分别记录。
强制结束预览及精确句柄处重新读取原生窗口、进度和 Output;备份客户端进行中或状态无法确认时拒绝强制结束。普通客户端的保存模态窗口不直接等同于正在写库。服务器、导入器和未知身份仍排除。
窗口快照记录桌面,预览展示完整映像路径并提供原生提示/输出查看,支持原有保存选择。仅在当前用户/会话、精确进程身份匹配,且当前备份窗口唯一存在启用的 Cancel/取消(标准 IDCANCEL=2)控件时,提供单独确认的取消请求;执行前再次匹配标题、桌面、窗口和按钮并校验 PID。该请求复用串行准入和库外留证,不能当作已取消/已释放/已修复,也没有未知菜单或前台鼠标回退。没有可识别控件时入口禁用,保留状态供重新检查。
回归覆盖初次/末次快照超时、逐项记录、报告落盘失败、活动状态变化、原生取消按钮变化、确认与取消选择、已释放事实不被报告错误覆盖。测试使用临时文件和 fake 原生边界,没有终止或取消真实 Mentor 事务。原生取消在具体 Mentor 版本/现场窗口上的效果仍待隔离实机验收;ICDB 自动修复继续暂停。
回退代码不恢复已结束进程或原生事务;工程数据恢复仍需一致人工备份或明确选择的原生历史,不自动重放写操作。
+7
View File
@@ -0,0 +1,7 @@
# 评论 275/283:结果口径接入
管理进度显示历史占用复核;结果分别呈现已确认的原生动作、变更清单完整性和历史会话释放事实。单项 RECOVERY_REQUIRED 但 safe_to_continue=True 时,不再声称整个队列停止;记录失败不伪装成原生上传/移除失败。
独立操作回归更新为实际历史记录归档流程:先证明当前占用已释放、保留归档字节,再进行普通验证;不自动 Repair,失败后不遗留新的占用锁。
本阶段相关 UI/独立操作/后端回归 29 passed。后端完整回归与最终集成回归另见验证记录。没有执行真实 Mentor 写操作;回退按依赖逆序,不代表恢复原生事务。
@@ -1,5 +1,7 @@
# CBB Batch Status Bridge Implementation Plan
Current behavior (2026-11-09): production status scans always query the bridge fresh and overwrite checkpoint evidence. `BatchStatusSession.scan` defaults to `resume=False`; `resume=True` remains an explicit low-level option for controlled read-only scans, not an exposed production resume feature. Its lightweight fingerprint cannot detect every native payload/database write, so saved READY/PENDING rows must not authorize production actions.
> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
**Goal:** Read 180 to 3000 CBB Verify states through one resilient x86 Mentor bridge session with checkpoint recovery and authoritative `READY/PENDING` results.
@@ -0,0 +1,56 @@
# PR #3 comment 249 implementation plan
**Goal:** Make the force-release contract explicit and keep shared native servers out of forced termination; replace the large PR as the merge entry with smaller, ordered review branches.
**Authorization:** The user requested implementation of comment 249. Earlier user decisions retain explicit user-confirmed residual release, manual backups and no release/issue closure before field acceptance. These decisions take precedence over changing Draft status merely to make a merge button available.
**Architecture:** Keep automatic verification and checking cooperative. A separate explicit force action can end supported document clients only after scope, exact process identity, current user SID and Windows session are checked. Native database servers and unknown processes remain observable and cannot be force terminated. Every force request must persist consent intent before any termination and still obtain independent library-release proof afterward.
**Review split:** Preserve #3 as an unmerged umbrella. Create a dependent review chain: (1) read-only owner probing and cooperative queue; (2) native backend, manifests, library coordination and transactions; (3) background checks and application UI integration; (4) residual-session release, its explicit contract, version and documentation. `management_transaction` imports management and lock APIs, and semantic manifests depend on recognized coordination checkpoints; they belong with the backend rather than pretending to be standalone read-only utilities. The existing #5 remains a prerequisite document review, not an automatically merged change.
## Global constraints
- Python/PySide6, Windows and Mentor EE7.9.5 remain the supported implementation stack.
- Never kill native servers or operate production libraries in this development task.
- No automatic Repair, Restore, version choice, issue closure, publishing or PR merge.
- Each split PR states visible behavior, failure safety and rollback.
- Preserve the current user test workspace; work in the PR worktree.
## Task 1: Force-release boundary
Files: `residual_sessions.py`, `ui/residual_release_dialog.py`, `tests/test_residual_sessions.py`, `tests/test_residual_release_dialog.py`.
- [x] Add failing tests: a scoped iCDB server cannot reach `_force_close`; direct force requests against server/unsupported names never obtain a terminate-capable handle; missing consent evidence causes zero termination calls; mixed client/server previews list only supported clients in the confirmation; rejected/reused process identities remain unchanged.
- [x] Define one shared force-eligibility function allowing only recognized document clients with confirmed scope and matching captured executable identity. Enforce it again at dispatch and at the exact handle boundary.
- [x] Persist `consent.json` with target library, exact preview identities and allowed/excluded targets before processing a force request. A write failure returns before native actions.
- [x] Keep excluded servers in the release result and require the existing independent release proof; process exit never means repaired/READY.
- [x] Run focused regressions and the complete suite with an isolated test mutex.
## Task 2: Version and operating contract
Files: `pyproject.toml`, `README.md`, `SOP.md`, `docs/ICDB-MANUAL-REPAIR-AND-UPDATE.md`, `docs/releases/v0.2.4-beta.1.md`.
- [x] Identify the actual tracked SOP before editing; update README, SOP and manual together.
- [x] Set `0.2.4b1` and document the user-confirmed client-only exception to the automatic no-force contract. Explain server handling, evidence, failure states, manual recovery and rollback.
- [x] Keep field ICDB recovery, Win10 and multiuser shared-library acceptance explicitly unverified.
## Task 3: Review branches
- [x] Inventory final files against main and assign each file to the smallest dependency-closed review stage. Keep UI/release hooks unavailable until their backend exists; do not publish branches with broken normal operation.
- [x] Create ordered branches with one concrete feature commit per review unit, no rewriting the existing umbrella history.
- [x] Test each review stage using appropriate unit/full checks; preserve all final regressions and verify the final source tree against the umbrella snapshot.
- [x] Push branches without force and create draft PRs with explicit base dependencies. Link them from #3 and record the splitting rationale.
## Task 4: Delivery
- [x] Commit and push the safety fixes and validation record to #3.
- [x] Verify remote heads, PR descriptions and clean review worktrees. Keep #3 and issues open and do not merge #5.
- [x] Prepare the final user report with PR links, passing tests and remaining field-acceptance limits.
## Delivery evidence
- Review chain: [#9](https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/9) → [#10](https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/10) → [#11](https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/11) → [#12](https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/12). Each base/head was read back from the API and matched local branches; all remain Draft and open.
- #3 is the unmerged umbrella with the review chain linked. #5 was not merged; issues #2/#4 remain open.
- Complete suites: 491 / 1,376 / 1,545 / 1,601 passed for the four review stages; umbrella 1,601; focused release tests 43. Validation hashes and limits: `docs/releases/2026-11-05-pr249-review-validation.json`.
- Final implementation files match byte for byte; compile and diff checks pass. Only review worktrees were committed; the original running r10 test workspace was preserved.
- No new frozen package or field ICDB recovery acceptance is claimed. No actual Mentor client/server termination or production-library operation was performed.
@@ -0,0 +1,370 @@
# CBB 原生操作生命周期与只读边界 Implementation Plan
> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking. 本计划默认在当前会话逐项执行,不隐含授权创建额外代理。
**Goal:** 消除保存阶段的自动强杀、管理工作进程失联后的无监督写入,以及 legacy 状态读取的隐式弹窗操作,同时保持工具可用和独立任务可继续。
**Architecture:** 复用现有 NativeQueue、LibraryLock、私有桌面和库外证据体系,补充一个保存事实解析模块与一个会话记录模块。先实现保存/停止策略,再补管理监督与跨重启观察,最后收紧读取路径并联调 GUI/CLI;不重写调度框架。
**Tech Stack:** Python 3.12、PySide6、pywin32、Windows、Mentor EE7.9.5、现有 32 位 cscript/VBS 与状态桥;不增加运行时依赖。
**Spec:** [2026-11-06-runtime-lifecycle-safety-design.md](../specs/2026-11-06-runtime-lifecycle-safety-design.md)
**Status:** 用户已于 2026-11-06 授权按本计划实施,正在独立分支执行。基线为 `5eb4b2d7ed09947f19310ba80ba3b9a2a09b6f7d`;本文件没有修改生产行为。
## Global Constraints
- 技术栈保持 Python 3.12、PySide6、pywin32、Mentor EE7.9.5,不增加运行时依赖。
- 检验、ICDB 批量检查、中心库管理保持独立;ICDB 自动 Repair 继续关闭。
- READY 条目继续排除维护;状态读取失败不伪造 READY,也不直接判定 ICDB 损坏。
- 自动停止、超时、失败清理、监督器失联都不得强杀可能正在原生写入的进程或 COM 保存 helper。
- UI、搜索、筛选、日志查看和只读浏览保持可用;一个未释放写入不得占着全局执行票据等待无限长时间。
- 同库仍在进行的写入不可并发启动;已释放的单项失败不得阻止独立项,其他独立库可排队获得执行机会。
- 历史 JSON 只触发身份及实际资源复核,不直接构成真实占用;报告错误不得伪造占用或永久禁用入口。
- 不自动选择历史版本、重新导入、恢复备份、删除原生锁/地址文件或修改 LMC 伪造状态。
- 不新增全库自动备份或全库内容哈希;此前确认的人工备份策略不变。
- 确认结束客户端与正常停止分开;本轮不增加“放弃正在进行的保存”按钮,不自动放宽服务器保护。
- 原生动作、证据读取、路径解析、进程查询和重启核对都放在非 GUI 线程;GUI 只更新视图和提交用户意图。
- 保留 VerifyRequest / VerifyResult 的现有参数及 safe_to_continue 兼容性,只追加有默认值的事实字段。
- 保持原有文件行尾,不进行与功能无关的整文件规范化。
- 开发验证只使用临时数据、模拟原生边界或隔离副本;不在生产库制造保存中断。
## 实施起点与三个交付单元
执行前重新读取 PR #10/#11/#12、main 和本地工作树,不将旧评论当成新实现的批准。另开 `codex/runtime-lifecycle-safety` 临时分支,使用独立工作树;保留用户原始脏工作区与已运行程序,不 force-push。
- **A:保存与停止闭环。** Task 1–4 作为一个功能单元验收;前面的模型和 primitive 可以独立提交,接线未完成时不交付候选程序。
- **B:管理失联与重启核对。** Task 5–6,在 A 的协议上扩展管理;不增加自动清除锁文件。
- **C:只读边界与交付验证。** Task 7–9;最终创建仅含本轮差分的草稿 PR,按 A/B/C 给出审查顺序。
### Task 1:建立保存事实解析与停止判据
**Files:**
- Create: `src/cbb_refresh_verify/operation_lifecycle.py`
- Create: `tests/test_operation_lifecycle.py`
**Interfaces:**
- Consumes: 本轮 helper 的标记序列与独立进程/资源观察结果;不读取文件或调用 COM。
- Produces: `classify_save_state(markers: Sequence[str], *, helper_exited: bool | None, native_released: bool | None, forced: bool = False) -> dict[str, str | bool]`。
- Produces: `stop_disposition(*, call_started: bool | None, call_returned: bool | None, identity_confirmed: bool) -> str`,返回 CANCEL_BEFORE_CALL / WAIT_FOR_RETURN / CLEANUP_RETURNED / OBSERVE_UNCONFIRMED。
- 返回字段固定为 save_state、close_state、helper_state、resource_state、interrupted、detail,取值见设计第 4 节;其他任务不得另造不兼容枚举。
- [x] **Step 1: 写失败测试,证明 Save/Close/释放不能互相替代。**
```python
from cbb_refresh_verify.operation_lifecycle import classify_save_state, stop_disposition
def test_saved_but_not_closed_is_not_unsaved_or_fully_released():
facts = classify_save_state(["SAVE_START", "SAVE_OK", "CLOSE_START"],
helper_exited=False, native_released=False)
assert facts["save_state"] == "COMMITTED"
assert facts["close_state"] == "UNKNOWN"
assert facts["resource_state"] == "RETAINED"
def test_process_exit_without_save_result_is_unknown():
facts = classify_save_state(["SAVE_START"], helper_exited=True, native_released=True)
assert facts["save_state"] == "UNKNOWN"
assert facts["resource_state"] == "RELEASED"
def test_an_unreturned_call_requires_waiting():
assert stop_disposition(call_started=True, call_returned=False,
identity_confirmed=True) == "WAIT_FOR_RETURN"
```
- [x] **Step 2: 运行并确认 RED。** `python -m pytest -q tests/test_operation_lifecycle.py`;预期缺少新模块/函数。再加取消前、SAVE_REJECTED、CLOSE_FAILED、日志截断/矛盾、forced=True 的行为断言,不把进程退出断言成保存完成。
- [x] **Step 3: 实现纯判据。** 核心映射从明确证据出发:
```python
save_state = ("COMMITTED" if "SAVE_OK" in markers else
"REJECTED" if "SAVE_REJECTED" in markers else
"NOT_STARTED" if "CANCELLED_BEFORE_SAVE" in markers else "UNKNOWN")
resource_state = ("RELEASED" if native_released is True and helper_exited is True else
"RETAINED" if native_released is False or helper_exited is False else "UNKNOWN")
```
不把 UNKNOWN 当 False;矛盾标记降为 UNKNOWN 并记录原因。`forced=True` 不能把 UNKNOWN 的保存状态升级为成功。
- [x] **Step 4: 运行该文件及 `tests/test_runtime_timing.py`,确认已有标记消费保持兼容。**
- [x] **Step 5: 提交。** `git commit -m "feat: distinguish save close and resource release facts"`。
### Task 2:保存 helper 的启动门和完整返回事实
**Files:**
- Modify: `src/cbb_refresh_verify/save_verified_block.vbs`
- Modify: `src/cbb_refresh_verify/desktop_worker.py:17`(launch_save_helper)
- Modify: `src/cbb_refresh_verify/mentor_adapter.py:818`(保存启动和等待)
- Modify: `tests/test_save_helper_runtime.py`
- Modify: `tests/test_failure_lifecycle.py`
- Create: `tests/test_save_helper_identity.py`(基线没有此文件,用于 helper 启动门和精确身份测试)
**Interfaces:**
- Consumes: Task 1 的 `classify_save_state`。
- Produces: VBS 标记 CANCELLED_BEFORE_SAVE / SAVE_START / SAVE_OK / SAVE_REJECTED / POST_SAVE_FAILED / CLOSE_START / CLOSE_OK / CLOSE_FAILED,以及既有 LICENSE_OK / IS_SAVED / QUIT_OK。
- `launch_save_helper(command, **kwargs)` 保持原调用兼容;生产调用附加第 4 个 VBS 参数 `start_gate`,helper 未放行时不进入 COM;已有三参数 harness 保留兼容覆盖。
- helper 启动门在 30 秒内未放行或取消时,以明确的启动前取消结果退出。门放行前记录精确 helper 身份,不靠事后按父 PID 猜测。
- [x] **Step 1: 扩展真实 cscript harness,先复现保存中取消和保存后关闭失败。** 继续使用 `tests/test_save_helper_runtime.py` 中实际 VBS 源码,仅替换 COM 创建。FakeDoc.Save 内创建取消文件后返回 True;FakeDoc.Close 返回 False 或延迟返回。断言:
```python
assert "SAVE_OK" in completed.stdout
assert "SAVE_REJECTED" not in completed.stdout
assert "CLOSE_FAILED" in completed.stdout
assert completed.returncode != 0
```
另写门未打开的 harness,在 FakeDoc.Save 中写 `save-called.txt`;取消/门超时后该文件必须不存在。
- [x] **Step 2: 运行 `python -m pytest -q tests/test_save_helper_runtime.py tests/test_failure_lifecycle.py -k "save or helper"`,确认新行为失败。** 不运行真实 Mentor 保存。
- [x] **Step 3: 按调用边界写标记,取消只阻止尚未开始的写入。** 局部错误检查示例:
```vb
CheckCancellationBeforeSave
WScript.Echo "SAVE_START"
On Error Resume Next
operationOK = doc.Save()
saveError = Err.Number
saveDescription = Err.Description
Err.Clear
On Error GoTo 0
If saveError <> 0 Then
WScript.Echo "SAVE_RESULT_UNKNOWN|" & CStr(saveError) & "|" & saveDescription
WScript.Quit 11
End If
If Not operationOK Then
WScript.Echo "SAVE_REJECTED"
WScript.Quit 4
End If
WScript.Echo "SAVE_OK"
```
把 `saveError/saveDescription` 加入 Option Explicit 的 Dim。IsSaved/后置同步/Close 使用同样局部检查并写各自原因,不抹掉 SAVE_OK;取消后只在精确文档、已保存且没有未知模态提示时做既有正常关闭,不再次 Save、不自动放弃脏文档。
- [x] **Step 4: 保存等待不再因取消文件出现就直接抛错丢弃监督。** 本任务先提供可观察的调用返回事实;Task 4 再统一所有生产停止入口,不能保留“保存脚本启动后随便 terminate”的中间交付。
- [x] **Step 5: 重跑真实 cscript harness、身份核对和 runtime_timing 回归,通过后提交。** `git commit -m "fix: preserve native save results across cancellation"`。
### Task 3:受控会话记录和后台观察交接
**Files:**
- Create: `src/cbb_refresh_verify/session_record.py`
- Create: `tests/test_session_record.py`
- Modify: `src/cbb_refresh_verify/desktop_worker.py`
- Modify: `src/cbb_refresh_verify/library_lock.py`
- Modify: `src/cbb_refresh_verify/network_path.py`
- Test: `tests/test_network_release_safety.py`
**Interfaces:**
- Produces: `write_session_record(path: Path, record: Mapping[str, object]) -> None`,原子 pending→replace,flush/fsync。
- Produces: `read_session_record(path: Path, *, evidence_root: Path) -> dict`,协议 `cbb-runtime-session`、version=1,最大 64 KiB,拒绝重解析点/越界路径。
- Produces: `reconcile_session_record(path: Path, *, evidence_root: Path) -> dict`,仅后台核对,返回 ACTIVE / RELEASED / IDENTITY_CHANGED / UNCONFIRMED、原因、匹配库及独立 release_proof;不终止进程、不执行写操作。
- Produces: `observe_retained_session(path: Path, *, evidence_root: Path, stop=None, progress=None) -> dict`,1 秒为观察间隔,单次进程查询有界;退出观察不表示原生进程退出。
- immutable `session.json`、单写者 `worker-state.json`、watchdog 的 `supervisor-lost.json`、观察者的 `observation-<observer_id>.json` 分开,不互相覆盖。
- 等待循环只使用精确句柄的零等待观察,不持有 `_guard`/全局原生互斥,不循环扫描全库。退出后才尝试独立释放证明;通道忙只延后该观察,不把其他库也设为不可用。
- [x] **Step 1: 写协议/重启身份测试。** 使用临时 JSON 和本测试创建的 sleep 子进程;在记录上替换 created 或 executable,确保返回 IDENTITY_CHANGED 而不是结束新进程。损坏 JSON、超过 64 KiB 和越界目录返回 UNCONFIRMED。
```python
def test_record_outside_controlled_root_is_observation_only(tmp_path):
from cbb_refresh_verify.session_record import reconcile_session_record
owned = tmp_path / "owned"
owned.mkdir()
external = tmp_path / "external-session.json"
external.write_text("{}", encoding="utf-8")
outcome = reconcile_session_record(external, evidence_root=owned)
assert outcome["status"] == "UNCONFIRMED"
assert outcome["release_proof"] is None
```
PID 复用测试在本测试创建的进程身份记录上替换 created/image,再走同一个 reconcile 入口;只模拟外部进程观察边界,不用实现返回值代替被测流程。
- [x] **Step 2: 运行 `python -m pytest -q tests/test_session_record.py`,确认 RED。**
- [x] **Step 3: 实现精确身份、记录路径约束与独立释放证明。** 核对 supervisor/worker/helper/native 的创建时间、映像和当前用户/Windows 会话。PID 不在快照中不能直接生成 RELEASED;复用现有 `release_retained_library` 证明,不删 `.lock`/sAddress.adr。
- [x] **Step 4: 增加后台交接测试。** A 库 helper 保持活跃,转观察后释放 NativeQueue action;同库写请求仍等待,B 库请求可运行。映射直到匹配库独立释放证明成立才归还;调用 `process.close()` 不得丢失观察所需的精确句柄。
- [x] **Step 5: 验证报告写失败不终止活跃调用、不把记录存在本身当占用。** 内存观察保留,下一次可重新核对。通过后提交 `git commit -m "feat: persist and reconcile owned native sessions"`。
### Task 4:统一用户停止、超时、finally、kill 和 watchdog
**Files:**
- Modify: `src/cbb_refresh_verify/verify_runner.py:231`、`:951`、`:1067`、`:1090`、`:1128`
- Modify: `src/cbb_refresh_verify/desktop_worker.py:25`、`:174`、`:182`、`:191`、`:224`
- Modify: `src/cbb_refresh_verify/mentor_adapter.py:830`、`:888`、`:1005`
- Modify: `src/cbb_refresh_verify/verify_runner.py`(VerifyResult 默认字段与兼容 auto_repair 停止传播)
- Modify: `src/cbb_refresh_verify/defaults.py`(NATIVE_WRITE_STOP_WAIT_SECONDS)
- Modify: `src/cbb_refresh_verify/native_repair.py:127`(增加 stop=None 并向现有 run_native_repair 传播;默认 auto_repair 不变)
- Test: `tests/test_failure_lifecycle.py`
- Test: `tests/test_abnormal_shutdown.py`
- Test: `tests/test_verify_worker_stability.py`
- Test: `tests/test_incident_safety.py`
**Interfaces:**
- Consumes: Task 1–3 的事实解析、停止判据与后台观察。
- `_stop_worker(process, *, timeout=5)` 保持 bool 返回兼容,但不再对未知/活跃写调用自动 kill;返回 False 表示尚未确认 worker 退出,调用方转受控观察。
- `VerifyResult.write_outcome: dict | None = None`、`session_record_path: Path | None = None` 追加在现有默认字段之后。
- `attempt_recovery(request, diagnosis, evidence_root, *, progress=None, stop=None)` 只补取消传播,不使其默认启用。
- 保存等待转后台的默认界限为 120 秒,放在 `defaults.py` 的 `NATIVE_WRITE_STOP_WAIT_SECONDS`;普通合作式停止等待保持 5 秒。
- [x] **Step 1: 替换冲突的旧期望并增加跨组件失败测试。** `test_stop_cancels_then_stops_automation_client_after_deadline` 不能再锁定保存中会 kill 的行为。用本测试创建的 helper 进程触发 `_stop_worker`,使 `kill` 和 `stop_automation_clients` 在活跃 Save 时被调用即失败:
```python
class SavingWorker:
def request_stop(self): self.requested = True
def wait(self, timeout): raise TimeoutError("native Save pending")
def poll(self): return None
def kill(self): raise AssertionError("pending Save was killed")
def stop_automation_clients(self): raise AssertionError("save helper was killed")
def test_cancel_preserves_pending_save():
worker = SavingWorker()
assert _stop_worker(worker, timeout=0) is False
assert worker.requested
```
这个 fake 只用于停止分派边界;另用实际 sleep/helper 与真实 VBS harness 验证生命周期,不只断言 mock 被调用。
- [x] **Step 2: 覆盖五个触发入口。** 用户停止、活动超时、对话框错误、finally 收尾、父进程退出分别触发;verify 生产配置 `stop_clients=True` 必须在测试中出现,不只测默认参数 False。
- [x] **Step 3: 接入同一阶段判据。** 原生调用可能进行中时只写取消意图;短等待后不关闭观察句柄,120 秒后交 Task 3 观察。只有已证明未进入写入的辅助进程允许终止。`DesktopWorker.kill` 及 watchdog 不能绕开此判据;保存 helper 的注册/启动门与 supervisor 失联竞争必须被覆盖。
- [x] **Step 4: 清理前采样并保留中断事实。** 更新 session-release/outcome/result/recovery-required 的生成逻辑。后续发现进程退出可以更新 resource_state,但不能把未知保存重写成成功;正常关闭失败保留 COMMITTED。`safe_to_continue=True` 不删除 write_outcome 中的失败。
- [x] **Step 5: 跑 `tests/test_operation_lifecycle.py tests/test_session_record.py tests/test_failure_lifecycle.py tests/test_abnormal_shutdown.py tests/test_verify_worker_stability.py tests/test_incident_safety.py`,通过后执行交付单元 A 的完整回归并提交。** `git commit -m "fix: honor native write boundaries on every stop path"`。
### Task 5:管理 worker 的监督器失联与事务边界取消
**Files:**
- Modify: `src/cbb_refresh_verify/native_management.py:132`、`:209`、`:694`、`:1131`
- Modify: `src/cbb_refresh_verify/management.py`(run_management / run_management_batch)
- Modify: `src/cbb_refresh_verify/management_transaction.py`(沿用原 nonce/lease 身份,不放宽验证)
- Create: `tests/test_management_supervision.py`
- Test: `tests/test_management_native_transaction.py`
- Test: `tests/test_management_retry_blockers.py`
**Interfaces:**
- Consumes: Task 3 的会话记录,Task 4 的阶段停止规则。
- 管理 job 新增 `cancel_signal`、`session_record_path`、`start_gate` 与 supervisor 创建身份;原 desktop/supervisor_pid/timeout_seconds 保持兼容。
- 新增私有 `_require_management_continuation(job, root)`:每次新原生调用前校验当前取消/监督器身份;被取消时抛已有管理错误并明确 native_started,而不是启动下一动作。
- 现有 `_require_add_supervisor(job, root)` 调用共享判据并保留 add deadline 检查,不移除已有安全检查。
- [x] **Step 1: 写无监督器和双阶段 update 测试。** 在 catalog 前使 supervisor 退出,确保不创建 LM;remove 调用内部模拟父进程退出但保留 native owner,返回后确保 add 不调用。断言部分完成与资源释放分别记录:
```python
assert result.status == "RECOVERY_REQUIRED" # remove 已发生,add 未发出
assert adapter.calls == ["remove"]
assert result.safe_to_continue is True # 本例原生 owner 已独立证明释放
assert not any(call[0] == "add" for call in adapter.calls if isinstance(call, tuple))
```
- [x] **Step 2: 对 catalog/add/remove 三条 worker 路由分别触发取消和父进程退出。** 正在 native wait/COM 调用时 helper/worker 存活,返回后不进入下一写动作;同目标不得自动重试。
- [x] **Step 3: 启动前持久登记并使用 gate,worker 在身份检查后 armed watchdog。** watchdog 只写自己的失联文件和取消信号,不异步退出正在做原生调用的工作进程。调用返回后执行正常自有会话收尾;未知弹窗记录并交已有提示入口。
- [x] **Step 4: 将仅依赖 `_pending_workers/_retained_workers` 的判据补为内存+可核对记录。** 内存列表继续用于同一进程的快速收尾,但不是唯一恢复入口;读不到 result.json 不能永久推断 owner 活着。
- [x] **Step 5: 跑管理三个测试文件与 `tests/test_native_management.py tests/test_native_export_guard.py`,通过后提交。** `git commit -m "fix: supervise management workers across cancellation and parent loss"`。
### Task 6:重启核对和相关库范围的可操作入口
**Files:**
- Modify: `src/cbb_refresh_verify/session_record.py`
- Modify: `src/cbb_refresh_verify/residual_sessions.py`
- Modify: `src/cbb_refresh_verify/ui/residual_release_dialog.py`
- Modify: `src/cbb_refresh_verify/ui/main_window.py`
- Test: `tests/test_session_record.py`
- Test: `tests/test_review_release_followup.py`
- Test: `tests/test_ui_lease_lifecycle.py`
- Test: `tests/test_background_queue_regressions.py`
**Interfaces:**
- Consumes: A/B 单元生成的 session.json/worker-state/observation 记录,保留已存在的 release_proof / released Signal 兼容。
- GUI 后台核对只显示状态,不自动点击原生提示、不重放新增/更新/移除、不仅按 Cbb 桌面前缀结束进程。
- NativeQueue 使用已有 ready 回调查询已缓存的相关库可执行性;ready 自身不做 WMI、文件读取或 SMB 路径解析。
- [x] **Step 1: 写“重启时 active、exited、PID reused、记录损坏”四个场景。** 检查行为结果而不是只检查某函数调用次数:active 可看原生阶段,exited 需独立证明,reused 没有 native 消息,损坏记录不禁用全部 UI。
- [x] **Step 2: 写 A 库等待、B 库可执行的队列测试。** 暂停保存的 A 保留关联租约,不占队列 active;B 可完成,A 同库新写操作不会进入。用户取消批次不会在恢复后自动重放。
- [x] **Step 3: 接入启动时的后台发现与已有“查看原生提示/处理保存/请求原生取消/重新检查”。** 未知窗口只展示真实原因;记录 helper 正在 Save/Close 时不向普通强制结束路径放行。不为“能释放”而新增任意杀进程按钮。
- [x] **Step 4: 证明成立后只解除匹配会话的历史记录和映射。** 关闭自有 desktop 句柄,不能宣称已删除仍有原生引用的桌面。报告保存失败单独展示,不把有效释放反转成真实占用。
- [x] **Step 5: 跑上述四个文件与 `tests/test_residual_sessions.py tests/test_residual_release_dialog.py`,再跑 A+B 完整回归后提交。** `git commit -m "fix: reconcile retained sessions without blocking independent libraries"`。
### Task 7:关闭 legacy 读取的隐式启动和肯定按钮回退
**Files:**
- Modify: `src/cbb_refresh_verify/library_reader.py:249`、`:340`
- Modify: `src/cbb_refresh_verify/mentor_adapter.py:1453`、`:1532`、`:1629`
- Modify: `src/cbb_refresh_verify/ui/main_window.py:114`
- Create: `tests/test_readonly_status_policy.py`
- Test: `tests/test_library_reader.py`
- Test: `tests/test_batch_status_faults.py`
- Modify: `tests/test_desktop_worker.py`(旧 reader 必启动私有桌面的测试改为可达新边界)
**Interfaces:**
- 默认 authoritative_status_map 入口不变;桥返回失败仍使用 UNKNOWN / UNAVAILABLE,不能升级为 PENDING/READY。
- `read_library_flag(library, name, *, timeout_seconds=90, log_path=None)` 与 `read_library_tree_image_map(library, names, *, timeout_seconds=45, log_path=None)` 保留调用签名和返回形状,但只观察已存在、匹配库身份的 LM;不自动启动新的 LM。
- 无匹配 LM 返回原有 FLAG_UNKNOWN 或 `{name: None}`,并附可读原因,不要求每次搜索/过滤都人工开 LM。
- `handle_known_dialogs` 留给需要的已验证写路径;状态读取不调用它,不修改其他流程的弹窗规则。
- [x] **Step 1: 写未知提示和无 LM 的回归。** 把 CreateDesktop/CreateProcess/PostMessage 与 affirmative handler 设为触发即失败,调用两个兼容 reader 和 `scan_open_library_status`:
```python
def test_status_read_does_not_launch_or_answer_when_lm_is_absent(monkeypatch):
import win32process
from cbb_refresh_verify import mentor_adapter
monkeypatch.setattr(win32process, "CreateProcess",
lambda *a, **kw: (_ for _ in ()).throw(AssertionError("implicit LM launch")))
# 该测试文件的 FakeDesktop.windows() 返回 [],不连接实际 LM。
assert mentor_adapter.read_library_tree_image_map("C:/Fixture/lib.lmc", ["A"]) == {"A": None}
```
再对“修复/覆盖/保存/未知许可提示”的窗口快照参数化,要求没有任何消息发送。
- [x] **Step 2: 运行 `python -m pytest -q tests/test_readonly_status_policy.py`,确认 RED。**
- [x] **Step 3: 去掉读取路径的自动创建和通用肯定应答。** 抽取最小既有 LM 树观察函数,使用 exact library identity,只读树项,不关闭用户打开的窗口;窗口不可读就返回未知。若发现其他调用方仍依赖隐式启动,迁到明确的原生工具打开入口或保留未知,不隐藏恢复原行为。
- [x] **Step 4: 跑 reader/bridge/UI 状态回归,并用混合 READY/PENDING/UNAVAILABLE 清单检查数量和筛选。** 默认桥成功时不改变结果;legacy 无 LM 不弹每项确认、不启动 Mentor、不伪造损坏。
- [x] **Step 5: 提交。** `git commit -m "fix: keep legacy status reads observational"`。
### Task 8:GUI/CLI 停止传播和结果展示联调
**Files:**
- Modify: `src/cbb_refresh_verify/verify_runner.py:1251`(run_batch 默认 verifier 转发 stop)
- Modify: `src/cbb_refresh_verify/ui/main_window.py:1504`
- Modify: `src/cbb_refresh_verify/ui/icdb_check_task.py`
- Modify: `src/cbb_refresh_verify/ui/management_dialog.py`
- Modify: `src/cbb_refresh_verify/application.py`(CLI 信号进入本批次 stop)
- Test: `tests/test_operation_queue.py`
- Test: `tests/test_ui_batch_safety.py`
- Test: `tests/test_background_queue_regressions.py`
- Test: `tests/test_application.py`
**Interfaces:**
- `run_batch(..., stop=event)` 的默认调用必须实际传 `run_verify(..., stop=event)`;外部注入 verifier 的原一参数签名保持不变。
- UI 接受 write_outcome 和 session_record_path,但不直接读取/resolve 这些路径;通过现有 ServiceWorker 接收事实。
- 文案分别为“已请求停止,等待当前保存返回”“保存已返回,关闭待确认”“已转后台观察,可查看状态”“当前状态无法确认”,只在对应事实成立时显示。
- [x] **Step 1: 写 CLI 默认 verifier 的 stop 传播测试与 GUI 延迟保存测试。** GUI 点击停止后 200ms 定时器仍可触发、搜索仍可编辑;新原生动作没有发出,进度不假称已保存或已释放。
- [x] **Step 2: 写 action 抛异常、队列取消和按库 ready 的回归。** 保留已存在的 Event 轮询取消语义;不用无通知无限 wait;一项异常释放队列票据,后续可执行独立项不被遗留票据挡住。
- [x] **Step 3: 接入停止来源与事实投影。** `auto_repair=False` 保持默认;兼容修复/复验的 stop 转发使用 Task 4 新签名。Save/Close 尚未确认时不启用本目标自动重试,也不禁用所有只读操作。
- [x] **Step 4: 跑队列、后台、界面和 CLI 测试,检查退出/重启后未完成状态仍可见。** GUI 线程中将 Path.resolve、WMI、文件读取替换为失败哨兵,确认事实展示不触发它们。
- [x] **Step 5: 提交。** `git commit -m "fix: propagate stop intent and display independent lifecycle facts"`。
### Task 9:完整回归、冻结包和隔离验收
**Files:**
- Modify: `README.md`
- Modify: `SOP.md`
- Modify: `docs/ICDB-MANUAL-REPAIR-AND-UPDATE.md`
- Create: `docs/releases/2026-11-06-runtime-lifecycle-validation.json`
- Create: `docs/releases/2026-11-06-runtime-lifecycle-candidate.md`
- Modify: `pyproject.toml`、`src/cbb_refresh_verify/__init__.py`(只在实际生成新候选包时递增为 0.2.4b2)
- Test: `tests/test_packaging.py`、`tests/test_packaging_status_bridge.py`
**Interfaces:**
- 验证记录必须关联最终 source tree SHA、测试 XML 哈希、实际版本、候选包 SHA-256、平台/Mentor 版本;缺少真实包或环境的项使用 not_run/unverified,不能写 passed。
- 旧 `0.2.4b1` / `5eb4b2d` 的 1,631 完整回归和 63 项专项为历史基线,不是本轮验收结果。
- [x] **Step 1: 在独立测试互斥/临时恢复目录下串行跑最终完整套件。** `python -m pytest -q` 必须来自新工作树 src,子进程 PYTHONPATH 同步。不要同时在同一 runner 跑另一个使用真实产品互斥或固定恢复根的全量任务。
- [x] **Step 2: 跑 `python -m compileall -q src tests` 和 `git diff --check`。** 保持源文件原行尾;记录预期 pywinauto COM 提示,新的警告另查。
- [x] **Step 3: 生成候选包前更新为 0.2.4b2,按现有 spec 打包并运行 frozen 入口冒烟。** 验证 verify-worker、management-worker、owner-probe、dialog-agent 的模块/脚本可加载;运行“冻结程序 + fake 保存/管理 native 边界”的父进程退出与停止测试。构建环境缺外部依赖时记录具体缺项,不宣称已产生候选包。
- [ ] **Step 4: 通过应用实际按钮运行隔离副本。** 单独记录 Save 前停止、Save 中停止、Close 延迟、管理父进程退出、legacy 未打开 LM、未知原生提示、独立库继续。缺实机场景保留未验收;不为验收在生产库保存中杀进程。
- [x] **Step 5: 更新文档与审查描述。** 写清默认停止规则、后台观察范围、只读未知状态、真实 COM 无法取消的限制、程序回退与数据恢复的区别。测试用例数量不得替代场景证据。
- [x] **Step 6: 提交候选记录,推送本轮差分草稿 PR。** `git commit -m "docs: record runtime lifecycle validation and candidate limits"`;按 A/B/C 提供审查顺序,不自动发布或关闭 ICDB 工单。若 main 已变化,重新整合并验证实际候选树。
## 验收映射与自查
| 审查问题 | 计划任务 | 必须看到的证据 |
| --- | --- | --- |
| P1-1 Save/Close 中自动强杀及误记安全 | 1、2、3、4、8 | 实际 helper 存活、无 terminate、保存/关闭/释放独立结果、跨入口停止一致 |
| P1-2 管理失联后无监督操作 | 3、5、6 | 启动门、精确持久身份、失联取消、调用返回后无后续写入、重启可核对 |
| P1-3 legacy 读取隐式启动/肯定应答 | 7、8 | 读取路径无 CreateProcess/点击/用户窗口关闭,未知状态可见 |
| 直接相关停止传播和保存状态记录 | 1、2、4、8 | CLI/GUI 同一 stop,兼容修复链不吞 stop;COMMITTED 与关闭失败并存 |
| 防止修复又导致整批/全工具卡死 | 3、6、8 | 有界等待后转观察;B 库执行,A 库活跃写入不并发;UI 始终可响应 |
| 源码、冻结包和现场验收口径 | 9 | 三层分别记录,缺哪层明确保留,不能互相代替 |
本轮排除项见设计第 13 节。源文件行号以基线供定位,实施时先用 rg 重新定位所有调用方,不直接按旧行号打补丁。
实施记录:用户已授权执行。实施在当前会话逐项完成;独立只读审查按 requesting-code-review 技能执行,没有并行修改源码。保存启动门依赖会话协议,故先提交协议 primitive(97a3de2);mentor_adapter 的停止与只读修改共用文件,在 8d3eab1 合并提交 A/C 基础接线。管理接线为 6278091,界面及按库交接为 cb11624,独立审查修正为 d9a79b6、051498b。三单元联通后统一验证、生成候选,未提前交付不完整的 A 单元。真实环境未验收项见候选记录。
@@ -0,0 +1,25 @@
# Latest reviews for PR #10/#11/#12
Scope: user selected only the 2026-11-09 comments #309/#310/#313. Earlier deep-review backlog stays out of scope. Preserve ICDB automatic Repair OFF, serial writes, current icon and independent-library admission. No merge to main, issue closure or release.
## PR #10
- [x] Reproduce BlockA releasing historical BlockB checkpoints and retained owners without naming BlockB in proof.
- [x] Record requested block, actual proof scope, checkpoint identities and foreign names in existing release proof. Preserve original checkpoint bytes and data_recovery_confirmed=False; no new blocking gate.
- [x] Run backend/lease regressions, compile/diff checks, commit on codex/review-native-backend.
## PR #11/#12
- [x] Forward-merge backend into UI, then UI into residual branch; no history rewriting.
- [x] Test one unreadable process/thread not affecting another, without fabricated modal windows; cover preview and action-time rechecks.
- [x] Enumerate only target-process threads/windows, store incomplete-read metadata per PID. Keep unknown/active writers protected rather than adding a blind force override.
- [x] Persist window identities and at most a 4 KiB output excerpt, preserve full preview in memory, stop duplicating activity output.
- [x] Carry exact preview block scope into native Cancel and recheck identity/scope before clicking.
- [x] Run residual/UI regressions, compile/diff checks, commit on residual branch.
## Chain tip
- [x] Merge residual fixes into PR #13, preserving ca39fb3 and icon.
- [x] Address #11's line-ending note once at the tip: application.py to LF; pin it and main_window.py with targeted .gitattributes exceptions. Preserve all other file endings and the global * -text rule; EditorConfig is unnecessary.
- [x] Independent review, integrated full suite, compile/diff checks, validation records and fast-forward pushes to existing PR branches.
- [x] Update PR descriptions with precise results and no new package/native acceptance. Leave #3 as historical umbrella, without merging or closing it.
@@ -0,0 +1,32 @@
# PR #13 comment 319: residual lifecycle fixes
**Goal:** Fix the P2 and three P3 findings against head `96e179b` without interrupting an in-flight native writer.
**Scope:** https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/13#issuecomment-319
**Constraints:** Python 3.12/PySide6/pywin32, no new dependencies; existing isolated worktree and icon retained; no automatic ICDB Repair, native process termination, merge, release or issue closure. Real Mentor and shared-network acceptance remains outstanding.
## 1. Release proof and queue state
- [x] Reproduce an ended UNCONFIRMED observer leaving its library unavailable after explicit successful release.
- [x] Snapshot retained entries before proof; acknowledge only those entries, for the matching library, while proof handles are still held. Failed proof and newer handoffs remain blocked. Test independent libraries and queue resumption.
- [x] Retain final observer states for the existing residual panel, including unreadable journals; expose fresh recheck through that panel.
## 2. Passive discovery
- [x] Test that discovery of an exited session never calls release proof or modifies library checkpoints.
- [x] Separate process observation from explicit reconciliation; startup observation stays read-only, reports ACTIVE/UNCONFIRMED and cannot claim RELEASED without proof.
- [x] Refresh journal/status information from the running residual dialog, keeping all I/O off the Qt thread.
## 3. Supervisor loss
- [x] Use harmless real subprocesses to test PRECALL/POSTCALL exit, NATIVE_CALL/CLOSING retention, missing/unreadable journals and failed cancellation/journal writes.
- [x] Synchronize phase transitions and helper registration with watchdog shutdown. Only matching durable and in-memory non-pending state allows worker-only exit; preserve active helpers and unknown states. Prevent subsequent calls after supervisor loss.
- [x] Recheck after retained calls return; do not kill a native child or interpret worker exit as release proof.
## 4. Dead code and validation
- [x] Delete unused `_terminate_pid`; replace obsolete test hooks with assertions at actual process termination boundaries.
- [x] Run focused regressions, full Python suite, compileall and diff checks. Record results and field-validation limits, then commit scoped changes. Do not rebuild or publish a candidate without documenting its source and validation.
Execution: implement inline in the existing isolated worktree; red/green tests per task. The user has authorized these fixes, so no further plan confirmation is needed.
@@ -0,0 +1,13 @@
# 占用范围、释放及原生确认实施计划
按用户已确认的误判复核方案执行,沿用隔离分支 codex/runtime-lifecycle-safety;原工作区和其他软件会话不改动。
1. 先以真实外部命令回放补充失败测试:未知外部观察项不接收关闭请求,目标独占资源证明成功后不阻止本库释放;真实本库进程、在途助手、数据库句柄或地址仍阻止。统一库证明与维护入口的进程候选判断,保留观测证据。
2. 修正释放结果:观察项与可操作目标分开,未允许且自然退出不计主动释放;对绑定身份已退出的旧 UNCONFIRMED 日志不再认定调用仍在执行。覆盖 PID 复用和读取失败。
3. 单项失败后,先尝试本库/本工程新鲜释放证明;证明完成才解除相应保留状态并继续,无法证明的共享占用仍停止相关写入。普通校验/管理在完整单项事务结束后让出队列,支持等待及取消。
4. 普通 PCB 打开接通精确读写提示的用户确认;复用现有异步选择模型,绑定 PID、工程、私有桌面、当前提示,取消/跳过不自动写入。自动 Repair 保持关闭;恢复副本也传递用户选择。识别未知提示不猜测点击。
5. 每项先红后绿;全量回归、编译、冻结代码/图标一致性;其他软件独立工程持续运行、真实目标句柄占用、停止后重试、提示确认/取消、真实校验与备份恢复。EDIF 崩溃单独保留原生失败证据,不跳过身份校验。尚未达到 READY 不宣称验收通过。
主要代码:library_lock.py、residual_sessions.py、session_record.py、verify_runner.py、management.py、ui/icdb_check_task.py、ui/management_dialog.py、ui/main_window.py、ui/residual_release_dialog.py、backup_recovery.py;原生确认按现有 pcb_restore_prompt 和 dialog_agent 模式实现。
验证记录写入 docs/releases,原始运行证据放 F 盘独立验收目录。开发与原生验收串行执行,所有原生操作局限隔离副本。
@@ -1,5 +1,7 @@
# CBB 批量状态桥设计
Current behavior (2026-11-09): production status scans always query the bridge fresh and overwrite checkpoint evidence. `BatchStatusSession.scan` defaults to `resume=False`; `resume=True` remains an explicit low-level option for controlled read-only scans, not an exposed production resume feature. Its lightweight fingerprint cannot detect every native payload/database write, so saved READY/PENDING rows must not authorize production actions.
## 目标
支持对 180 到 3000 个 CBB 进行 Mentor authoritative Verify 状态读取,要求单条失败隔离、过程可恢复、结果可审计,并避免现有“一条 CBB 启动一个桥进程”的启动开销。
@@ -0,0 +1,201 @@
# CBB 原生操作生命周期与只读边界设计草案
状态:用户已批准并授权实施。运行时实现与验证见同日计划和候选记录;不代表现场 ICDB 恢复验收。
范围已由用户确认:三条 P1,加上直接相关的停止与状态记录问题。
代码基线:`5eb4b2d7ed09947f19310ba80ba3b9a2a09b6f7d`。审查来源:[PR #12 评论 297](https://git.simgreek.com/Skyworth_Digital_Hardware/cbb-refresh-verify-795/pulls/12#issuecomment-297)。评论分析的是旧 `dc536da`;以下结论重新对照当前基线,不把旧评论当作最新提交的复审。
## 1. 要解决的实际问题
1. 用户停止、超时、失败清理和监督器失联可经不同路径结束保存 helper。杀掉调用 `doc.Save()` 的 cscript 后,退出查询又可能将它描述为已安全释放。
2. 管理工作进程缺少统一的监督器失联处理和可跨重启核对的会话记录。新增路径已有 `_require_add_supervisor`,但 catalog、remove、COM 等待和收尾覆盖不一致。
3. legacy 状态读取会启动私有桌面的 LM,并调用偏向肯定按钮的通用弹窗处理器。默认权威桥路径没有这项图标读取回退,不能把风险泛化成所有读取都会执行它。
本轮修复停止与监督行为,不能据此宣布现场 ICDB 失联已修好,或推断已经验证了保存中断的数据后果。
## 2. 方案比较与选择
| 方案 | 代价与限制 | 选择 |
| --- | --- | --- |
| 按异常逐处增加判断 | 改动初期较少,但用户停止、finally、kill、watchdog 容易再分叉;不能解决重启后的身份核对 | 不采用 |
| 小型事实模型、按原生阶段停止、复用现有队列和租约 | 集中约束危险动作,同时保留现有串行操作、后台检查和确认入口;可分三个交付单元审查 | 推荐 |
| 重写调度与自动化框架 | 改动大,可能重新引入阻塞,且难以定位回归 | 不采用 |
本轮只增加保存事实解析和受限会话记录两个小模块,不引入通用流程引擎、任务总线或新依赖。
## 3. 全局约束
- 技术栈保持 Python 3.12、PySide6、pywin32、Mentor EE7.9.5,不增加运行时依赖。
- 检验、ICDB 批量检查、中心库管理保持独立;ICDB 自动 Repair 继续关闭。
- READY 条目继续排除维护;状态读取失败不伪造 READY,也不直接判定 ICDB 损坏。
- 自动停止、超时、失败清理、监督器失联都不得强杀可能正在原生写入的进程或 COM 保存 helper。
- UI、搜索、筛选、日志查看和只读浏览保持可用;一个未释放写入不得占着全局执行票据等待无限长时间。
- 同库仍在进行的写入不可并发启动;已释放的单项失败不得阻止独立项,其他独立库可排队获得执行机会。
- 历史 JSON 只触发身份及实际资源复核,不直接构成真实占用;报告错误不得伪造占用或永久禁用入口。
- 不自动选择历史版本、重新导入、恢复备份、删除原生锁/地址文件或修改 LMC 伪造状态。
- 不新增全库自动备份或全库内容哈希;此前确认的人工备份策略不变。
- 确认结束客户端与正常停止分开;本轮不增加“放弃正在进行的保存”按钮,不自动放宽服务器保护。
- 原生动作、证据读取、路径解析、进程查询和重启核对都放在非 GUI 线程;GUI 只更新视图和提交用户意图。
- 保留 VerifyRequest / VerifyResult 的现有参数及 safe_to_continue 兼容性,只追加有默认值的事实字段。
- 保持原有文件行尾,不进行与功能无关的整文件规范化。
- 开发验证只使用临时数据、模拟原生边界或隔离副本;不在生产库制造保存中断。
## 4. 保存、关闭和释放分开记录
新增 `operation_lifecycle.py`,提供纯函数,不导入 Qt、COM 或管理模块。
`classify_save_state(markers, *, helper_exited, native_released, forced=False)` 返回:
| 字段 | 值 | 能证明什么 |
| --- | --- | --- |
| save_state | NOT_STARTED / COMMITTED / REJECTED / UNKNOWN | COMMITTED 只来自本次 helper 的 SAVE_OK;缺标记不能推断“没有写入” |
| close_state | CLOSED / FAILED / UNKNOWN | 只由本次 CLOSE_OK / CLOSE_FAILED 明确记录;退出码不替代关闭证据 |
| helper_state | EXITED / RUNNING / UNKNOWN | 来自精确进程句柄或身份核对;PID 消失不证明保存完整 |
| resource_state | RELEASED / RETAINED / UNKNOWN | 来自独立资源证明;与保存成功分开 |
| interrupted | bool | 记录异常结束,不用进程消失掩盖中断 |
| detail | str | 说明证据缺失、矛盾或中断的具体原因;正常结果为空字符串 |
SAVE_OK 表示原生 Save 返回成功,不表示工程内容、中心库权威状态或历史版本已验收。关闭失败时仍保留已取得的保存事实。证据读取失败保留 UNKNOWN,不自动将 save_state 改为 REJECTED。
运行中的事实只能由当前操作 nonce、helper 身份和新日志更新;不拼接上轮 save.log。发生矛盾时采用 UNKNOWN 并保留原始标记。
新增兼容字段 `VerifyResult.write_outcome: dict | None = None`、`VerifyResult.session_record_path: Path | None = None`;现有 release_proof 含义不变。
safe_to_continue 继续表达本任务是否已释放后续原生执行所需资源,不能表达保存已通过。若保存结果未知但资源已独立证明释放,结果保留失败/待核对,允许其他独立项;同一目标不自动重放。若当前仍有原生写入,只保留相关库的写入限制,不全局锁住应用。
## 5. 原生调用阶段与停止语义
原生调用使用以下最小阶段:PRECALL、NATIVE_CALL、POSTCALL、CLOSING、RELEASE_CHECK、FINISHED、UNCONFIRMED。保存 helper 启动但本轮阶段缺失时视为可能进入 NATIVE_CALL,不能据此终止。
| 当前事实 | 用户停止/超时后的行为 | 后续项 |
| --- | --- | --- |
| 原生调用尚未发出且身份已确认 | 取消启动,不再进入新的原生动作;正常释放已创建的辅助资源 | 用户取消的批次不再启动后续项 |
| Save/Close/管理 COM 写调用可能仍在进行 | 写停止意图,继续观察调用返回;不杀 helper、worker、PCB 或服务器 | 同库写入等待;UI 和独立库继续可用 |
| Save 已返回,文档未关闭 | 记录 COMMITTED 或 REJECTED;仅对身份一致、已确认可正常关闭的文档执行现有正常收尾 | 无新 Save,不自动放弃未保存修改 |
| 原生调用已返回,资源已释放,但单项失败 | 保存失败及证据,保持同目标待核对,不自动重放 | 非用户取消批次可继续独立项 |
| 无法确认调用是否返回或身份变化 | 留存原始证据,转后台观察/人工处理;不猜测已退出或强制结束 | 只限制无法确认的相关库写入 |
5 秒用于普通合作式停止的短等待。进入原生写入后的前台任务最多再等待 120 秒;该时长只是转后台观察的界限,不是终止进程的时限。后台观察使用独立工作线程/任务,不占 GUI 线程,不占 NativeQueue 的全局 action 票据。
必须在后台观察交接完成、关联库租约及会话记录保存后,才能释放原执行票据。若交接存储失败,保留内存监督并提示记录错误;不得因此调用 kill,也不得声称释放完成。
同库后台写入的独占由现有 LibraryLock / retained lease 约束,其他库仍通过既有串行通道运行。不能解除另一个活跃写入者的文件锁;合作锁也不能宣称约束了别的工程师直接打开 Mentor。
观察循环优先使用已打开进程句柄的零等待查询,不每秒重跑整库扫描或完整 WMI。等待时不得持有 `_guard` 或全局原生互斥;只有候选会话退出后才尝试独立释放证明。证明通道正忙则延后观察,不将其当作新的库故障,不挤占其他库的执行机会。
用户取消批次后不自动恢复该批次。后台释放确认可以解除匹配会话状态,但后续写入仍来自已经排队的独立请求或用户新的操作,不能重放历史写入。
## 6. 保存 helper 与所有退出路径一致
- `_stop_worker`、`DesktopWorker.kill`、`stop_automation_clients`、`close`、verify watchdog 和失败 finally 使用同一阶段判据。不能只修用户停止按钮。
- 终止许可必须是已证明没有原生写调用的精确辅助进程;缺阶段、缺身份、helper 日志不可读都不是终止许可。
- 已启动保存时,worker 不再仅因取消文件出现就放弃监督。先等当前 COM 返回,再决定正常关闭或保留原生提示。
- VBS 在首次 Save 前检查取消。Save 返回后写 SAVE_OK / SAVE_REJECTED,再进行只读后置核对及有条件正常收尾;不让取消检查丢掉已经返回的保存事实。
- VBS 对 Save/IsSaved/后置同步/Close 分别记录阶段和明确失败原因。接触这些调用的错误处理必须是局部检查,不用一个覆盖整份脚本的 On Error Resume Next 掩盖失败。
- Close(True) 若仍可能写入,同样属于受保护原生调用;不得因 Save 已成功就强杀正在 Close 的 helper。
- 父进程意外退出与用户停止采用相同的“停止新动作、观察现有调用”原则。现有 `stop_clients=True` 不能绕过保存保护。
- 本轮不提供自动强制放弃保存。已有人工释放入口若遇到本工具明确登记的活跃 Save/Close,要显示实际阶段,不能把它当普通僵死客户端。
## 7. 可跨重启核对的会话记录
新增 `session_record.py`,只处理本工具受控的库外任务记录。它不执行 UI 操作,不从记录读取命令再运行,不依据文件存在就判占用。
每个会话记录:protocol/version、operation_id、operation、library、block、原生阶段、supervisor/worker/helper/native 的 PID/创建时间/完整映像、当前用户 SID/Windows 会话、desktop、取消文件、证据位置、原生调用是否可能仍在进行、保存/关闭/释放事实、记录时间和观察结果。
有可能新建原生进程时,先记录启动意图;CreateProcess 后记录精确身份。监督器打开工作进程的句柄后,工作进程才能获准进入原生动作;避免 GUI 在首条记录之前消失的空窗。
每个操作目录的 `session.json` 由启动方记录不可变身份;`worker-state.json` 只由 worker 写,`supervisor-lost.json` 只由 watchdog 写,后台观察写自己的 `observation-<observer_id>.json`。不能让多个进程覆盖同一状态文件。保存 helper 增加启动门:Python 写入 helper 精确身份后才放行进入 COM;取消、启动方消失或 30 秒内没有放行时,在 COM 调用前退出。旧测试直接运行 VBS 的无门参数模式只作兼容测试,生产启动必须使用门。
不按桌面名前缀批量删除桌面,不按旧 PID 杀进程。重启时:
1. 后台读取受控目录内受认可协议的记录,限制大小并拒绝目录链接和越界证据路径。
2. 精确核对当前进程创建时间、映像和用户/Windows 会话;PID 复用记为 IDENTITY_CHANGED,不对替代进程做动作。
3. 身份一致且仍活跃的会话显示原生阶段、输出和已有保存/取消选项;不自动继续写操作。
4. 进程确已退出后仍调用现有资源释放证明;确认后归档会话记录并刷新相关视图。
5. 未知记录只显示无法确认的原因和重新检查入口,不因为一份损坏 JSON 永久禁止所有操作。
CloseDesktop 只关闭本工具持有的句柄;原生进程或窗口仍引用桌面时,不宣称桌面已删除。不得把“清理桌面”作为结束正在写库进程的借口。
## 8. 管理 worker 的监督与取消
复用 `_require_add_supervisor` 的阶段检查思路,为 catalog/add/remove 都设置 supervisor identity、统一取消文件和阶段记录,而不是在管理入口外加一个无限等待。
- PRECALL:监督器失联或取消后,不再启动 LM、导入器、移除命令或下一阶段。
- NATIVE_CALL:watchdog 只记录失联和停止意图;COM 或外部 CleanWGLib 调用仍在进行时不强制退出其调用进程。
- POSTCALL:不得因为上一调用返回成功就继续 update 的下一次 add 或其他后续写入。关闭已确认空闲的自有原生会话,未知弹窗保留并暴露。
- update 的 remove 成功而 add 未提交时返回“部分完成/需恢复”,保留该目标检查点,不把它报成纯取消,也不重新导入。
- 正常完成和超时记录都能在重启后独立核对;不再仅依赖 `_retained_workers` 内存列表和 worker 自己写出的单个 safe_to_continue。
watchdog 不能解决不可取消 COM 调用的所有平台限制。验收要求是不再有无记录、无观察、可继续执行新写入的孤儿自动化;不承诺任意卡死的原生进程都能自动安全退出。
## 9. 状态读取的默认边界
推荐默认:普通读取不为 legacy 图标探测自动启动 LM,也不自动应答原生弹窗。
优先使用当前权威状态桥;桥不可用或无可观察的既有 LM 时,条目返回现有 UNKNOWN / UNAVAILABLE 及具体原因。无法查询不能等同于已知 PENDING,更不能等同于 ICDB 损坏。
`scan_open_library_status` 可以只观察用户已经打开、库身份匹配的 LM 树,不点任何按钮,不替用户关闭该 LM。发现模态提示时返回无法确认并保留可查看的提示信息。
`read_library_flag` / `read_library_tree_image_map` 的兼容 API 也必须服从上述边界,不能让隐藏直接调用继续启动生产库 LM 并走通用肯定按钮策略。需要自动创建 LM 的旧诊断能力不属于本轮交付;明确的用户“打开原生工具”入口保持原用途。
不能为了禁止状态读取自动点击,粗暴更改所有验证流程的已验证弹窗白名单。按调用目的分开,读取路径绝不调用肯定优先/fallback 操作器。
## 10. 批次与停止信号
GUI、CLI run_batch、ICDB 后台检查将 stop 传入同一请求链。兼容 auto_repair 分支补齐 stop 传播,但默认仍为关闭,不能把兼容性测试当作重新启用修复的授权。
队列取消、当前调用待返回、当前项失败和资源未释放分别记录。不能通过给全局 stop Event 置位,把一个库的观察状态扩散为所有独立库都停止。
本轮不把 Condition 改成无超时等待,除非所有取消/ready 变化都已有可靠通知;避免修复停止时引入永久等待。
## 11. 测试与验收
### 必须通过的跨组件场景
| 场景 | 必须观察到的结果 |
| --- | --- |
| Save 前取消 | Save 未调用;无新原生写入 |
| Save 调用中取消、超时、未知弹窗或父进程退出 | helper 存活、没有强杀调用;停止意图与当前阶段已记录 |
| Save 返回成功、Close 卡住 | 保存为 COMMITTED,关闭未确认;不得报无写入或全部完成 |
| helper 突然退出且没有返回标记 | 保存 UNKNOWN,原生释放另行核对;不能因进程消失报安全保存 |
| 前台等待超过 120 秒 | 转后台观察;UI 和独立库任务可运行,同库写入不能并发 |
| 管理 GUI 在 catalog/add/remove/update 两阶段之间退出 | 无新动作越过停止边界;已发生动作、待核对范围和精确会话身份可恢复显示 |
| 重启后 PID 复用、记录损坏或报告保存失败 | 不杀替代进程,不重放操作,不制造全局永久锁 |
| 权威桥不可用、LM 没打开 | 状态无法确认;没有 CreateProcess、恢复/覆盖/肯定点击 |
| 既有 LM 出现未知/修复/覆盖/保存提示 | 只观察并返回 UNKNOWN,不点击、不关闭用户会话 |
使用真实 cscript + 替换外部 COM 创建的 VBS harness 验证保存顺序;使用临时 sleep/helper 进程验证父进程退出、超时和进程身份;不在生产 Mentor 中途保存时做破坏试验。
全量 Python 测试必须串行运行,测试互斥使用独立命名空间,恢复临时根与源库分离;编译检查、diff 检查和 frozen helper 入口冒烟完成后才描述候选程序。冻结包先在 fake 原生边界跑同一保存/停止场景。
真实 Mentor 仅在一致隔离副本做延迟保存、当前原生取消、正常收尾、重新打开与权威状态核对;Win10、网络共享库、跨机器并发分别记录,缺环境时保留未验收项。
## 12. 分阶段交付和停止条件
交付单元 A:保存事实、helper/worker 停止规则和有界等待交接。
交付单元 B:管理 worker 的失联、取消、持久记录和重启核对。
交付单元 C:legacy 只读边界、GUI/CLI 停止接入、冻结包与隔离环境验收。
每个单元有独立回归和提交;三单元完成前不以局部通过宣称生命周期修复完成。原有 #10–#12 的旧评论不视为本轮新实现的批准。
实现从当前集成基线另开临时分支 `codex/runtime-lifecycle-safety`,不得覆盖原运行工作区,不 force-push 历史分支。执行时重新读取远端 PR/main 状态,按实际基线选择差分,避免重复合入已合并改动。先交付一个仅包含本轮运行时差分的草稿 PR,并在描述中按 A/B/C 给出审查顺序。
出现以下任一情况停止扩展并回到设计核对:没有办法在保留 COM 调用的同时交接观察、需要自动终止真实服务器、需要删除原生锁或地址才能让测试通过、需要引入新全局门禁才能规避身份不明。
## 13. 本轮明确排除
一般化跨机器写入互斥、全库/非目标工程内容证明、PRJ 重写与 before 字节补齐、UIA 无界线程整改、桥部署预检、CLI 全库范围确认、CI/打包外部路径/README 整理均单独排期。
其中与本轮直接相关的未知进程身份核对、停止传播及 Save/Close 事实拆分已纳入;不把排除项写成已修复。
## 14. 草案自查
- 三条 P1 均有明确生产调用点、拟修改边界和跨组件验收场景。
- 没有把取消等同于强制结束,也没有把退出等同于保存成功。
- 没有通过新增历史记录硬门禁解决问题;活动写入仍保持真实资源约束。
- 没有依赖“调用 watchdog(stop_clients=False) 就万事解决”的错误假设。
- 只读回退范围与权威桥路径区分;未知状态不伪造 READY/PENDING。
- 用户已通过“按照当前的优化吧”授权实施;真实环境验收仍单独记录,不以测试计数代替。
+1 -1
View File
@@ -3,7 +3,7 @@ requires = ["setuptools>=84.1.0"]
build-backend = "setuptools.build_meta"
[project]
name = "cbb-refresh-verify"
version = "0.2.3b1"
version = "0.2.4b2"
requires-python = ">=3.12,<3.13"
dependencies = ["PySide6>=6.7,<7", "pywin32>=306", "pywinauto>=0.6.9,<0.7", "Pillow>=13.0.0,<14"]
+1 -1
View File
@@ -1 +1 @@
__version__ = "0.2.3b1"
__version__ = "0.2.4b2"
+194 -151
View File
@@ -1,151 +1,194 @@
from __future__ import annotations
import argparse, os, sys
from pathlib import Path
from typing import Sequence
from .runtime_logging import configure
def configure_qt_application(app):
"""Apply a predictable light palette across machines and Qt themes."""
from PySide6.QtGui import QColor, QPalette
app.setStyle('Fusion')
palette = app.palette()
colors = {
QPalette.ColorRole.Window: '#f5f6f8',
QPalette.ColorRole.WindowText: '#202124',
QPalette.ColorRole.Base: '#ffffff',
QPalette.ColorRole.AlternateBase: '#eef1f5',
QPalette.ColorRole.Text: '#202124',
QPalette.ColorRole.Button: '#eef1f5',
QPalette.ColorRole.ButtonText: '#202124',
QPalette.ColorRole.Highlight: '#2f6fed',
QPalette.ColorRole.HighlightedText: '#ffffff',
QPalette.ColorRole.ToolTipBase: '#ffffff',
QPalette.ColorRole.ToolTipText: '#202124',
QPalette.ColorRole.PlaceholderText: '#6b7280',
}
for role, color in colors.items():
palette.setColor(role, QColor(color))
for role in (QPalette.ColorRole.WindowText, QPalette.ColorRole.Text, QPalette.ColorRole.ButtonText):
palette.setColor(QPalette.ColorGroup.Disabled, role, QColor('#858b94'))
app.setPalette(palette)
app.setStyleSheet(
'QWidget { color: #202124; }'
'QTableWidget, QLineEdit, QTextEdit, QSpinBox { '
'background: #ffffff; color: #202124; selection-color: #202124; }'
'QHeaderView::section { background: #e9edf2; color: #202124; }'
'QPushButton, QToolButton { color: #202124; }'
'QPushButton:disabled, QToolButton:disabled { color: #858b94; background: #e3e6eb; }'
)
return app
def resolve_data_root(cli_value: Path | None) -> Path:
if cli_value is not None: return cli_value.expanduser().resolve()
if getattr(sys, "frozen", False): return Path(sys.executable).resolve().parent / "data"
local = os.environ.get("LOCALAPPDATA")
return (Path(local) / "CbbRefreshVerify" if local else Path.home() / "AppData" / "Local" / "CbbRefreshVerify").resolve()
def main(argv: Sequence[str] | None = None) -> int:
parser = argparse.ArgumentParser(prog="CbbRefreshVerify")
parser.add_argument('--verify-worker', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--repair-worker', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--owner-probe', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--native-backup-worker', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--management-worker', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--dialog-agent', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--collect-failure-logs', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--batch-lmc', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--block', action='append', default=[], help=argparse.SUPPRESS)
parser.add_argument('--libspec', default='', help=argparse.SUPPRESS)
parser.add_argument('--dms-login-config', default='', help=argparse.SUPPRESS)
parser.add_argument('--mentor-bin', default='', help=argparse.SUPPRESS)
parser.add_argument("--smoke-test", action="store_true", help=argparse.SUPPRESS); parser.add_argument("--data-root", type=Path, help=argparse.SUPPRESS)
args = parser.parse_args(list(argv) if argv is not None else None)
if args.owner_probe:
from .owner_probe import owner_probe_main
return owner_probe_main(args.owner_probe)
if args.native_backup_worker:
from .native_backup import native_backup_worker_main
return native_backup_worker_main(args.native_backup_worker)
if args.management_worker:
from .native_management import management_worker_main
return management_worker_main(args.management_worker)
if args.repair_worker:
from .native_repair_ui import repair_worker_main
return repair_worker_main(args.repair_worker)
if args.collect_failure_logs:
from .verify_diagnostics import failure_logs_main
return failure_logs_main(args.collect_failure_logs)
if args.dialog_agent:
from .dialog_agent import dialog_agent_main
return dialog_agent_main(args.dialog_agent)
if args.verify_worker:
from .mentor_adapter import worker_main
import contextlib
with args.verify_worker.with_name('worker.log').open('a', encoding='utf-8') as log:
with contextlib.redirect_stdout(log), contextlib.redirect_stderr(log):
return worker_main(args.verify_worker)
context = configure(resolve_data_root(args.data_root)); logger = context.logger; logger.info("application_start")
try:
if args.batch_lmc:
from .verify_runner import (VerifyRequest, native_lm_verify_enabled,
run_batch, write_batch_report, failure_report_path)
from .library_reader import read_library_fast
from .batch_status import authoritative_status_map
from .network_path import LibraryPathLease, select_library_file
with LibraryPathLease.acquire(args.batch_lmc) as lease:
mentor_path = select_library_file(lease.mentor_path)
logger.info(
'library_path_lease display=%s mentor=%s drive=%s remote=%s created=%s',
lease.display_path, mentor_path, lease.drive or '-',
lease.remote_root or '-', lease.created)
names = list(args.block)
if not names:
library = read_library_fast(
mentor_path,
status_reader=lambda path, names: authoritative_status_map(
names, path, context.root/'status-scan.jsonl'),
)
if library.error: raise RuntimeError(library.error)
names = [record.name for record in library.records if record.status != 'READY']
if not names: raise RuntimeError('中心库未读取到需要校验的 CBB')
requests = [VerifyRequest(mentor_path, name, context.root/'evidence',
libspec=args.libspec, dms_login_config=args.dms_login_config,
mentor_bin=args.mentor_bin, logging_context=context,
mapped_drive=lease.drive,
mapped_remote_root=lease.remote_root,
native_lm_verify=native_lm_verify_enabled())
for name in names]
try:
results = run_batch(requests)
except Exception:
lease.retain_for_recovery()
raise
if any(not result.safe_to_continue for result in results):
lease.retain_for_recovery()
report = write_batch_report(requests, results, context.root/'reports')
logger.info('batch_report path=%s', report)
logger.info('failure_report path=%s', failure_report_path(report))
code = 0 if all(result.status == 'PASS' for result in results) else 2
logger.info('application_clean_shutdown exit_code=%s', code)
return code
from .ui.main_window import MainWindow, application_icon_path
if args.smoke_test:
os.environ.setdefault('QT_QPA_PLATFORM', 'offscreen')
from PySide6.QtWidgets import QApplication
app = QApplication.instance() or QApplication([sys.executable])
configure_qt_application(app)
window = MainWindow(logging_context=context); window.show()
icon_path = application_icon_path()
if icon_path.is_file():
from PySide6.QtGui import QIcon
app.setWindowIcon(QIcon(str(icon_path)))
if args.smoke_test:
from PySide6.QtCore import QTimer
logger.info('gui_smoke_window_created title=%s', window.windowTitle())
QTimer.singleShot(250, app.quit)
code = int(app.exec()); logger.info("application_clean_shutdown exit_code=%s", code); return code
except Exception: logger.exception("application_startup_failure"); return 1
finally:
for handler in logger.handlers: handler.flush()
from __future__ import annotations
import argparse, os, sys
from contextlib import contextmanager
from pathlib import Path
from typing import Sequence
from .runtime_logging import configure
@contextmanager
def batch_stop_signals(stop):
import signal
previous = {}
try:
for name in ('SIGINT', 'SIGTERM', 'SIGBREAK'):
signum = getattr(signal, name, None)
if signum is not None:
previous[signum] = signal.signal(signum, lambda *_: stop.set())
yield
finally:
for signum, handler in previous.items():
signal.signal(signum, handler)
def configure_qt_application(app):
"""Apply a predictable light palette across machines and Qt themes."""
from PySide6.QtGui import QColor, QPalette
app.setStyle('Fusion')
palette = app.palette()
colors = {
QPalette.ColorRole.Window: '#f5f6f8',
QPalette.ColorRole.WindowText: '#202124',
QPalette.ColorRole.Base: '#ffffff',
QPalette.ColorRole.AlternateBase: '#eef1f5',
QPalette.ColorRole.Text: '#202124',
QPalette.ColorRole.Button: '#eef1f5',
QPalette.ColorRole.ButtonText: '#202124',
QPalette.ColorRole.Highlight: '#2f6fed',
QPalette.ColorRole.HighlightedText: '#ffffff',
QPalette.ColorRole.ToolTipBase: '#ffffff',
QPalette.ColorRole.ToolTipText: '#202124',
QPalette.ColorRole.PlaceholderText: '#6b7280',
}
for role, color in colors.items():
palette.setColor(role, QColor(color))
for role in (QPalette.ColorRole.WindowText, QPalette.ColorRole.Text, QPalette.ColorRole.ButtonText):
palette.setColor(QPalette.ColorGroup.Disabled, role, QColor('#858b94'))
app.setPalette(palette)
app.setStyleSheet(
'QWidget { color: #202124; }'
'QTableWidget, QLineEdit, QTextEdit, QSpinBox { '
'background: #ffffff; color: #202124; selection-color: #202124; }'
'QHeaderView::section { background: #e9edf2; color: #202124; }'
'QPushButton, QToolButton { color: #202124; }'
'QPushButton:disabled, QToolButton:disabled { color: #858b94; background: #e3e6eb; }'
)
return app
def resolve_data_root(cli_value: Path | None) -> Path:
if cli_value is not None: return cli_value.expanduser().resolve()
if getattr(sys, "frozen", False): return Path(sys.executable).resolve().parent / "data"
local = os.environ.get("LOCALAPPDATA")
return (Path(local) / "CbbRefreshVerify" if local else Path.home() / "AppData" / "Local" / "CbbRefreshVerify").resolve()
def _positive_int(value):
try:
number = int(value)
except ValueError:
raise argparse.ArgumentTypeError('must be a positive integer') from None
if number <= 0:
raise argparse.ArgumentTypeError('must be a positive integer')
return number
def main(argv: Sequence[str] | None = None) -> int:
parser = argparse.ArgumentParser(prog="CbbRefreshVerify")
parser.add_argument('--verify-worker', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--repair-worker', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--owner-probe', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--native-backup-worker', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--management-worker', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--dialog-agent', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--collect-failure-logs', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--batch-lmc', type=Path, help=argparse.SUPPRESS)
parser.add_argument('--block', action='append', default=[], help=argparse.SUPPRESS)
parser.add_argument('--limit', type=_positive_int, help='Maximum requested CBB names for --batch-lmc (default: all)')
parser.add_argument('--libspec', default='', help=argparse.SUPPRESS)
parser.add_argument('--dms-login-config', default='', help=argparse.SUPPRESS)
parser.add_argument('--mentor-bin', default='', help=argparse.SUPPRESS)
parser.add_argument("--smoke-test", action="store_true", help=argparse.SUPPRESS); parser.add_argument("--data-root", type=Path, help=argparse.SUPPRESS)
args = parser.parse_args(list(argv) if argv is not None else None)
if args.owner_probe:
from .owner_probe import owner_probe_main
return owner_probe_main(args.owner_probe)
if args.native_backup_worker:
from .native_backup import native_backup_worker_main
return native_backup_worker_main(args.native_backup_worker)
if args.management_worker:
from .native_management import management_worker_main
return management_worker_main(args.management_worker)
if args.repair_worker:
from .native_repair_ui import repair_worker_main
return repair_worker_main(args.repair_worker)
if args.collect_failure_logs:
from .verify_diagnostics import failure_logs_main
return failure_logs_main(args.collect_failure_logs)
if args.dialog_agent:
from .dialog_agent import dialog_agent_main
return dialog_agent_main(args.dialog_agent)
if args.verify_worker:
from .mentor_adapter import worker_main
import contextlib
with args.verify_worker.with_name('worker.log').open('a', encoding='utf-8') as log:
with contextlib.redirect_stdout(log), contextlib.redirect_stderr(log):
return worker_main(args.verify_worker)
context = configure(resolve_data_root(args.data_root)); logger = context.logger; logger.info("application_start")
try:
if args.batch_lmc:
from .verify_runner import (VerifyRequest, native_lm_verify_enabled,
run_batch, write_batch_report, failure_report_path)
from .library_reader import read_library_fast
from .batch_status import authoritative_status_map
from .network_path import LibraryPathLease, select_library_file
with LibraryPathLease.acquire(args.batch_lmc) as lease:
mentor_path = select_library_file(lease.mentor_path)
logger.info(
'library_path_lease display=%s mentor=%s drive=%s remote=%s created=%s',
lease.display_path, mentor_path, lease.drive or '-',
lease.remote_root or '-', lease.created)
names = list(args.block)
if not names:
library = read_library_fast(
mentor_path,
status_reader=lambda path, names: authoritative_status_map(
names, path, context.root/'status-scan.jsonl'),
)
if library.error: raise RuntimeError(library.error)
names = [record.name for record in library.records if record.status != 'READY']
if not names: raise RuntimeError('中心库未读取到需要校验的 CBB')
total = len(names)
if args.limit is not None:
names = names[:args.limit]
logger.info('batch_selection selected=%s total=%s remaining=%s limit=%s',
len(names), total, total - len(names), args.limit or 'all')
requests = [VerifyRequest(mentor_path, name, context.root/'evidence',
libspec=args.libspec, dms_login_config=args.dms_login_config,
mentor_bin=args.mentor_bin, logging_context=context,
mapped_drive=lease.drive,
mapped_remote_root=lease.remote_root,
native_lm_verify=native_lm_verify_enabled())
for name in names]
def retain_mapping(reason):
lease.retain_for_recovery()
logger.warning(
'batch_drive_retained drive=%s remote=%s display=%s mentor=%s reason=%s '
'guidance=请打开此中心库的“释放残留会话…”入口,确认原生会话安全释放后再释放映射;'
'原生写入仍可能运行时请保留映射。',
lease.drive or '-', lease.remote_root or '-', lease.display_path,
mentor_path, reason)
try:
from threading import Event
stop = Event()
with batch_stop_signals(stop):
results = run_batch(requests, stop=stop)
except Exception:
retain_mapping('batch_exception')
raise
if any(not result.safe_to_continue for result in results):
retain_mapping('native_release_unconfirmed')
report = write_batch_report(requests, results, context.root/'reports')
logger.info('batch_report path=%s', report)
logger.info('failure_report path=%s', failure_report_path(report))
code = 0 if all(result.status == 'PASS' for result in results) else 2
logger.info('application_clean_shutdown exit_code=%s', code)
return code
from .ui.main_window import MainWindow, application_icon_path
if args.smoke_test:
os.environ.setdefault('QT_QPA_PLATFORM', 'offscreen')
from PySide6.QtWidgets import QApplication
app = QApplication.instance() or QApplication([sys.executable])
configure_qt_application(app)
window = MainWindow(logging_context=context); window.show()
icon_path = application_icon_path()
if icon_path.is_file():
from PySide6.QtGui import QIcon
app.setWindowIcon(QIcon(str(icon_path)))
if args.smoke_test:
from PySide6.QtCore import QTimer
logger.info('gui_smoke_window_created title=%s', window.windowTitle())
QTimer.singleShot(250, app.quit)
code = int(app.exec()); logger.info("application_clean_shutdown exit_code=%s", code); return code
except Exception: logger.exception("application_startup_failure"); return 1
finally:
for handler in logger.handlers: handler.flush()
Binary file not shown.

Before

Width:  |  Height:  |  Size: 264 KiB

After

Width:  |  Height:  |  Size: 120 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 4.8 KiB

After

Width:  |  Height:  |  Size: 1.1 MiB

@@ -1,7 +1 @@
<svg xmlns="http://www.w3.org/2000/svg" width="64" height="64" viewBox="0 0 64 64">
<rect x="3" y="3" width="58" height="58" rx="10" fill="#0f766e"/>
<rect x="11" y="12" width="42" height="32" rx="4" fill="#f8fafc"/>
<rect x="15" y="16" width="34" height="24" rx="2" fill="#dbeafe"/>
<path d="M20 29l6 6 14-14" fill="none" stroke="#16a34a" stroke-width="4" stroke-linecap="round" stroke-linejoin="round"/>
<rect x="25" y="47" width="14" height="4" rx="2" fill="#f8fafc"/>
</svg>
<svg xmlns="http://www.w3.org/2000/svg" width="1024" height="1024" viewBox="0 0 1024 1024"><image width="1024" height="1024" href="data:image/png;base64,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 truncated

Before

Width:  |  Height:  |  Size: 489 B

After

Width:  |  Height:  |  Size: 1.5 MiB

+61 -15
View File
@@ -29,14 +29,55 @@ def _paths(root):
def _idle_check(library, project, pcb):
from .mentor_adapter import pcb_processes, _process_details
from .icdb_ownership import unreleased_addresses
if pcb_processes() or any(_process_details(name) for name in (
'LibraryManager.exe', 'viewdraw.exe')):
raise ValueError('请先正常关闭 PCB、Library Manager 和原理图编辑器,再恢复备份。')
from .mentor_adapter import _process_details
from .icdb_ownership import database_directories, unreleased_addresses
from contextlib import ExitStack
import win32con, win32file
directories = database_directories(library, project, pcb)
scope = (Path(project).parent, *directories)
for name in ('ExpeditionPCB.exe', 'LibraryManager.exe', 'viewdraw.exe',
'iCDBProjectBackup.exe', 'ReuseBlockImporter.exe'):
for pid, executable, command, parent in _process_details(name):
match = _native_scope_match(command, roots=scope, files=(library, project, pcb))
if match is True:
raise ValueError(f'TARGET_NATIVE_OWNER_ACTIVE: 目标工程或共享库有原生会话,PID={pid}')
# Pathless clients do not identify this target. The address and
# exclusive file probes below still reject actual target occupancy.
addresses = unreleased_addresses(library, project, pcb)
if addresses:
raise ValueError(f'数据库会话尚未释放,停止恢复:{addresses}')
raise ValueError(f'TARGET_DATABASE_ADDRESS_UNCONFIRMED: 地址记录需原生确认;未据此判定正在占用:{addresses}')
with ExitStack() as stack:
paths = [Path(library), Path(pcb).parent/'Layout/LayoutDB.lyt',
*(directory/'icdb.dat' for directory in directories)]
for path in dict.fromkeys(paths):
if path.is_file():
handle = win32file.CreateFile(str(path), win32con.GENERIC_READ, 0, None,
win32con.OPEN_EXISTING, 0, None)
stack.callback(handle.Close)
def _native_scope_match(command, *, roots=(), files=()):
"""Match complete native path arguments; a missing identity remains unknown."""
from .failed_session import _command_args
if not command:
return None
expected_files = {Path(path).resolve() for path in files}
expected_roots = tuple(Path(path).resolve() for path in roots)
identities = []
for arg in _command_args(command)[1:]:
value = arg.split('=', 1)[-1].strip('"')
path = Path(value)
if not path.is_absolute():
continue
key = arg.split('=', 1)[0].lstrip('-/').casefold() if '=' in arg else ''
if (path.suffix.casefold() in ('.prj', '.pcb', '.lmc')
or path.name.casefold() in ('database', '__reusableblockssymb', 'icdb.dat')
or key in ('project', 'icdb', 'database', 'db', 'lmc', 'prj', 'pcb')):
identities.append(path.resolve())
if not identities:
return None
return any(path in expected_files or any(path == root or path.is_relative_to(root)
for root in expected_roots) for path in identities)
def _project_text(path):
@@ -69,14 +110,15 @@ def _source_check(source, name, original):
if not pcb.is_relative_to(source.parent.resolve()):
raise ValueError('备份 PCB 引用越出工程目录。')
if not all(path.is_file() for path in (pcb, pcb.parent/'Layout/LayoutDB.lyt',
pcb.parent/'pcb.prj', source.parent/'database/icdb.dat')):
raise ValueError('备份不完整:需要原理图数据库、PCB、布局数据库和 PCB 工程文件。')
source.parent/'database/icdb.dat')):
raise ValueError('备份不完整:需要原理图数据库、PCB 和布局数据库。')
project_refs = re.findall(r'^KEY ProjectFile "([^"]+)"', _project_text(pcb), re.M)
for value in project_refs:
require_relative(value)
if len(project_refs) != 1 or (pcb.parent/project_refs[0]).resolve() != source.resolve():
raise ValueError('备份 PCB 必须引用所选工程,不能引用其他位置的工程。')
for project in (source, pcb.parent/'pcb.prj'):
from .maintenance_entry import maintenance_project_files
for project in maintenance_project_files(source, pcb):
data = _project_text(project)
for value in re.findall(r'^KEY iCDBDir "([^"]+)"', data, re.M):
require_relative(value)
@@ -100,7 +142,7 @@ def _align_library(project, pcb, library):
ensure_project_central_library(path, library)
def recover_from_backup(request, source, *, progress=None, verifier=None, idle_check=None):
def recover_from_backup(request, source, *, progress=None, verifier=None, idle_check=None, version_prompt=None):
"""Validate in copies before changing Work; preserve all recovery evidence."""
import pythoncom, win32con, win32event
from .network_path import absolute_preserving_drive
@@ -140,9 +182,12 @@ def recover_from_backup(request, source, *, progress=None, verifier=None, idle_c
return report
def save(result):
record.update({'installed': installed, 'result': asdict(result)})
root.mkdir(parents=True, exist_ok=True)
report = root/'recovery-result.json'
report.write_text(json.dumps(record, ensure_ascii=False, default=str, indent=2), encoding='utf-8')
try:
root.mkdir(parents=True, exist_ok=True)
report.write_text(json.dumps(record, ensure_ascii=False, default=str, indent=2), encoding='utf-8')
except OSError as error:
return replace(result, stderr=(result.stderr+'\n恢复报告写入失败:'+str(error)).strip())
return replace(result, evidence_path=report)
from .com_runtime import initialize_com, uninitialize_com
initialized = initialize_com()
@@ -196,7 +241,8 @@ def recover_from_backup(request, source, *, progress=None, verifier=None, idle_c
preflight(validation)
emit('validate_source', '正在副本中打开备份工程并执行原生校验…')
release_confirmed = False
source_result = verifier(validation, progress=progress, flag_reader=read_library_flag_if_available)
choice_options = {'version_prompt': version_prompt} if version_prompt is not None else {}
source_result = verifier(validation, progress=progress, flag_reader=read_library_flag_if_available, **choice_options)
release_confirmed = source_result.safe_to_continue
record['source_validation'] = asdict(source_result)
if (source_result.status != 'PASS' or source_result.library_flag != 'GREEN'
@@ -230,7 +276,7 @@ def recover_from_backup(request, source, *, progress=None, verifier=None, idle_c
release_confirmed = False
final = verifier(replace(request, lmc_file=library, evidence_dir=root/'target-verify',
auto_repair=False, native_lm_verify=False, interactive_input=False),
progress=progress, flag_reader=read_library_flag_if_available)
progress=progress, flag_reader=read_library_flag_if_available, **choice_options)
release_confirmed = final.safe_to_continue
if final.status == 'PASS' and (final.library_flag != 'GREEN' or not final.safe_to_continue):
final = replace(final, status='FAIL', stderr='目标工程恢复后的 Ready 状态未确认。')
@@ -239,7 +285,7 @@ def recover_from_backup(request, source, *, progress=None, verifier=None, idle_c
return save(final)
except Exception as error:
return save(VerifyResult('FAIL' if installed else 'BLOCKED', stderr=str(error),
safe_to_continue=False))
safe_to_continue=release_confirmed))
finally:
try:
if library_lock is not None:
+22 -9
View File
@@ -48,6 +48,7 @@ class BatchStatusRecord:
detail: str = ""
fingerprint: str = ""
timestamp: str = ""
block_type: str = "RULP"
@property
def completed(self) -> bool:
@@ -106,9 +107,11 @@ def fingerprint_library(lmc: Path) -> LmcFingerprint:
# Metadata alone is not a safe checkpoint key: on network shares and on
# file systems with coarse timestamp resolution, two different LMC
# contents can have the same size and mtime. Include the actual file
# digest so a changed central library can never reuse an old status row.
# digest to invalidate changed LMC bytes; payload edits remain outside this key.
content_sha256 = _sha256_file(path)
combined = hashlib.sha256()
combined.update(str(path).casefold().encode('utf-8', errors='surrogatepass'))
combined.update(b'\0')
combined.update(content_sha256.encode("ascii"))
combined.update(b"\0" + str(stat.st_size).encode("ascii"))
combined.update(b"\0" + str(stat.st_mtime_ns).encode("ascii") + b"\0")
@@ -165,6 +168,7 @@ class BatchCheckpoint:
detail=str(payload.get("detail") or ""),
fingerprint=str(payload.get("fingerprint") or ""),
timestamp=str(payload.get("timestamp") or ""),
block_type=str(payload.get("block_type") or "RULP"),
)
return cls(records)
@@ -182,7 +186,12 @@ def _now() -> str:
class BatchStatusSession:
"""Supervised serial bridge session with checkpoint recovery."""
"""Supervised serial bridge session; fresh status queries by default.
Checkpoints record scan evidence. Explicit ``resume=True`` is a low-level
opt-in for controlled read-only scans, not a production freshness proof:
the lightweight fingerprint does not cover native payload/database edits.
"""
def __init__(self, bridge: StatusBridgeClient, *, timeout_seconds: float = 10.0,
rotate_every: int = 500) -> None:
@@ -192,7 +201,7 @@ class BatchStatusSession:
self.restart_count = 0
def scan(self, names: Iterable[str], lmc_file: Path, checkpoint_path: Path,
*, block_type: str = "RULP", resume: bool = True) -> list[BatchStatusRecord]:
*, block_type: str = "RULP", resume: bool = False) -> list[BatchStatusRecord]:
fingerprint = fingerprint_library(Path(lmc_file))
if resume:
checkpoint = BatchCheckpoint.load(checkpoint_path)
@@ -212,7 +221,7 @@ class BatchStatusSession:
for index, name in enumerate(names):
previous = checkpoint.get(name)
if (previous and previous.status in {'READY', 'PENDING'}
and previous.fingerprint == fingerprint.sha256):
and previous.fingerprint == fingerprint.sha256 and previous.block_type == block_type):
results.append(previous)
continue
try:
@@ -224,11 +233,12 @@ class BatchStatusSession:
block_name=name, status=result.status, source=result.source,
raw_status=result.raw_status, detail=result.detail,
fingerprint=fingerprint.sha256, timestamp=_now(),
block_type=block_type,
)
except BridgeUnavailable as error:
record = BatchStatusRecord(block_name=name, status="UNAVAILABLE",
detail=str(error), fingerprint=fingerprint.sha256,
timestamp=_now())
timestamp=_now(), block_type=block_type)
except BridgeResponseError as error:
message = str(error)
lower_message = message.casefold()
@@ -242,15 +252,15 @@ class BatchStatusSession:
else "UNAVAILABLE" if unavailable
else "MALFORMED" if malformed else "ERROR")
record = BatchStatusRecord(block_name=name, status=status, detail=message,
fingerprint=fingerprint.sha256, timestamp=_now())
if server is not None and status not in {"UNAVAILABLE", "MALFORMED"}:
fingerprint=fingerprint.sha256, timestamp=_now(), block_type=block_type)
if server is not None and status != "UNAVAILABLE":
try:
server.restart()
except BridgeUnavailable:
pass
except Exception as error: # bridge failures isolate one CBB
record = BatchStatusRecord(block_name=name, status="ERROR", detail=str(error),
fingerprint=fingerprint.sha256, timestamp=_now())
fingerprint=fingerprint.sha256, timestamp=_now(), block_type=block_type)
if server is not None:
try:
server.restart()
@@ -258,7 +268,7 @@ class BatchStatusSession:
pass
checkpoint.append(checkpoint_path, record)
results.append(record)
if record.status in {"UNAVAILABLE", "MALFORMED"} and index + 1 < len(names):
if record.status == "UNAVAILABLE" and index + 1 < len(names):
# Startup/protocol failures cannot be repaired by retrying
# the same request contract for every CBB.
# Preserve a terminal record for each remaining name so a
@@ -267,6 +277,7 @@ class BatchStatusSession:
unavailable_record = BatchStatusRecord(
block_name=remaining, status=record.status, detail=record.detail,
fingerprint=fingerprint.sha256, timestamp=_now(),
block_type=block_type,
)
checkpoint.append(checkpoint_path, unavailable_record)
results.append(unavailable_record)
@@ -285,6 +296,8 @@ def authoritative_status_map(names: Iterable[str], lmc_file: Path, checkpoint_pa
timeout_seconds: float = 10.0) -> dict[str, str]:
"""Return only bridge-backed statuses for an optional UI integration.
Production always queries fresh statuses, including after native writes;
saved READY/PENDING rows are never reused here.
A missing bridge is represented as ``UNAVAILABLE``/``UNKNOWN`` by callers;
it is never promoted to ``READY``.
"""
+3
View File
@@ -1,3 +1,6 @@
"""Shared execution defaults for GUI, batch requests and integration."""
DEFAULT_TIMEOUT_SECONDS = 600
DIALOG_AGENT_STARTUP_TIMEOUT_SECONDS = 30
WORKER_START_GATE_TIMEOUT_SECONDS = DIALOG_AGENT_STARTUP_TIMEOUT_SECONDS + 15
NATIVE_WRITE_STOP_WAIT_SECONDS = 120
+116 -28
View File
@@ -1,6 +1,8 @@
"""Launch automation on a private Windows desktop, never the input desktop."""
import subprocess
import os
import logging
from contextlib import contextmanager
from pathlib import Path
import threading
import win32con
@@ -12,46 +14,113 @@ import win32service
_automation_clients = []
_automation_clients_lock = threading.Lock()
_supervisor_lost = threading.Event()
_native_phases = {}
def launch_save_helper(command, **kwargs):
def _phase_pending(phase):
if phase not in ('PRECALL', 'POSTCALL', 'RELEASE_CHECK', 'FINISHED', 'NATIVE_CALL', 'CLOSING'):
return None
from .operation_lifecycle import stop_disposition
return stop_disposition(call_started=phase != 'PRECALL',
call_returned=phase in ('POSTCALL', 'RELEASE_CHECK', 'FINISHED'),
identity_confirmed=True) == 'WAIT_FOR_RETURN'
@contextmanager
def worker_phase(operation_id, phase, *, continuation=False):
with _automation_clients_lock:
existing = continuation and _native_phases.get(operation_id) == phase and _phase_pending(phase) is True
if continuation and not existing:
raise RuntimeError('WORKER_STATE_UNCONFIRMED: 尚无已派发的原生调用可继续登记')
if _supervisor_lost.is_set() and phase in ('NATIVE_CALL', 'CLOSING') and not existing:
raise RuntimeError('WORKER_CANCELLED: 监督器已退出,未提交新的原生调用')
_native_phases[operation_id] = phase
yield
def _record_write_pending(record_path, operation_id=None):
if record_path is None:
return None
from .session_record import read_session_record
path = Path(record_path).with_name('worker-state.json')
try:
state = read_session_record(path, evidence_root=path.parent)
if operation_id is not None and state['operation_id'] != operation_id:
return None
return _phase_pending(state.get('phase'))
except (OSError, ValueError):
return None
def launch_save_helper(command, *, start_gate=None, identity_path=None, operation_id=None, **kwargs):
"""Register the save client atomically with respect to supervisor loss."""
with _automation_clients_lock:
if _supervisor_lost.is_set():
raise RuntimeError('VERIFY_CANCELLED: 监督器已退出,未启动新的保存客户端')
helper = subprocess.Popen(command, **kwargs)
_automation_clients.append(helper)
if start_gate is not None:
from .session_record import process_identity, read_session_record, write_session_record
if identity_path is None or not operation_id:
raise ValueError('保存启动门缺少会话记录')
state = {'operation_id': operation_id, 'processes': []}
identity_path = Path(identity_path)
if identity_path.exists():
state = read_session_record(identity_path, evidence_root=identity_path.parent)
if state['operation_id'] != operation_id:
raise ValueError('保存记录不属于当前操作')
if helper.poll() is None:
state['processes'].append({'role': 'helper', **process_identity(helper.pid)})
state['phase'] = 'NATIVE_CALL'
write_session_record(identity_path, state)
Path(start_gate).touch()
return helper
def start_supervisor_watchdog(supervisor_pid, *, cancel_path=None, stop_clients=False):
"""A crashed supervisor must not leave unsupervised automation writing.
def start_supervisor_watchdog(supervisor_pid, *, cancel_path=None, stop_clients=False,
session_record_path=None, operation_id=None, supervisor_identity=None):
"""Record stop intent; a watchdog cannot interrupt an in-flight COM call.
Stop only this Python worker and its registered automation clients. Native
database owners remain alive, including a COM call already in progress.
stop_clients is retained for old callers, never permission to kill Save.
"""
import win32api
parent = win32api.OpenProcess(win32con.SYNCHRONIZE, False, int(supervisor_pid))
if supervisor_identity is not None:
from .session_record import process_identity, IDENTITY_KEYS
actual = process_identity(int(supervisor_pid))
if any(actual[k] != supervisor_identity.get(k) for k in IDENTITY_KEYS):
parent.Close()
raise RuntimeError('监督器进程身份已变化,未启动原生操作')
def watch():
try:
win32event.WaitForSingleObject(parent, win32event.INFINITE)
if cancel_path is not None:
# Let native calls finish before the worker closes its exact
# document. Killing a pending automation client can abandon it.
Path(cancel_path).touch()
if not stop_clients:
return
threading.Event().wait(5)
with _automation_clients_lock:
for helper in _automation_clients:
try:
if helper.poll() is None:
helper.terminate()
except OSError:
pass
_supervisor_lost.set()
if cancel_path is not None:
try:
Path(cancel_path).touch()
except OSError:
logging.getLogger(__name__).warning('Unable to record orphan cancellation', exc_info=True)
if session_record_path is not None:
from .session_record import write_session_record
try:
write_session_record(Path(session_record_path).with_name('supervisor-lost.json'),
{'operation_id': operation_id, 'phase': 'SUPERVISOR_LOST'})
except (OSError, ValueError):
logging.getLogger(__name__).warning('Unable to record supervisor loss', exc_info=True)
while True:
with _automation_clients_lock:
pending = _record_write_pending(session_record_path, operation_id)
# The same gate blocks new calls/helpers after loss. A stale
# POSTCALL on disk cannot override an active in-memory call.
if (pending is False and _phase_pending(_native_phases.get(operation_id)) is False
and all(helper.poll() is not None for helper in _automation_clients)):
os._exit(75) # This Python worker only; native children remain untouched.
threading.Event().wait(.2)
finally:
parent.Close()
if cancel_path is None or stop_clients:
os._exit(75)
thread = threading.Thread(target=watch, name='supervisor-watchdog', daemon=True)
thread.start()
@@ -80,6 +149,8 @@ class DesktopWorker:
self.companions = []
self.companion_pids = []
self.job_bound = False
self.native_worker = any(flag in command for flag in (
'--verify-worker', '--management-worker', '--repair-worker', '--native-backup-worker'))
# Track the session, but never use kill-on-close: PCB descendants can
# still own a database transaction when the Python supervisor exits.
try:
@@ -99,7 +170,13 @@ class DesktopWorker:
except Exception:
# Kill-through-job is a cleanup guarantee, not a launch requirement.
self.job_bound = False
except Exception:
except Exception as error:
# No worker means no native descendants; a partial launch is uncertain.
error.worker_launch_release = {
'process_created': self.handle is not None,
'safe_to_continue': self.handle is None,
'worker_exited': self.handle is None,
'remaining_pids': [self.pid] if self.handle is not None else []}
if self.handle is not None:
try:
win32process.TerminateProcess(self.handle, 1)
@@ -156,7 +233,8 @@ class DesktopWorker:
expected = {Path(__file__).with_name(name).resolve()
for name in ('save_verified_block.vbs', 'close_failed_block.vbs')}
clients = []
for row in query_owners(('cscript.exe',), timeout_seconds=3):
# Interpreter + COM startup measures ~4.75s on the field machine.
for row in query_owners(('cscript.exe',), timeout_seconds=10):
if row.get('parent') != self.pid:
continue
args = _command_args(row['command'])
@@ -173,24 +251,25 @@ class DesktopWorker:
return {'safe_to_continue': False, 'remaining_pids': [], 'error': str(error)}
def stop_automation_clients(self):
from .mentor_adapter import _terminate_pid
for pid in self._save_clients():
_terminate_pid(pid)
# An absent return marker is not permission to interrupt Save/Close.
return self._save_clients()
@property
def returncode(self):
return self.poll()
def kill(self):
"""Stop automation clients only; preserve native PCB/iCDB sessions."""
"""Native workers use cooperative stop, including compatibility callers."""
if getattr(self, 'native_worker', False) or hasattr(self, 'cancel_path'):
if hasattr(self, 'cancel_path'):
self.cancel_path.touch()
return False
try:
win32process.TerminateProcess(self.handle, 1)
except Exception:
pass
# The worker is now stopped and cannot spawn another save helper.
try:
self.wait(timeout=5)
self.stop_automation_clients()
except Exception:
pass
for companion in self.companions:
@@ -202,6 +281,13 @@ class DesktopWorker:
def bind_cancellation(self, path):
self.cancel_path = Path(path)
def bind_session(self, job):
from .session_record import register_worker_session
register_worker_session(self, job)
def native_write_pending(self):
return _record_write_pending(getattr(self, 'session_record_path', None))
def request_stop(self):
if not hasattr(self, 'cancel_path'):
# Compatibility for diagnostic workers without the Verify protocol.
@@ -221,6 +307,8 @@ class DesktopWorker:
try:
if self.poll() is None:
self.request_stop()
if getattr(self, 'native_worker', False) or hasattr(self, 'cancel_path'):
return False # The observer keeps this exact process/desktop handle.
except Exception:
pass
for companion in self.companions:
+377
View File
@@ -0,0 +1,377 @@
"""Process-bound native opening check. This module never acquires PCB COM.
The caller must bind the newly launched PCB command/PID before entering here.
The check runs in that worker's private desktop and closes only that session.
"""
from __future__ import annotations
import ctypes
from ctypes import wintypes
import heapq
import hashlib
import ntpath
import os
from pathlib import Path
import re
import time
from .dialog_probe import _bounded_call
from .icdb_recovery import LOG_LIMIT, _read_log_delta, decode_log, snapshot_log
_MAX_WINDOWS = 128
_MAX_CHILDREN = 512
def _normal(path):
return ntpath.normcase(ntpath.normpath(str(path).strip()))
def _log_paths(job):
project = Path(job['project']).parent
pcb_logs = Path(job['pcb']).parent / 'LogFiles'
paths = [pcb_logs / 'PCBAutoSave.txt', pcb_logs / 'PCBLogFile.txt']
directory = project / 'database/cdbsvr/log'
directories = []
if directory.is_dir():
with os.scandir(directory) as entries:
directories = [entry.name for entry in entries if entry.is_dir(follow_symlinks=False)]
# Native names start with sortable YYYY-MM-DD HH.MM.SS. Only the
# newest candidates can describe this launch; history is never read.
for name in heapq.nlargest(8, directories):
with os.scandir(directory / name) as entries:
logs = heapq.nlargest(2, (entry.name for entry in entries
if entry.is_file(follow_symlinks=False) and entry.name.casefold().endswith('.txt')))
paths.extend(directory / name / log for log in logs)
return paths, directories
def _read_logs(job):
result = {}
paths, directories = _log_paths(job)
for path in paths:
try:
record = snapshot_log(path)
record['mtime_ns'] = path.stat().st_mtime_ns
result[str(path)] = record
except FileNotFoundError:
result[str(path)] = snapshot_log(path)
return {'files': result, 'server_directories': directories}
def _read_bounded(function):
try:
return _bounded_call(function, 2)
except TimeoutError as error:
# Abort this check, rather than spawning another reader while a slow
# share still owns the previous read. The daemon performs reads only.
raise RuntimeError('DIAGNOSTIC_LOG_READ_TIMEOUT: native evidence unavailable') from error
def snapshot_diagnostic_logs(job) -> dict:
"""Capture pre-launch content, so copied/stale successes cannot pass.
Store the returned JSON-compatible value at job['diagnostic_log_baseline'].
File count, content size and the entire read are bounded; failure is closed.
"""
started = time.time_ns()
return {'started_ns': started, **_read_bounded(lambda: _read_logs(job))}
def _fresh_logs_read(job, baseline):
paths, _ = _log_paths(job)
fresh = {}
for path in paths:
try:
stat = path.stat()
old = baseline['files'].get(str(path))
if stat.st_mtime_ns < baseline['started_ns']:
continue
if old is None:
if path.parent.name in baseline.get('server_directories', ()):
continue # An unobserved historical file cannot prove freshness.
old = {'size': 0, 'tail_sha256': hashlib.sha256(b'').hexdigest()}
if 'error' in old or (old.get('mtime_ns') == stat.st_mtime_ns and old['size'] == stat.st_size):
continue
content, appended = _read_log_delta(path, old)
if not content and stat.st_mtime_ns != old.get('mtime_ns') and stat.st_size == old['size']:
# Native PCBLog is overwritten with the same successful text.
content, _ = _read_log_delta(path, {'size': 0, 'tail_sha256': ''})
if not appended and old.get('truncated') and stat.st_size > LOG_LIMIT:
continue
fresh[str(path)] = decode_log(content)
except FileNotFoundError:
continue
return fresh
def _fresh_logs(job, baseline):
return _read_bounded(lambda: _fresh_logs_read(job, baseline))
def _load_evidence(job, pcb_pid, baseline, *, require_autosave=True):
logs = _fresh_logs(job, baseline)
project = Path(job['project']).parent
pcb = str(Path(job['pcb']))
auto_path = str(Path(job['pcb']).parent / 'LogFiles/PCBAutoSave.txt')
read_path = str(Path(job['pcb']).parent / 'LogFiles/PCBLogFile.txt')
names = re.findall(r'^Job Name:\s*(.*?)\s*$', logs.get(auto_path, ''), re.M | re.I)
if names and any(_normal(name) != _normal(pcb) for name in names):
return None
if require_autosave and not names:
return None
if 'Successfully read the design.' not in logs.get(read_path, ''):
return None
database = _normal(project / 'database')
project_file = _normal(job['project'])
for path, text in logs.items():
if path in (auto_path, read_path):
continue
db_paths = re.findall(r'S: Project path \[(.*?)\]', text)
prj_paths = re.findall(r'S: Project file \(\.prj\) path \[(.*?)\]', text)
startup = re.search(r'S: Started up automatically by iCDB Client: [^\r\n]*:' + str(pcb_pid) + r'\s*$', text, re.M)
client = re.search(r'Registered sub-client \[Expedition PCB[^\r\n]*process ID \[' + str(pcb_pid) + r'\]', text)
if (db_paths and all(_normal(value) == database for value in db_paths)
and prj_paths and all(_normal(value) == project_file for value in prj_paths)
and startup and client and 'S: iCDB database loaded' in text):
return {'pcb': pcb, 'pcb_pid': pcb_pid,
'logs': [auto_path, read_path, path], 'document_identity': 'native_path_pid_logs'}
return None
def _target_frame(windows, name):
matches = []
for window in windows:
if not window['class'].casefold().startswith('afx:') or not window.get('enabled', True):
continue
if not window['text'].casefold().startswith('expedition pcb') or f'[{name}]'.casefold() not in window['text'].casefold():
continue
mdi = {child['hwnd'] for child in window['children'] if child['class'].casefold() == 'mdiclient'}
if any(child.get('parent') in mdi and child['text'].casefold() == name.casefold()
and child['class'].casefold().startswith('afx:') for child in window['children']):
matches.append(window)
return matches[0] if len(matches) == 1 else None
def _windows(pid):
"""Read only this PID on the current desktop; cap text and enumeration work."""
import win32api
import win32gui
import win32process
user = ctypes.WinDLL('user32', use_last_error=True)
user.GetWindowTextW.argtypes = [wintypes.HWND, wintypes.LPWSTR, ctypes.c_int]
user.GetWindowTextW.restype = ctypes.c_int
deadline = time.monotonic() + .2
def text(hwnd):
if time.monotonic() >= deadline:
raise RuntimeError('DIAGNOSTIC_WINDOW_LIMIT: window scan exceeded 0.2s')
buffer = ctypes.create_unicode_buffer(4096)
# All handles belong to the external PCB PID: GetWindowText reads
# Windows' cached caption without sending a message to its UI thread.
# Custom controls may have no cached text; missing evidence fails closed.
user.GetWindowTextW(hwnd, buffer, len(buffer))
return buffer.value
result = []
seen = 0
def collect(hwnd, _):
nonlocal seen
if win32process.GetWindowThreadProcessId(hwnd)[1] != pid or not win32gui.IsWindowVisible(hwnd):
return True
seen += 1
if seen > _MAX_WINDOWS:
raise RuntimeError('DIAGNOSTIC_WINDOW_LIMIT: too many windows')
cls = win32gui.GetClassName(hwnd)
if cls.casefold() in {'uac_inputindicatoroverlaywnd', 'uac input indicator'}:
return True
children = []
count = 0
def child(ch, _):
nonlocal count
count += 1
if count > _MAX_CHILDREN:
raise RuntimeError('DIAGNOSTIC_WINDOW_LIMIT: too many controls')
child_cls = win32gui.GetClassName(ch)
if (cls == '#32770' or cls.casefold() == 'internet explorer_hidden'
or child_cls.casefold() == 'mdiclient' or child_cls.casefold().startswith('afx:')):
children.append({'hwnd': ch, 'class': child_cls, 'text': text(ch),
'id': win32gui.GetDlgCtrlID(ch), 'parent': win32gui.GetParent(ch),
'enabled': bool(win32gui.IsWindowEnabled(ch))})
return True
if cls == '#32770' or cls.casefold().startswith('afx:') or cls.casefold() == 'internet explorer_hidden':
win32gui.EnumChildWindows(hwnd, child, None)
caption = text(hwnd)
if cls.casefold() == 'internet explorer_hidden' and not caption and not children:
left, top, right, bottom = win32gui.GetWindowRect(hwnd)
if left == right and top == bottom:
return True
result.append({'hwnd': hwnd, 'pid': pid, 'class': cls, 'text': caption,
'enabled': bool(win32gui.IsWindowEnabled(hwnd)), 'children': children,
'owner': int(win32gui.GetWindow(hwnd, 4) or 0)})
return True
win32api.SetLastError(0)
try:
win32gui.EnumWindows(collect, None)
except Exception as error:
if getattr(error, 'winerror', None) not in (0, 2) or result:
raise
return result
class _NativeSession:
def __init__(self, job, pid, launcher):
if launcher is None or not callable(getattr(launcher, 'poll', None)):
raise RuntimeError('DIAGNOSTIC_LAUNCHER_REQUIRED: retain the exact launch process handle')
import win32api
import win32service
import win32con
import win32event
import win32gui
self.handles = []
self.pid = int(pid)
self.launcher = launcher
self.gui = win32gui
self.event = win32event
desktop = win32service.GetThreadDesktop(win32api.GetCurrentThreadId())
actual = win32service.GetUserObjectInformation(desktop, 2)
expected = str(job.get('desktop', '')).split('\\')[-1]
if not expected or expected.casefold() in {'default', 'winlogon', 'disconnect'} or actual.casefold() != expected.casefold():
raise RuntimeError('DIAGNOSTIC_DESKTOP_MISMATCH')
self.pcb = win32api.OpenProcess(win32con.SYNCHRONIZE, False, int(pid))
self.handles.append(self.pcb)
def pcb_alive(self):
return self.event.WaitForSingleObject(self.pcb, 0) != 0
def launcher_alive(self):
return self.launcher.poll() is None
def post(self, hwnd, message):
import win32process
if not self.pcb_alive() or win32process.GetWindowThreadProcessId(hwnd)[1] != self.pid:
raise RuntimeError('DIAGNOSTIC_STALE_WINDOW: target window identity changed')
self.gui.PostMessage(hwnd, message, 0, 0)
def close(self):
for handle in self.handles:
handle.Close()
self.handles.clear()
def _cancel(job):
if job.get('cancel_signal') and Path(job['cancel_signal']).exists():
raise RuntimeError('VERIFY_CANCELLED: native opening check cancelled')
def _pending_prompt(windows, *, closing=False, frame=None, session=None, clicked=None, pcb_name=''):
for window in windows:
known_titles = {'expedition pcb', 'expedition pcb - reusable block'}
if (window['class'].casefold().startswith('afx:') and not window.get('owner')
and (window['text'].casefold() in known_titles
or frame is not None and window['hwnd'] == frame['hwnd'])):
continue
text = ' '.join([window['text'], *(child['text'] for child in window['children'])]).casefold()
if not closing and ('select the desired product license' in text or window['text'] == 'Tip of the Day'):
return True # The caller's PID-scoped DialogAgent owns startup choices.
standard_save = (pcb_name and window['class'] == '#32770' and window['text'] == 'Expedition PCB'
and any(child['class'].casefold() == 'static'
and child['text'].strip().casefold() == f'Save changes to {pcb_name}?'.casefold()
for child in window['children'])
and {child.get('id') for child in window['children'] if child['class'].casefold() == 'button'} == {2, 6, 7})
if closing and window.get('owner') == frame['hwnd'] and (standard_save or 'do you want to save the reusable block cell' in text):
targets = [child for child in window['children'] if child['class'].casefold() == 'button'
and child.get('id') == 7 and child['text'].replace('&', '').strip().casefold() in {'no', '否(n)', '否'}
and child.get('enabled', True)]
if len(targets) == 1:
if window['hwnd'] not in clicked:
session.post(targets[0]['hwnd'], 0xF5)
clicked.add(window['hwnd'])
continue
raise RuntimeError('DIAGNOSTIC_CLOSE_UNCONFIRMED: unknown native prompt preserved; '
f"pid={window['pid']} hwnd={window['hwnd']} class={window['class']!r} "
f"title={window['text']!r}")
return False
def check_target_window(job, pcb_pid, *, emit, error_reader, launcher_process=None) -> dict:
"""Prove this launch opened its native document, close it, and prove exit.
PASS describes opening only; it does not imply refresh READY or unoccupied
shared library data. Unknown prompts are preserved for existing recovery.
"""
baseline = job.get('diagnostic_log_baseline')
if not isinstance(baseline, dict) or 'started_ns' not in baseline or 'files' not in baseline:
raise RuntimeError('DIAGNOSTIC_BASELINE_REQUIRED: snapshot before native launch')
session = _NativeSession(job, int(pcb_pid), launcher_process)
try:
deadline = time.monotonic() + max(.01, float(job['timeout']))
while True:
_cancel(job)
error = _read_bounded(error_reader) if error_reader else None
if error:
raise RuntimeError(str(error))
if not session.pcb_alive():
raise RuntimeError('DIAGNOSTIC_OPEN_UNCONFIRMED: target PCB exited before loaded evidence')
windows = _windows(int(pcb_pid))
if any(window['pid'] != int(pcb_pid) for window in windows):
raise RuntimeError('DIAGNOSTIC_PID_MISMATCH')
frame = _target_frame(windows, Path(job['pcb']).stem)
# Startup splash/control windows are transient. Unknown windows
# veto closing only once the target actually has loaded evidence.
pending = (_pending_prompt(windows, frame=frame, pcb_name=Path(job['pcb']).name)
if frame is not None else True)
evidence = (_load_evidence(job, int(pcb_pid), baseline, require_autosave=False)
if frame is not None and not pending else None)
if evidence is not None:
emit(stage='diagnostic_native_loaded_candidate', **evidence)
break
if time.monotonic() >= deadline:
raise RuntimeError('DIAGNOSTIC_OPEN_UNCONFIRMED: current target MDI/path/PID logs missing')
time.sleep(.1)
# Re-read the exact PID/frame before the sole close request.
_cancel(job)
windows = _windows(int(pcb_pid))
_pending_prompt(windows, frame=frame, pcb_name=Path(job['pcb']).name)
current = _target_frame(windows, Path(job['pcb']).stem)
if current is None or current['hwnd'] != frame['hwnd']:
raise RuntimeError('DIAGNOSTIC_CLOSE_UNCONFIRMED: target frame changed')
session.post(frame['hwnd'], 0x10)
emit(stage='diagnostic_wm_close', pcb_pid=int(pcb_pid), hwnd=frame['hwnd'])
close_deadline = time.monotonic() + 30
clicked = set()
last_pending = ''
while session.pcb_alive() or session.launcher_alive():
# Cancellation stops future work; an accepted normal close must
# still finish before its queue slot can be safely released.
windows = _windows(int(pcb_pid)) if session.pcb_alive() else []
previous_clicks = set(clicked)
pending_error = None
try:
_pending_prompt(windows, closing=True, frame=frame, session=session, clicked=clicked,
pcb_name=Path(job['pcb']).name)
except RuntimeError as error:
if not str(error).startswith('DIAGNOSTIC_CLOSE_UNCONFIRMED: unknown native prompt'):
raise
# WM_CLOSE creates and destroys native windows asynchronously;
# preserve them and wait for populated controls or process exit.
pending_error = error
if str(error) != last_pending:
last_pending = str(error)
emit(stage='diagnostic_close_pending', detail=last_pending)
for hwnd in clicked - previous_clicks:
emit(stage='diagnostic_save_no', pcb_pid=int(pcb_pid), hwnd=hwnd)
if time.monotonic() >= close_deadline:
if pending_error is not None:
raise pending_error
raise RuntimeError('DIAGNOSTIC_CLOSE_UNCONFIRMED: target PCB/launcher still running')
time.sleep(.1)
# EE7.9 buffers PCBAutoSave's complete Job Name until normal close.
# The candidate permits closing only; PASS still needs that exact path.
evidence = _load_evidence(job, int(pcb_pid), baseline)
if evidence is None:
raise RuntimeError('DIAGNOSTIC_OPEN_UNCONFIRMED: fresh full PCB path not confirmed after close')
emit(stage='diagnostic_native_open_confirmed', **evidence)
return {'status': 'PASS', 'check': 'native_open_only', 'native_processes_exited': True,
'library_ownership': 'NOT_ASSESSED', 'pcb_pid': int(pcb_pid),
'launcher_pid': getattr(launcher_process, 'pid', None), 'evidence': evidence,
'detail': 'Native target opened and its PCB/launcher exited; refresh readiness and shared library ownership not assessed.'}
finally:
session.close()
+47 -10
View File
@@ -236,6 +236,14 @@ class DialogActionExecutor:
def perform(self, snapshot: WindowSnapshot, action: dict) -> ActionResult:
channels = tuple(action.get("channels", ()))
action_name = action.get("action")
if action_name == 'user_confirmed_pcb_open':
try:
from .pcb_open_consent import validate_action_window
if channels != ('win32_button_click',):
raise ValueError('PCB_OPEN_ACTION_CHANNEL_CHANGED')
validate_action_window(snapshot, action)
except Exception as error:
return ActionResult(error_code='DATABASE_RECOVERY_REQUIRED', detail=str(error))
if action_name == 'user_selected_pcb_version':
try:
from .pcb_restore_prompt import validate_action_window
@@ -324,6 +332,7 @@ class DialogAgentServer:
self.channel = JsonChannel(connection)
self.channel.send({"token": paths.token}, timeout_seconds=10)
self.observed_snapshots = {}
self.observed_pids = None
@classmethod
def from_config(cls, config_path: Path, *, probe, executor=None):
@@ -348,7 +357,15 @@ class DialogAgentServer:
try:
phase = DialogPhase(request["phase"])
if request.get("op") == "observe":
snapshots = self.probe.capture(phase)
pids = request.get('pids')
if pids is not None and (not isinstance(pids, list)
or any(type(pid) is not int or pid <= 0 for pid in pids)):
raise ValueError('DIALOG_PID_SCOPE_INVALID')
self.observed_pids = set(pids) if pids is not None else None
snapshots = (self.probe.capture(phase, pids=self.observed_pids) if pids is not None
else self.probe.capture(phase))
if pids is not None:
snapshots = tuple(w for w in snapshots if w.pid in self.observed_pids)
self.observed_snapshots.clear()
for snapshot in snapshots:
self.observed_snapshots[stable_fingerprint(snapshot)] = snapshot
@@ -356,7 +373,9 @@ class DialogAgentServer:
return
if request.get("op") == "perform":
expected = self.observed_snapshots.get(request.get("fingerprint"))
if expected is None:
if (expected is None or int(request.get('hwnd', 0)) != expected.hwnd
or self.observed_pids is not None and (expected.pid not in self.observed_pids
or not self.probe.matches_identity(expected))):
self._response(request, result=ActionResult(
error_code="STALE_DIALOG",
detail="窗口已消失或指纹已变化",
@@ -422,27 +441,45 @@ class DialogAgentClient:
raise
@classmethod
def start(cls, worker, root: Path, *, timeout_seconds: float = 2.0):
def start(cls, worker, root: Path, *, timeout_seconds: float = 2.0, stop=None):
client = cls.prepare(root, desktop_name=worker.name, timeout_seconds=timeout_seconds)
try:
from .defaults import DIALOG_AGENT_STARTUP_TIMEOUT_SECONDS
worker.start_companion(dialog_agent_command(client.paths.config))
client._accept(timeout_seconds=10.0)
# Frozen imports can exceed ten seconds; PCB remains behind its gate.
client._accept(timeout_seconds=DIALOG_AGENT_STARTUP_TIMEOUT_SECONDS, stop=stop)
return client
except Exception:
client.close()
raise
def _accept(self, timeout_seconds):
def _accept(self, timeout_seconds, *, stop=None):
if self.channel is not None:
return
self.listener.settimeout(timeout_seconds)
connection, _address = self.listener.accept()
deadline = time.monotonic() + timeout_seconds
while True:
if stop is not None and stop.is_set():
raise InterruptedError('VERIFY_CANCELLED: 弹窗代理启动已取消,未启动 PCB')
remaining = deadline - time.monotonic()
if remaining <= 0:
raise TimeoutError('dialog-agent startup timed out')
self.listener.settimeout(min(.2, remaining))
try:
connection, _address = self.listener.accept()
break
except socket.timeout:
continue
channel = JsonChannel(connection)
try:
connection.setsockopt(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1)
hello = channel.receive(timeout_seconds)
remaining = deadline - time.monotonic()
if remaining <= 0:
raise TimeoutError('dialog-agent handshake timed out')
hello = channel.receive(min(self.timeout_seconds, remaining))
if hello.get("token") != self.paths.token:
raise DialogAgentProtocolError("dialog-agent authentication failed")
if stop is not None and stop.is_set():
raise InterruptedError('VERIFY_CANCELLED: 弹窗代理启动已取消,未启动 PCB')
self.channel = channel
except Exception:
channel.close()
@@ -473,8 +510,8 @@ class DialogAgentClient:
raise DialogAgentProtocolError(f"{response['error_code']}: {response.get('detail', '')}")
return response.get("result")
def observe(self, phase: DialogPhase) -> tuple[WindowSnapshot, ...]:
result = self._request("observe", phase)
def observe(self, phase: DialogPhase, *, pids=None) -> tuple[WindowSnapshot, ...]:
result = self._request("observe", phase, **({'pids': sorted(pids)} if pids is not None else {}))
return tuple(WindowSnapshot.from_record(item) for item in (result or ()))
def perform(self, *, hwnd: int, fingerprint: str, action: dict,
+15 -1
View File
@@ -265,9 +265,23 @@ class WindowProbe:
except Exception:
return None
def capture(self, phase: DialogPhase) -> tuple[WindowSnapshot, ...]:
def matches_identity(self, snapshot):
try:
return (self.win32gui.IsWindowVisible(snapshot.hwnd)
and self.win32process.GetWindowThreadProcessId(snapshot.hwnd)[1] == snapshot.pid
and self.win32gui.GetClassName(snapshot.hwnd) == snapshot.window_class)
except Exception:
return False
def capture(self, phase: DialogPhase, *, pids=None) -> tuple[WindowSnapshot, ...]:
snapshots = []
for hwnd in self._handles():
if pids is not None:
try:
if self.win32process.GetWindowThreadProcessId(hwnd)[1] not in pids:
continue
except Exception:
continue
try:
window_class = str(self.win32gui.GetClassName(hwnd) or "")
except Exception:
+6 -2
View File
@@ -100,11 +100,15 @@ class DialogBackend(Protocol):
class AgentDialogBackend:
"""Adapt the private-desktop JSONL client to the supervisor backend."""
def __init__(self, client) -> None:
def __init__(self, client, *, pids=None) -> None:
self.client = client
self.pids = pids
def observe(self, phase: DialogPhase):
return self.client.observe(phase)
if self.pids is None:
return self.client.observe(phase)
pids = set(self.pids())
return tuple(w for w in self.client.observe(phase, pids=pids) if w.pid in pids)
def perform(self, snapshot: WindowSnapshot, action: dict, phase: DialogPhase) -> ActionResult:
return self.client.perform(
+63 -22
View File
@@ -24,13 +24,42 @@ def _command_args(command):
kernel.LocalFree(ctypes.cast(args, wintypes.HLOCAL))
def _startup_snapshot(hwnd, pcb_pid):
import win32gui, win32process
from .dialog_models import ControlSnapshot, DialogPhase, WindowSnapshot
if win32process.GetWindowThreadProcessId(hwnd)[1] != pcb_pid:
raise ValueError('STARTUP_WINDOW_OWNER_CHANGED')
children = []
win32gui.EnumChildWindows(hwnd, lambda child, _: children.append(child), None)
controls = tuple(ControlSnapshot(child, win32gui.GetDlgCtrlID(child) & 0xffff,
win32gui.GetClassName(child), win32gui.GetWindowText(child),
enabled=bool(win32gui.IsWindowEnabled(child))) for child in children)
return WindowSnapshot(hwnd, pcb_pid, 'ExpeditionPCB.exe', 'pcb',
win32gui.GetClassName(hwnd), win32gui.GetWindowText(hwnd), DialogPhase.PCB_STARTING, controls)
def _cancel_target(snapshot):
from .post_repair_consent import exact_prompt
from .dialog_rules import pcb_version_choice
if exact_prompt(snapshot):
return 'read_write_access', next(c for c in snapshot.controls if c.control_id == 7)
if pcb_version_choice(snapshot) is not None:
buttons = [c for c in snapshot.controls if c.class_name == 'Button']
cancel = [c for c in buttons if c.control_id == 2 and c.name == 'Cancel' and c.enabled]
if (len(cancel) == 1 and {c.control_id for c in buttons} == {6212, 6213, 6206, 1, 2}
and any(c.control_id == 6211 and c.name == '* Cancelling will load no design.'
for c in snapshot.controls)):
return 'pcb_version_choice', cancel[0]
return None
def close_failed_startup(job, pcb_pid, *, baseline_pids=(), emit=lambda **event: None,
timeout_seconds=20, launcher_process=None):
"""Normally close our empty PCB after a diagnosed database startup failure.
No COM attach: there is no document, and the global ROT can name another
session. Unknown windows, live database addresses and other desktops are
retained. WM_CLOSE is cooperative; no save prompt is accepted or killed.
No COM attach: the global ROT can name another session. Only the exact
read/write question is declined; no ownership is taken or version restored.
Unknown windows and documents are retained. WM_CLOSE is cooperative.
"""
import win32api, win32con, win32gui, win32process
from .mentor_adapter import _process_details
@@ -43,23 +72,6 @@ def close_failed_startup(job, pcb_pid, *, baseline_pids=(), emit=lambda **event:
if launcher_process is not None:
result['launcher_pid'] = launcher_process.pid
try:
diagnosed = False
diagnosis_path = root/'diagnostics/icdb-recovery.json'
if diagnosis_path.is_file():
diagnosis = json.loads(diagnosis_path.read_text('utf-8'))
diagnosed = diagnosis.get('error_code') == 'ICDB_REPAIR_REQUIRED'
dialogs_path = root/'dialogs.jsonl'
if not diagnosed and dialogs_path.is_file():
for line in dialogs_path.read_text('utf-8').splitlines():
try:
dialog = json.loads(line)
except ValueError:
continue
if (dialog.get('pid') == pcb_pid and dialog.get('phase') == 'PCB_STARTING'
and dialog.get('error_code') == 'MENTOR_DATABASE_LOAD_FAILED'):
diagnosed = True
if not diagnosed:
raise ValueError('STARTUP_FAILURE_UNCONFIRMED')
desktop = job.get('desktop', '')
if (job.get('interactive_input') or not desktop.startswith('CbbVerify_')
or _current_desktop_name() != desktop):
@@ -69,14 +81,20 @@ def close_failed_startup(job, pcb_pid, *, baseline_pids=(), emit=lambda **event:
expected = os.path.normcase(os.path.abspath(job['pcb']))
deadline = time.monotonic() + max(0, timeout_seconds)
sent = False
declined_hwnd = None
declined_kind = None
cancelled_open = False
while True:
records = [r for r in _process_details('ExpeditionPCB.exe') if r[0] == pcb_pid]
if not records:
if declined_hwnd is not None:
cancelled_open = True
result['cancelled_prompt'] = declined_kind
addresses = unreleased_addresses(job['lmc_file'], job['project'], job['pcb'])
launcher_exited = launcher_process is None or launcher_process.poll() is not None
result['release'] = {'pcb_exited': True, 'launcher_exited': launcher_exited,
'addresses': addresses,
'safe_to_continue': launcher_exited and not addresses}
'safe_to_continue': launcher_exited and (cancelled_open or not addresses)}
if result['release']['safe_to_continue']:
result['status'] = 'CLOSED'
result.pop('detail', None)
@@ -100,6 +118,29 @@ def close_failed_startup(job, pcb_pid, *, baseline_pids=(), emit=lambda **event:
windows.append(hwnd)
win32api.SetLastError(0)
win32gui.EnumWindows(observe, None)
if declined_hwnd is not None and declined_hwnd not in windows:
cancelled_open = True
result['cancelled_prompt'] = declined_kind
if declined_hwnd is None:
for hwnd in windows:
if win32gui.GetClassName(hwnd) != '#32770':
continue
snapshot = _startup_snapshot(hwnd, pcb_pid)
target = _cancel_target(snapshot)
if target is None:
continue
# Re-read immediately before declining; never select a version.
current = _startup_snapshot(hwnd, pcb_pid)
if _cancel_target(current) != target or current != snapshot:
raise ValueError('STARTUP_CANCEL_PROMPT_CHANGED')
declined_kind, cancel = target
win32gui.PostMessage(cancel.handle, win32con.BM_CLICK, 0, 0)
declined_hwnd = hwnd
emit(stage='startup_open_declined', pcb_pid=pcb_pid, hwnd=hwnd,
control_id=cancel.control_id, prompt=declined_kind)
break
if declined_hwnd is not None:
continue
main = [w for w in windows if win32gui.GetClassName(w).startswith('Afx:')
and win32gui.GetWindowText(w).strip() in (
'Expedition PCB', 'Expedition PCB - Reusable Block')
@@ -131,7 +172,7 @@ def close_failed_startup(job, pcb_pid, *, baseline_pids=(), emit=lambda **event:
win32api.SetLastError(0)
win32gui.EnumChildWindows(main[0], child, None)
addresses = unreleased_addresses(job['lmc_file'], job['project'], job['pcb'])
if len(main) == 1 and not unknown and not documents and not addresses:
if len(main) == 1 and not unknown and not documents and (cancelled_open or not addresses):
win32gui.PostMessage(main[0], win32con.WM_CLOSE, 0, 0)
sent = True
emit(stage='startup_close_requested', pcb_pid=pcb_pid, hwnd=main[0])
+1 -1
View File
@@ -21,7 +21,7 @@ def maintenance_counts(results):
'checked_without_repair': sum(r.status == 'CHECKED' for r in results),
'issues_found': sum(r.error_category == 'ICDB_REPAIR_REQUIRED' for r in results),
'in_use': sum(r.error_category == 'ICDB_DATABASE_IN_USE' for r in results),
'unconfirmed': sum(r.status not in ('CHECKED', 'SKIPPED', 'CANCELLED')
'unconfirmed': sum(r.status not in ('PASS', 'CHECKED', 'SKIPPED', 'CANCELLED')
and r.error_category not in ('ICDB_REPAIR_REQUIRED', 'ICDB_DATABASE_IN_USE') for r in results),
'cancelled': sum(r.status == 'CANCELLED' for r in results),
'skipped': sum(r.status == 'SKIPPED' for r in results),
+33 -1
View File
@@ -17,6 +17,37 @@ _SESSION_LOSS = re.compile(
r'\bthe\s+last\s+server\s+session\s+on\s+different\s+server\s+machine\s+'
r'lost\s+the\s+connection\s+to\s+the\s+database\b', re.IGNORECASE)
_ICDB_MESSAGE = re.compile(r'(?<![A-Za-z0-9_])icdb(?![A-Za-z0-9_.\\/])', re.I)
_FAILURE_MESSAGE = re.compile(
r'\b(?:error|failed|failure|unable|cannot|could\s+not|lost|corrupt(?:ed|ion)?|'
r'inconsistent|unsupported|refused|denied)\b|失败|错误|异常|无法|损坏|失联|拒绝|未完成', re.I)
def _icdb_error_paragraphs(text, *, error_message=False):
text = str(text or '').replace('\r', '')
for record in _session_losses(text):
yield record['native_message'].strip()
for paragraph in text.splitlines():
# A pathname or our own category label is not native error content.
message = re.sub(r'\[[^\]\n]*[\\/][^\]\n]*\]', '', paragraph)
if _ICDB_MESSAGE.search(message) and (error_message or _FAILURE_MESSAGE.search(message)):
yield paragraph.strip()
def icdb_error_content(text, *, error_message=False):
"""Extract failure paragraphs mentioning iCDB; codes/healthy text do not qualify."""
return '\n\n'.join(dict.fromkeys(_icdb_error_paragraphs(text, error_message=error_message)))
def read_icdb_error(path, baseline):
"""Retain only new native error content, with the same bounded freshness rules."""
data, appended = _read_log_delta(path, baseline)
if not appended and (not baseline or baseline.get('truncated') or 'icdb_errors' not in baseline):
return ''
previous = baseline.get('icdb_errors', [])
return '\n\n'.join(message for message in _icdb_error_paragraphs(decode_log(data))
if appended or ' '.join(message.split()) not in previous)
def parse_session_loss(text):
"""Keep the native target verbatim; never infer a local repair target."""
@@ -68,9 +99,10 @@ def snapshot_log(path):
losses = [_session_identity(record) for record in _session_losses(decode_log(tail))]
return {'size': size, 'tail_sha256': hashlib.sha256(tail).hexdigest(),
'truncated': size > LOG_LIMIT,
'icdb_errors': [' '.join(message.split()) for message in _icdb_error_paragraphs(decode_log(tail))],
'session_losses': losses}
except FileNotFoundError:
return {'size': 0, 'tail_sha256': hashlib.sha256(b'').hexdigest()}
return {'size': 0, 'tail_sha256': hashlib.sha256(b'').hexdigest(), 'icdb_errors': []}
except OSError as error:
return {'error': str(error)}
+1
View File
@@ -109,6 +109,7 @@ def refresh_integration(pi, *, emit=lambda **event: None, settle_seconds=5, time
emit(stage='integration_action_unavailable', action=method,
initial_status=state[name], reason='mentor_not_allowed')
raise IntegrationError(f'{name}: required but not allowed')
check_error() # The permission getter itself can block past cancellation.
emit(stage='integration_action', action=method, initial_status=state[name],
reason='required_verification' if name not in executed else 'resynchronize')
started = clock()
+249 -34
View File
@@ -17,6 +17,16 @@ class LibraryBusy(OSError):
pass
def inspect_residual_sessions(library, *, block_names=()):
from .residual_sessions import inspect_residual_sessions as inspect
return inspect(library, block_names=block_names)
def release_residual_sessions(plan, evidence_dir, *, stop=None, force=False):
from .residual_sessions import release_residual_sessions as release
return release(plan, evidence_dir, stop=stop, force=force)
def _key(path):
return str(Path(path).resolve()).casefold()
@@ -53,32 +63,83 @@ def checkpoint_record(path):
def has_blocker(library):
path = blocker_path(library)
if path.with_suffix('.pending').exists():
return True
if not path.exists():
return False
record = checkpoint_record(path)
# Complete snapshots retain every file, including their source lease
# checkpoint. Only a valid relocated snapshot record can be ignored.
return record is None or _key(record['library']) == _key(library)
for marker in (path, path.with_suffix('.pending')):
if marker.exists():
record = checkpoint_record(marker)
if record is None or _key(record['library']) == _key(library):
return True
return False
def _persist_blocker(library, status='OWNER_RELEASE_UNCONFIRMED', *, lease_id):
def has_retained_library(library):
"""Include leases retained without a JSON checkpoint, such as PCB checks."""
path = Path(library).resolve()
key = _key(path.with_name('.' + path.stem + '.cbb-upload.lock'))
with _guard:
owner = _owners.get(key)
return bool(owner and owner['retained'])
def recheck_previous_release(library, evidence_dir, *, block_name=None, required=False, progress=None):
"""Historical records trigger fresh proof, never process termination or repair."""
if not required:
with _guard:
owner = _owners.get(LibraryLock(library).key)
marker = blocker_path(library)
record = checkpoint_record(marker)
if (owner and owner['thread'] == threading.get_ident() and owner['depth'] > 0
and not owner['retained'] and not marker.with_suffix('.pending').exists()
and record and record['lease_id'] == owner['lease_id']
and record['pid'] == os.getpid() and record['status'] == 'OPERATION_IN_PROGRESS'
and record.get('block_name') == owner.get('block_name')
and owner.get('block_name') in (None, block_name)
and 'export_analysis' not in record):
return None # Nested Verify belongs to this live transaction, not a past crash.
if not (required or has_blocker(library) or has_retained_library(library)):
return None
def notify(stage, **details):
if progress is not None:
try:
progress({'stage': stage, 'block': block_name or '', **details})
except Exception:
import logging
logging.getLogger(__name__).warning('Release progress observer failed', exc_info=True)
notify('previous_release_check', detail='正在复核历史记录对应的实际占用')
try:
proof = release_retained_library(library, evidence_dir, block_name=block_name)
except Exception as error:
notify('previous_release_unconfirmed', detail=str(error))
raise
notify('previous_release_confirmed', release_proof=proof,
detail='已确认会话释放并归档历史记录;这不表示历史数据已恢复')
return proof
def _persist_blocker(library, status='OWNER_RELEASE_UNCONFIRMED', *, lease_id, block_name=None):
import win32api, win32process
path = blocker_path(library)
pending = path.with_suffix('.pending')
previous = checkpoint_record(path)
export_record = ({'export_analysis': previous['export_analysis']}
if previous and previous['lease_id'] == lease_id and 'export_analysis' in previous else {})
if block_name is not None:
export_record['block_name'] = block_name
elif previous and previous['lease_id'] == lease_id and 'block_name' in previous:
export_record['block_name'] = previous['block_name']
with pending.open('w', encoding='utf-8') as stream:
json.dump({**export_record, 'protocol': 'cbb-refresh-verify-library-lease', 'version': 1,
'lease_id': lease_id, 'library': str(library), 'status': status,
'machine': socket.gethostname(),
'supervisor_created': win32process.GetProcessTimes(win32api.GetCurrentProcess())['CreationTime'].isoformat(),
'pid': os.getpid(), 'time': datetime.now(timezone.utc).isoformat()}, stream)
stream.flush(); os.fsync(stream.fileno())
pending.replace(path)
class LibraryLock:
def __init__(self, library, *, safe_to_release=None, checkpoint=True):
def __init__(self, library, *, safe_to_release=None, checkpoint=True, block_name=None):
library = Path(library).resolve()
self.library = library
self.path = library.with_name('.' + library.stem + '.cbb-upload.lock')
@@ -86,6 +147,7 @@ class LibraryLock:
self.entered = False
self.safe_to_release = safe_to_release
self.checkpoint = checkpoint
self.block_name = block_name
self.maintenance_guard = None
self.maintenance_mutex = None
@@ -105,6 +167,13 @@ class LibraryLock:
except Exception:
released = False
if released:
if owner.get('checkpoint', True):
marker = blocker_path(self.library)
record = checkpoint_record(marker)
if (record is None or record['lease_id'] != owner['lease_id']
or 'export_analysis' in record or marker.with_suffix('.pending').exists()):
raise LibraryBusy('协调检查点身份或导出会话尚未确认,保持库锁。')
marker.unlink()
owner['handle'].Close()
del _owners[self.key]
owner = None
@@ -126,6 +195,8 @@ class LibraryLock:
win32con.GENERIC_READ | win32con.GENERIC_WRITE, 0, None,
win32con.OPEN_ALWAYS, win32con.FILE_ATTRIBUTE_NORMAL, None)
except Exception as error:
if getattr(error, 'winerror', None) not in (32, 33):
raise
raise LibraryBusy(f'中心库已锁定或无法创建协调锁:{self.path}: {error}') from error
try:
data = (f'machine={socket.gethostname()}; pid={os.getpid()}; '
@@ -140,14 +211,16 @@ class LibraryLock:
import pywintypes
win32file.SetFileTime(handle, None, None, pywintypes.Time(datetime.now()))
if self.checkpoint:
_persist_blocker(self.library, 'OPERATION_IN_PROGRESS', lease_id=lease_id)
_persist_blocker(self.library, 'OPERATION_IN_PROGRESS', lease_id=lease_id,
block_name=self.block_name)
except Exception:
handle.Close()
raise
_owners[self.key] = {'thread': threading.get_ident(), 'depth': 1,
'handle': handle, 'retained': False, 'lease_id': lease_id,
'checkpoint': self.checkpoint,
'release_probe': self.safe_to_release if not self.checkpoint else None}
'block_name': self.block_name,
'release_probe': self.safe_to_release}
self.entered = True
return self
@@ -241,6 +314,84 @@ def _supervisor_alive(pid):
handle.Close()
def _recorded_supervisor_alive(record):
"""None means legacy/unknown identity; the native/file guard still applies."""
if record.get('machine') != socket.gethostname() or not record.get('supervisor_created'):
return None
import pywintypes, win32api, win32con, win32process
try:
handle = win32api.OpenProcess(win32con.SYNCHRONIZE | 0x1000, False, record['pid'])
except pywintypes.error as error:
return False if error.winerror == 87 else None
try:
return (win32process.GetProcessTimes(handle)['CreationTime'].isoformat() == record['supervisor_created']
and _supervisor_alive(record['pid']))
finally:
handle.Close()
def _owner_matches_scope(item, library, roots):
"""False requires an explicit other project; a missing path stays unknown."""
import ntpath
from .failed_session import _command_args
command = str(item.get('command') or '')
if not command:
return None
try:
args = _command_args(command)
except (OSError, ValueError):
return None
candidates = []
worker_flags = {'--verify-worker', '--repair-worker', '--management-worker', '--native-backup-worker'}
for index, arg in enumerate(args[1:], 1):
if arg.casefold() in worker_flags and index + 1 < len(args):
try:
request = Path(args[index + 1])
if request.stat().st_size > 1024 * 1024:
return None
record = json.loads(request.read_text(encoding='utf-8'))
for key in ('library', 'lmc_file', 'project', 'database'):
if isinstance(record.get(key), str):
candidates.append(record[key])
except (OSError, ValueError, TypeError, AttributeError):
return None
value = arg.split('=', 1)[1] if arg.startswith('-') and '=' in arg else arg
if (ntpath.isabs(value) and (ntpath.splitext(value)[1].casefold() in ('.prj', '.pcb', '.lmc')
or arg.casefold().startswith(('-project=', '-icdb='))
or args[index - 1].casefold() in ('-icdb', '-project', '-p'))):
candidates.append(value)
normalized_roots = [ntpath.normcase(ntpath.normpath(str(root))) for root in roots]
expected_library = ntpath.normcase(ntpath.normpath(str(library)))
for candidate in candidates:
path = ntpath.normcase(ntpath.normpath(candidate))
if path == expected_library:
return True
for root in normalized_roots:
try:
if ntpath.commonpath((path, root)) == root:
return True
except ValueError:
pass
return False if candidates else None
def _owner_is_blocking(item, library, roots):
"""Unknown external clients need resource proof, not a global exit rule."""
scope = _owner_matches_scope(item, library, roots)
if scope is not None:
return scope
command = str(item.get('command') or '').casefold()
flags = ('--verify-worker', '--repair-worker', '--management-worker', '--native-backup-worker')
if any(flag in command for flag in flags):
return True # An unreadable in-flight request is not proof of another library.
from .failed_session import _command_args
try:
return any(Path(arg).name.casefold() in ('close_failed_block.vbs', 'save_verified_block.vbs')
for arg in _command_args(command)[1:])
except (OSError, ValueError):
return False
def _enter_maintenance(lease):
"""Transfer only a valid known lease, after project admission under lock."""
import win32con, win32event, win32file
@@ -266,13 +417,16 @@ def _enter_maintenance(lease):
raise LibraryBusy('本进程库锁仍在使用或检查点身份不符')
handle = existing['handle']
else:
if record is not None and record['pid'] != os.getpid() and _supervisor_alive(record['pid']):
if record is not None and record['pid'] != os.getpid() and _recorded_supervisor_alive(record) is True:
raise LibraryBusy('原监督进程仍存在,不能接续其库锁')
handle = win32file.CreateFile(str(lease.path), win32con.GENERIC_READ | win32con.GENERIC_WRITE,
0, None, win32con.OPEN_ALWAYS, win32con.FILE_ATTRIBUTE_NORMAL, None)
# All target/process/address/exclusive-file checks finish BEFORE
# any old checkpoint replacement. Keep the exact sidecar open.
from .native_export_guard import prove_export_copies_released
if record is not None and _recorded_supervisor_alive(record) is None:
from .maintenance_entry import assert_native_idle
assert_native_idle()
with prove_export_copies_released(lease.library), lease.maintenance_guard() as proof:
if (marker.read_bytes() if marker.exists() else None) != old:
raise LibraryBusy('维护检查期间检查点变化')
@@ -285,7 +439,8 @@ def _enter_maintenance(lease):
lease_id = uuid4().hex
_persist_blocker(lease.library, 'OPERATION_IN_PROGRESS', lease_id=lease_id)
_owners[lease.key] = {'thread': threading.get_ident(), 'depth': 1, 'handle': handle,
'retained': False, 'lease_id': lease_id}
'retained': False, 'lease_id': lease_id, 'checkpoint': True,
'release_probe': lease.safe_to_release, 'block_name': None}
lease.maintenance_proof = proof
lease.entered = True
lease.maintenance_mutex = mutex
@@ -315,16 +470,20 @@ def owned_snapshot(path):
@contextmanager
def _prove_library_released(library, block_name=None):
def _prove_library_released(library, block_name=None, *, allow_unknown_checkpoint=False):
"""Hold all native database read handles during explicit lease release."""
import stat
import win32con, win32file
from .owner_probe import query_owners
from .scoped_manifest import scan_tree
from .native_management import _retained_workers
from .native_export_guard import assert_retained_exports_idle, prove_export_copies_released
from .native_engineering_identity import _OWNER_NAMES
assert_retained_exports_idle()
if any(worker.poll() is None for worker in _retained_workers):
library = Path(library).resolve()
assert_retained_exports_idle(library)
if any(worker.poll() is None and (allow_unknown_checkpoint
or not getattr(worker, 'cbb_library', None) or _key(worker.cbb_library) == _key(library))
for worker in _retained_workers):
raise LibraryBusy('原生管理工作进程尚未返回,不能解除协调阻止。')
from .com_runtime import initialize_com, uninitialize_com
initialized = initialize_com()
@@ -334,56 +493,82 @@ def _prove_library_released(library, block_name=None):
'cscript.exe', 'wscript.exe', 'python.exe', 'pythonw.exe',
'CbbRefreshVerify.exe', *sorted(_OWNER_NAMES))
scope_root = (Path(library).resolve().parent/'ReusableBlocksLP'/block_name/'Work').resolve() if block_name else None
shared_root = Path(library).resolve().parent/'SymbolLibs'
if allow_unknown_checkpoint and scope_root is not None:
raise LibraryBusy('未知检查点必须执行完整库释放证明。')
observed = []
def check_owners():
owners = []
observed.clear()
from .session_record import active_write_records
if active_write_records(library):
raise LibraryBusy('本库原生调用或助手尚未确认返回,请等待当前操作结束。')
for item in query_owners(names):
if item['pid'] == os.getpid():
continue
if scope_root is not None and item.get('command'):
command = str(item['command']).casefold().replace('/', '\\')
if str(scope_root).casefold().replace('/', '\\') not in command:
if not allow_unknown_checkpoint:
roots = (scope_root, shared_root) if scope_root is not None else (Path(library).parent,)
if not _owner_is_blocking(item, library, roots):
if _owner_matches_scope(item, library, roots) is None:
observed.append(item)
continue
automation = item['name'].casefold() in ('python.exe', 'pythonw.exe', 'cbbrefreshverify.exe')
command = item['command'].casefold()
if not automation or not command or any(flag in command for flag in (
'--verify-worker', '--repair-worker', '--management-worker')):
'--verify-worker', '--repair-worker', '--management-worker', '--native-backup-worker')):
owners.append(item)
if owners:
raise LibraryBusy(f'原生进程仍存在,不能解除协调阻止:{owners}')
return owners
check_owners()
entries = scan_tree(Path(library).parent)
if scope_root is None:
entries = scan_tree(Path(library).parent)
else:
entries = {Path(library): Path(library).lstat()}
for root in (scope_root, shared_root):
if root.exists():
entries.update(scan_tree(root))
addresses = [str(path) for path in entries if path.name.casefold() == 'saddress.adr'
and (scope_root is None or path.resolve().is_relative_to(scope_root))]
and (scope_root is None or path.is_relative_to(scope_root)
or path.is_relative_to(shared_root))]
if addresses:
raise LibraryBusy(f'库中仍有数据库地址记录,不能自动解除:{addresses}')
native = [path for path in entries if path.is_file() and (
path.suffix.casefold() == '.lmc'
# scan_tree already rejects links; only native candidates need file opens.
native = [path for path, metadata in entries.items() if stat.S_ISREG(metadata.st_mode) and (
path == library
or (path.name.casefold() in ('icdb.dat', 'layoutdb.lyt')
and (scope_root is None or path.resolve().is_relative_to(scope_root))))]
if Path(library).resolve() not in native:
and (scope_root is None or path.is_relative_to(scope_root)
or path.is_relative_to(shared_root))))]
if library not in native:
raise LibraryBusy('目标库身份缺失,不能解除协调阻止。')
with ExitStack() as stack:
export_copies = stack.enter_context(prove_export_copies_released(library))
export_proof = (prove_export_copies_released(library, allow_unknown_checkpoint=True)
if allow_unknown_checkpoint else prove_export_copies_released(library))
export_copies = stack.enter_context(export_proof)
for path in native:
handle = win32file.CreateFile(str(path), win32con.GENERIC_READ, 0, None,
win32con.OPEN_EXISTING, win32con.FILE_ATTRIBUTE_NORMAL, None)
stack.callback(handle.Close)
owners = check_owners()
yield {'native_owners': owners, 'addresses': addresses, 'export_copies': export_copies,
yield {'native_owners': owners, 'observed_owners': list(observed),
'addresses': addresses, 'export_copies': export_copies,
'unknown_checkpoint_recovery': allow_unknown_checkpoint,
'exclusive_files': [str(path) for path in native]}
finally:
uninitialize_com(initialized)
def release_retained_library(library, evidence_dir, *, block_name=None):
"""Explicit user-requested release, after fresh owner/address/access proof.
"""Reconcile historical leases after fresh owner/address/access proof.
The upload sidecar is never deleted. This acknowledges process release,
not the correctness of a partially changed library or its recovery state.
"""
import win32con, win32event, win32file
library = Path(library).resolve(strict=True)
from .session_record import (retained_session_snapshot, acknowledge_library_release,
reconciled_retained_sessions)
retained = retained_session_snapshot(library)
evidence = Path(evidence_dir).resolve()
if evidence.is_relative_to(library.parent) or library.parent.is_relative_to(evidence):
raise ValueError('释放检查证据必须位于目标库之外。')
@@ -405,17 +590,46 @@ def release_retained_library(library, evidence_dir, *, block_name=None):
win32con.GENERIC_READ | win32con.GENERIC_WRITE, 0, None,
win32con.OPEN_ALWAYS, win32con.FILE_ATTRIBUTE_NORMAL, None)
except Exception as error:
if getattr(error, 'winerror', None) not in (32, 33):
raise
raise LibraryBusy('其他进程仍持有库锁,不能解除。') from error
proof_context = (_prove_library_released(library, block_name=block_name)
if block_name else _prove_library_released(library))
marker = blocker_path(library)
checkpoints = (marker, marker.with_suffix('.pending'))
before = {path: path.read_bytes() if path.exists() else None for path in checkpoints}
unknown = any(data is not None and checkpoint_record(path) is None for path, data in before.items())
records = [checkpoint_record(path) for path, data in before.items() if data is not None]
scoped = (block_name if block_name and all(record and record.get('block_name') == block_name
for record in records) and (owner is None or owner.get('block_name') == block_name) else None)
reconciled = [{'kind': 'checkpoint', 'path': str(path), 'recognized': metadata is not None,
**{key: (metadata or {}).get(key) for key in ('block_name', 'lease_id', 'pid', 'status', 'time')}}
for path, data in before.items() if data is not None
for metadata in (checkpoint_record(path),)]
if owner:
reconciled.append({'kind': 'retained_owner', 'path': str(lease.path),
'block_name': owner.get('block_name'), 'lease_id': owner['lease_id'],
'status': 'OWNER_RELEASE_UNCONFIRMED'})
foreign = sorted({item['block_name'] for item in reconciled
if block_name and item.get('block_name') and item['block_name'] != block_name})
proof_context = (_prove_library_released(library, allow_unknown_checkpoint=True) if unknown else
_prove_library_released(library, block_name=scoped)
if scoped else _prove_library_released(library))
with proof_context as proof:
marker = blocker_path(library)
checkpoints = (marker, marker.with_suffix('.pending'))
if any(path.exists() and checkpoint_record(path) is None for path in checkpoints):
raise LibraryBusy('协调检查点格式不明或已损坏,保留原文件供人工审查。')
if unknown and proof.get('unknown_checkpoint_recovery') is not True:
raise LibraryBusy('未知检查点缺少完整库释放证明,保留原文件。')
if any((path.read_bytes() if path.exists() else None) != data for path, data in before.items()):
raise LibraryBusy('释放检查期间协调检查点变化,保留原文件。')
root = evidence/('release-' + uuid4().hex)
root.mkdir(parents=True, exist_ok=False)
reconciled_sessions = reconciled_retained_sessions(retained, block_name=scoped)
reconciled_blocks = sorted({item['block_name'] for item in (*reconciled, *reconciled_sessions)
if item.get('block_name')})
record = {'library': str(library), 'status': 'RELEASE_CONFIRMED', **proof,
'requested_block_name': block_name, 'proof_scope': 'BLOCK' if scoped else 'LIBRARY',
'proof_block_name': scoped, 'reconciled_checkpoints': reconciled,
'reconciled_foreign_blockers': sorted(set(foreign) | {
block for block in reconciled_blocks if block_name and block != block_name}),
'reconciled_blocks': reconciled_blocks, 'reconciled_sessions': reconciled_sessions,
'unknown_checkpoint_archived': unknown,
'data_recovery_confirmed': False, 'evidence_path': str(root/'release.json')}
(root/'release.json').write_text(json.dumps(record, ensure_ascii=False, indent=2), encoding='utf-8')
for index, checkpoint in enumerate(checkpoints):
@@ -429,6 +643,7 @@ def release_retained_library(library, evidence_dir, *, block_name=None):
if owner:
owner['handle'].Close()
del _owners[lease.key]
acknowledge_library_release(record, retained, block_name=scoped)
return record
finally:
if external is not None:
+8 -41
View File
@@ -284,9 +284,8 @@ def scan_open_library_status(library: Path, names, *, tree_image_reader=None,
image_status_map=image_status_map)
status = reconcile_statuses(tree_statuses, fallback, names)
uncalibrated = [name for name in names if status.get(name, 'UNKNOWN') == 'UNKNOWN']
# The UI must never use UNKNOWN as a skip signal. Queue it as Pending
# so it receives the real Verify flow instead of being silently omitted.
effective = {name: ('PENDING' if value == 'UNKNOWN' else value)
# Missing observations remain unknown; eligibility is decided by the caller.
effective = {name: (value if tree_images.get(name) is not None else 'UNKNOWN')
for name, value in status.items()}
logger.info(
'library_status_sources names=%s image_histogram=%s current_image_map=%s cached_image_map=%s image_status_map=%s cache_actions=%s filesystem=%s calibrated_tree=%s final=%s uncalibrated_sample=%s conflicts=%s ready_anchor_ids=%s unseen_tree_ids=%s',
@@ -328,46 +327,14 @@ def scan_open_library_status(library: Path, names, *, tree_image_reader=None,
return effective
try:
from pywinauto import Desktop
from pywinauto.controls.common_controls import TreeViewWrapper
from cbb_refresh_verify.mentor_adapter import _tree_image, same_path
target = None
for window in Desktop(backend='win32').windows():
title = window.window_text()
if title.startswith('Library Manager:') and same_path(title.split(':', 1)[1].strip(), library):
target = window
break
if target is None:
if tree_image_reader is None:
from cbb_refresh_verify.mentor_adapter import read_library_tree_image_map
tree_image_reader = read_library_tree_image_map
return finish(tree_image_reader(library, names))
trees = [w for w in target.descendants() if 'TreeView' in w.class_name()]
if len(trees) != 1:
return finish({})
tree = TreeViewWrapper(trees[0].handle)
root_name = Path(library).name
images = {}
missing = []
for name in names:
try:
item = tree.get_item(['Central Libraries', root_name, 'Reusable Blocks', 'Logical-Physical', name], exact=True)
from cbb_refresh_verify.mentor_adapter import _tree_image_details
details = _tree_image_details(item, tree)
images[name] = details['image_id']
logger.info('library_tree_image_item name=%s image_id=%s selected_image_id=%s state_image_id=%s image_color=%s selected_image_color=%s state_image_color=%s',
name, details['image_id'], details['selected_image_id'],
details['state_image_id'], details['image_color'],
details['selected_image_color'], details['state_image_color'])
except Exception:
images[name] = None
missing.append(name)
logger.info('library_tree_image_probe requested=%s read=%s missing=%s',
len(names), len(names) - len(missing), missing[:20])
return finish(images)
if tree_image_reader is None:
from .mentor_adapter import read_library_tree_image_map
tree_image_reader = read_library_tree_image_map
return finish(tree_image_reader(library, names))
except Exception as error:
logger.warning('library_status_probe_failed path=%s error=%s', library, error)
return finish({})
return {name: 'UNKNOWN' for name in names}
def read_library(path: Path, *, dispatch=None):
+64 -67
View File
@@ -15,12 +15,29 @@ from uuid import uuid4
def assert_native_idle(target=None):
from .owner_probe import query_owners
from .failed_session import _command_args
from .native_management import _retained_workers
from .native_export_guard import assert_retained_exports_idle
from .native_engineering_identity import _OWNER_NAMES
assert_retained_exports_idle()
if any(worker.poll() is None for worker in _retained_workers):
from .library_lock import _key, _owner_matches_scope
library = None
roots = ()
if target is not None:
from .backup_recovery import _project_text
roots = (Path(target.project).parent, Path(target.database))
try:
lines = re.findall(r'^KEY CentralLibrary\b[^\r\n]*', _project_text(target.project), re.M)
references = [re.fullmatch(r'KEY CentralLibrary "([^"\r\n]+)"\s*', line) for line in lines]
if not references or any(match is None for match in references):
raise ValueError('CentralLibrary references are missing or invalid.')
libraries = {(Path(target.project).parent/match[1]).resolve(strict=True) for match in references}
if len(libraries) == 1 and all(path.suffix.casefold() == '.lmc' and path.is_file() for path in libraries):
library = libraries.pop()
roots += (library.parent/'SymbolLibs',)
except (OSError, ValueError):
pass # Missing/ambiguous original references cannot narrow retained ownership.
assert_retained_exports_idle(library)
if any(worker.poll() is None and (library is None or not getattr(worker, 'cbb_library', None)
or _key(worker.cbb_library) == _key(library)) for worker in _retained_workers):
raise ValueError('MAINTENANCE_OWNER_ACTIVE: 管理工作进程尚未返回')
names = ('ExpeditionPCB.exe', 'LibraryManager.exe', 'viewdraw.exe', 'iCDBProjectBackup.exe',
'ReuseBlockImporter.exe', 'CleanWGLib.exe', 'cscript.exe', 'wscript.exe',
@@ -32,18 +49,10 @@ def assert_native_idle(target=None):
name, command = item['name'].casefold(), item['command'].casefold()
automation = name in ('python.exe', 'pythonw.exe', 'cbbrefreshverify.exe')
if not automation or not command or any(flag in command for flag in (
'--verify-worker', '--repair-worker', '--management-worker')):
'--verify-worker', '--repair-worker', '--management-worker', '--native-backup-worker')):
owners.append(item)
if target is not None:
expected = os.path.normcase(os.path.abspath(str(target.project)))
scoped = []
for item in owners:
command = str(item.get('command') or '')
if not command:
scoped.append(item)
elif expected in [os.path.normcase(os.path.abspath(arg)) for arg in _command_args(command)[1:]]:
scoped.append(item)
owners = scoped
owners = [item for item in owners if _owner_matches_scope(item, library, roots) is not False]
if owners:
raise ValueError(f'MAINTENANCE_OWNER_ACTIVE: 请正常关闭原生程序/操作助手:{owners}')
@@ -215,13 +224,9 @@ def maintenance_target_guard(request, expected=None, *, stop=None):
def run_icdb_check(request, *, progress=None, stop=None):
"""Inspect through Project Backup; never launch PCB, Repair or Restore."""
from .verify_runner import VerifyResult
from .library_lock import LibraryLock, LibraryBusy
from .native_repair import run_native_repair, RepairTarget
from .native_backup import native_backup_project
from .backup_recovery import _project_text
from .native_occupancy import confirmed_inspection
"""Open PCB diagnostically and classify this attempt's native error content."""
from .verify_runner import VerifyResult, run_verify
from .icdb_recovery import icdb_error_content
root = Path(request.evidence_dir).resolve()/('icdb-check-'+uuid4().hex[:12])
library_root = Path(request.lmc_file).resolve().parent
if root.is_relative_to(library_root) or library_root.is_relative_to(root):
@@ -229,10 +234,7 @@ def run_icdb_check(request, *, progress=None, stop=None):
stderr='检查记录目录需位于中心库之外。')
record = {'operation': 'icdb_batch_check', 'automatic_repair': False,
'library': str(request.lmc_file), 'block': request.block_name}
released = True
def check():
if stop is not None and stop.is_set():
raise InterruptedError('已取消检查。')
started = False
def save(result):
report = root/'result.json'
recorded = replace(result, evidence_path=report)
@@ -245,51 +247,46 @@ def run_icdb_check(request, *, progress=None, stop=None):
return result
return recorded
try:
check()
root.mkdir(parents=True, exist_ok=False)
with LibraryLock(request.lmc_file, checkpoint=False, safe_to_release=lambda: released):
project = native_backup_project(request.lmc_file, request.block_name)
values = re.findall(r'^KEY iCDBDir "([^"\r\n]*)"', _project_text(project), re.M)
if (len(values) != 1 or not values[0] or PureWindowsPath(values[0]).drive
or PureWindowsPath(values[0]).root
or (project.parent/values[0]).resolve() != (project.parent/'database').resolve()):
raise ValueError('无法确认工程的 ICDB 数据库路径。')
target = RepairTarget(project, project.parent/'database')
record['target'] = asdict(target)
check()
if progress:
progress({'stage': 'icdb_ownership_check', 'block': request.block_name,
'detail': '正在检查 ICDB 状态'})
released = False
inspection = run_native_repair(target, root/'native-inspection',
mentor_bin=request.mentor_bin, timeout_seconds=min(30, request.timeout_seconds),
progress=progress, inspection_only=True, stop=stop)
record['native_inspection'] = inspection
released = (inspection.get('release', {}).get('native_exited') is True
or inspection.get('native_started') is False
and inspection.get('native_pid') is None and inspection.get('worker_pid') is None)
if ((stop is not None and stop.is_set())
or inspection.get('status') in ('CANCELLED', 'CANCELLED_UNRELEASED')):
return save(VerifyResult('CANCELLED', safe_to_continue=released, stderr='已取消 ICDB 检查。'))
if not confirmed_inspection(target, inspection):
return save(VerifyResult('FAIL', safe_to_continue=released,
error_category='ICDB_INSPECTION_UNCONFIRMED',
stderr='无法确认 ICDB 状态。' + str(inspection.get('detail', ''))))
if inspection['status'] == 'IN_USE':
return save(VerifyResult('BLOCKED', safe_to_continue=released,
error_category='ICDB_DATABASE_IN_USE', stderr='工程正在使用,暂未完成 ICDB 检查。'))
if inspection['repair_required']:
return save(VerifyResult('FAIL', safe_to_continue=released,
error_category='ICDB_REPAIR_REQUIRED', stderr='发现 ICDB 问题,未执行修复。'))
return save(VerifyResult('CHECKED', safe_to_continue=released,
stdout='ICDB 检查未发现异常,无需修复。'))
except InterruptedError:
return save(VerifyResult('CANCELLED', safe_to_continue=released, stderr='已取消 ICDB 检查。'))
except LibraryBusy as error:
return save(VerifyResult('BLOCKED', safe_to_continue=False,
error_category='LIBRARY_IN_USE', stderr=str(error)))
if stop is not None and stop.is_set():
return save(VerifyResult('CANCELLED', stderr='已取消 ICDB 检查。'))
if progress:
progress({'stage': 'icdb_startup_check', 'block': request.block_name,
'detail': '正在打开工程并检查本次错误信息'})
started = True
startup = run_verify(replace(request, maintenance=False, auto_repair=False,
diagnostic_only=True, native_lm_verify=False, post_repair_proof=None,
evidence_dir=root/'pcb-open'), progress=progress, stop=stop)
record['native_open'] = asdict(startup)
if startup.status == 'CANCELLED' or stop is not None and stop.is_set():
return save(replace(startup, status='CANCELLED', recovery_attempts=()))
if startup.status == 'PASS' and startup.safe_to_continue:
return save(replace(startup, status='CHECKED', library_flag='UNKNOWN',
stdout='工程打开正常,未发现 ICDB 错误;未执行 CBB 校验。', recovery_attempts=()))
content = icdb_error_content(startup.stderr, error_message=True)
if (startup.evidence_path
and startup.evidence_path.resolve().is_relative_to(root/'pcb-open')):
try:
saved = json.loads(startup.evidence_path.with_name('diagnostics').joinpath(
'icdb-error-content.json').read_text('utf-8'))
native = icdb_error_content(saved.get('native_message', ''))
if native:
content = '\n'.join(dict.fromkeys(filter(None, (content, native))))
record['native_error'] = saved
except (OSError, ValueError, AttributeError):
pass
if content:
return save(replace(startup, status='FAIL', library_flag='UNKNOWN',
error_category='ICDB_REPAIR_REQUIRED',
stderr='发现 ICDB 错误,未执行修复。\n'+content, recovery_attempts=()))
category = startup.error_category
if not category or category.startswith('ICDB_'):
category = 'ICDB_INSPECTION_UNCONFIRMED'
return save(replace(startup, status='FAIL' if startup.status=='PASS' else startup.status,
error_category=category, library_flag='UNKNOWN', recovery_attempts=(),
stderr=startup.stderr or '工程打开或释放未完成,尚无法确认 ICDB 状态。'))
except Exception as error:
return save(VerifyResult('FAIL', safe_to_continue=released,
return save(VerifyResult('FAIL', safe_to_continue=not started,
error_category='ICDB_INSPECTION_UNCONFIRMED', stderr=f'无法确认 ICDB 状态:{error}'))
+131 -30
View File
@@ -8,7 +8,7 @@ from uuid import uuid4
from .library_lock import LibraryLock, LibraryBusy
from .scoped_manifest import (ManifestScope, build_manifest, build_exclusive_library_manifest,
copy_verified, hash_file, scan_tree)
comparison_rules, copy_verified, hash_file, scan_tree)
@dataclass(frozen=True)
@@ -31,6 +31,7 @@ class ManagementRequest:
mapped_drive: str = ''
mapped_remote_root: str = ''
transaction_token: str = ''
check_previous_release: bool = False
@dataclass(frozen=True)
@@ -71,6 +72,8 @@ class ManagementResult:
safe_to_continue: bool
detail: str = ''
evidence_path: Path | None = None
release_proof: dict | None = None
change_report: dict | None = None
def validate_name(value):
@@ -150,14 +153,23 @@ def unfinished_operations(root):
for path in sorted(Path(root).glob('*/checkpoint.json')):
try:
record = json.loads(path.read_text(encoding='utf-8'))
if record.get('stage') != 'complete' or record.get('safe_to_continue') is False:
if (record.get('stage') != 'complete' or record.get('safe_to_continue') is False
or record.get('status') == 'RECOVERY_REQUIRED'):
results.append({**record, 'evidence_path': str(path)})
except (OSError, ValueError):
results.append({'status': 'CHECKPOINT_UNREADABLE', 'evidence_path': str(path)})
return results
def run_management(request, *, adapter=None, verifier=None, idle_check=None, progress=None):
def _may_have_mutated(record):
prewrite = {'preflight', 'release_check', 'catalog_before', 'baseline', 'stage_source', 'prepare_identity'}
history = record.get('history')
return (record.get('stage') not in prewrite | {'complete'} or record.get('status') == 'RECOVERY_REQUIRED'
or not isinstance(history, list) or not history
or any(not isinstance(item, dict) or item.get('stage') not in prewrite for item in history))
def run_management(request, *, adapter=None, verifier=None, idle_check=None, progress=None, stop=None, version_prompt=None):
from .verify_runner import VerifyRequest, run_verify
from .backup_recovery import _idle_check
import win32con, win32event
@@ -174,6 +186,21 @@ def run_management(request, *, adapter=None, verifier=None, idle_check=None, pro
owner_safe = True
mutex = None
acquired = False
release_proof = None
original = None
change_report = None
def invoke(callback, *args):
nonlocal owner_safe
if stop is not None and stop.is_set():
raise InterruptedError('MANAGEMENT_CANCELLED: 已停止后续原生操作')
owner_safe = False
try:
result = callback(*args)
except Exception as error:
owner_safe = getattr(error, 'safe_to_continue', False) is True
raise
owner_safe = result.safe_to_continue
return result
def checkpoint(next_stage, **details):
nonlocal stage
stage = next_stage
@@ -184,14 +211,46 @@ def run_management(request, *, adapter=None, verifier=None, idle_check=None, pro
if progress:
progress({'stage': stage, 'block': request.name, 'operation': request.operation, **details})
def finish(status, detail='', safe=True):
nonlocal change_report
if mutated and original is not None and change_report is None:
change_report = {'status': 'UNCONFIRMED', 'scope': original['scope'],
**comparison_rules(scope),
'unrelated_cbb_content_checked': False,
'detail': '原生会话尚未释放,未读取写后内容'}
if safe:
try:
after = build_exclusive_library_manifest(library_root, scope=scope, progress=progress)
_write_json(root/'library-after-manifest.json', after)
before_files, after_files = original['files'], after['files']
change_report.update(status='RECORDED', detail='目标工程及共享依赖的内容差分',
added=sorted(after_files.keys()-before_files.keys()),
deleted=sorted(before_files.keys()-after_files.keys()),
modified=sorted(path for path in before_files.keys() & after_files.keys()
if before_files[path] != after_files[path]))
except Exception as error:
change_report['detail'] = f'写后内容未确认:{error}'
try:
_write_json(root/'library-changes.json', change_report)
except OSError as error:
change_report.update(status='UNCONFIRMED', evidence_error=str(error),
detail=f'变更记录保存失败:{error}')
if change_report['status'] != 'RECORDED':
detail = (detail+'\n'+change_report['detail']).strip()
result = ManagementResult(request.operation, request.library, request.name, request.partition,
request.source, request.auto_verify, 'complete', status, safe, detail, report)
request.source, request.auto_verify, 'complete', status, safe, detail, report, release_proof, change_report)
if report:
_write_json(report, {**asdict(result), 'history': history, 'backup_policy': 'manual'})
try:
_write_json(report, {**asdict(result), 'history': history, 'backup_policy': 'manual'})
except OSError as error:
result = replace(result, detail=f'{detail}\n操作记录写入失败:{error}'.strip())
return result
from .com_runtime import initialize_com, uninitialize_com
initialized = initialize_com()
try:
if stop is not None and stop.is_set():
return finish('CANCELLED', '已取消,未启动原生管理操作')
if hasattr(adapter, 'bind_stop'):
adapter.bind_stop(stop)
if request.operation not in ('add', 'update', 'remove'):
raise ValueError('不支持的管理操作。')
validate_name(request.name); validate_name(request.partition)
@@ -207,9 +266,13 @@ def run_management(request, *, adapter=None, verifier=None, idle_check=None, pro
capability = adapter.capabilities().get(request.operation, Capability(False, '未验证的原生接口'))
if not capability.supported:
return finish('UNSUPPORTED', capability.reason)
unfinished_target = None
for unfinished in unfinished_operations(evidence):
if unfinished.get('library') and Path(unfinished['library']).resolve() == canonical_library:
return finish('BLOCKED', f'存在未完成/需恢复任务,禁止自动重放:{unfinished["evidence_path"]}', False)
if (unfinished.get('library') and Path(unfinished['library']).resolve() == canonical_library
and _may_have_mutated(unfinished) and (not unfinished.get('name')
or str(unfinished['name']).casefold() == request.name.casefold())):
unfinished_target = unfinished
break
source = pcb = None
if request.operation in ('add', 'update'):
if request.source is None:
@@ -219,16 +282,26 @@ def run_management(request, *, adapter=None, verifier=None, idle_check=None, pro
or evidence.is_relative_to(source.parent)
or source.parent.is_relative_to(evidence)):
raise ValueError('来源工程、目标库与操作证据必须使用独立目录。')
root = evidence/('management-' + uuid4().hex)
root.mkdir(parents=True, exist_ok=False)
report = root/'checkpoint.json'
checkpoint('release_check')
mutex = win32event.CreateMutex(None, False, r'Local\CbbRefreshVerify795')
acquired = win32event.WaitForSingleObject(mutex, 0) in (win32con.WAIT_OBJECT_0, win32con.WAIT_ABANDONED)
if not acquired:
return finish('BLOCKED', '已有 CBB 校验/管理/恢复任务运行。', False)
return finish('BLOCKED', '已有 CBB 校验/管理/恢复任务运行;本次未启动。')
from .library_lock import recheck_previous_release
release_proof = recheck_previous_release(library, root/'previous-release', block_name=request.name,
required=request.check_previous_release or bool(unfinished_target
and unfinished_target.get('safe_to_continue') is not True), progress=progress)
if release_proof is not None:
checkpoint('release_check', release_proof=release_proof)
if unfinished_target:
return finish('BLOCKED', f'存在未完成/需恢复任务,禁止自动重放:{unfinished_target["evidence_path"]}')
if hasattr(adapter, 'bind_library_mapping'):
adapter.bind_library_mapping(request.mapped_drive, request.mapped_remote_root)
with LibraryLock(library, safe_to_release=lambda: owner_safe):
root = evidence/('management-' + uuid4().hex)
root.mkdir(parents=True, exist_ok=False)
report = root/'checkpoint.json'
with LibraryLock(library, block_name=request.name, safe_to_release=lambda: owner_safe or (
callable(getattr(adapter, 'release_confirmed', None)) and adapter.release_confirmed(library))):
work = library.parent/'ReusableBlocksLP'/request.name/'Work'
target_project, target_pcb = work/(request.name+'.prj'), work/'PCB/pcb.pcb'
if request.operation != 'add':
@@ -244,7 +317,7 @@ def run_management(request, *, adapter=None, verifier=None, idle_check=None, pro
idle_check(library, source, pcb)
checkpoint('catalog_before')
owner_safe = False
before = adapter.catalog(library, root/'before')
before = invoke(adapter.catalog, library, root/'before')
owner_safe = before.safe_to_continue
if not owner_safe or Path(before.library).resolve() != canonical_library or before.flow != 1:
return finish('BLOCKED', '原生目录身份/流程或所有者释放未确认。仅支持已验证 DxDesigner 流程。', owner_safe)
@@ -283,34 +356,36 @@ def run_management(request, *, adapter=None, verifier=None, idle_check=None, pro
if build_exclusive_library_manifest(library_root, scope=scope, progress=progress) != original:
raise OSError('内容基线记录后中心库发生变化,停止写入。')
if request.operation in ('remove', 'update'):
if stop is not None and stop.is_set():
return finish('CANCELLED', '已取消,未开始移除', owner_safe)
checkpoint('native_remove')
mutated = True; owner_safe = False
native = adapter.remove(request, root/'remove')
native = invoke(adapter.remove, request, root/'remove')
owner_safe = native.safe_to_continue
if not owner_safe or not native.success:
return finish('RECOVERY_REQUIRED', native.detail, False)
owner_safe = False
return finish('RECOVERY_REQUIRED', native.detail, owner_safe)
idle_check(library, target_project, target_pcb)
owner_safe = False
after = adapter.catalog(library, root/'after-remove')
after = invoke(adapter.catalog, library, root/'after-remove')
owner_safe = after.safe_to_continue
if (not owner_safe or Path(after.library).resolve() != canonical_library or after.flow != 1
or any(e.name == request.name and e.partition == request.partition for e in after.entries)):
return finish('RECOVERY_REQUIRED', '原生移除后目录未确认目标缺失。', False)
return finish('RECOVERY_REQUIRED', '原生移除后目录未确认目标缺失。', owner_safe)
removed = True
checkpoint('removed')
if request.operation == 'remove':
return finish('REMOVED', '已由原生操作移除,重新读取目录确认不存在;应用未创建备份。')
if stop is not None and stop.is_set():
return finish('RECOVERY_REQUIRED' if mutated else 'CANCELLED',
'已停止后续导入;已发生的原生变更需要核对', owner_safe)
checkpoint('native_add')
mutated = True; owner_safe = False
native = adapter.add(staged, root/'add')
native = invoke(adapter.add, staged, root/'add')
owner_safe = native.safe_to_continue
if not owner_safe or not native.success:
return finish('RECOVERY_REQUIRED' if removed or not owner_safe else 'UPLOAD_FAILED',
native.detail, owner_safe and not removed)
owner_safe = False
native.detail, owner_safe)
idle_check(library, target_project, target_pcb)
after = adapter.catalog(library, root/'after-add')
after = invoke(adapter.catalog, library, root/'after-add')
owner_safe = after.safe_to_continue
matches = [entry for entry in after.entries if entry.name == request.name
and entry.partition == request.partition]
@@ -319,7 +394,7 @@ def run_management(request, *, adapter=None, verifier=None, idle_check=None, pro
len(matches) == 1 and matches[0].engineering_identity == identity)
if (not owner_safe or Path(after.library).resolve() != canonical_library or after.flow != 1
or len(matches) != 1 or not proof_matches):
return finish('RECOVERY_REQUIRED', '原生新增后名称、分区或来源工程内容未确认一致。', False)
return finish('RECOVERY_REQUIRED', '原生新增后名称、分区或来源工程内容未确认一致。', owner_safe)
uploaded = True
checkpoint('uploaded')
if not request.auto_verify:
@@ -327,8 +402,10 @@ def run_management(request, *, adapter=None, verifier=None, idle_check=None, pro
# Adapter catalog cannot return until its exact native owner exits.
checkpoint('verify')
owner_safe = False
options = {'version_prompt': version_prompt} if version_prompt is not None else {}
verified = verifier(VerifyRequest(library, request.name, root/'verify',
mapped_drive=request.mapped_drive, mapped_remote_root=request.mapped_remote_root), progress=progress)
mapped_drive=request.mapped_drive, mapped_remote_root=request.mapped_remote_root),
progress=progress, stop=stop, **options)
owner_safe = verified.safe_to_continue
checkpoint('verified', verification=asdict(verified))
good = verified.status == 'PASS' and verified.library_flag == 'GREEN' and owner_safe
@@ -336,12 +413,14 @@ def run_management(request, *, adapter=None, verifier=None, idle_check=None, pro
verified.stderr or verified.library_flag_detail, owner_safe)
except Exception as error:
if uploaded:
return finish('UPLOADED_VERIFY_FAILED', str(error), False)
return finish('UPLOADED_VERIFY_FAILED', str(error), owner_safe)
status = 'RECOVERY_REQUIRED' if mutated else ('UPLOAD_FAILED' if request.operation in ('add', 'update') else 'BLOCKED')
# Bad input is never described as a native upload attempt.
if root is None and isinstance(error, ValueError):
status = 'BLOCKED'
return finish(status, str(error), not mutated and owner_safe and not isinstance(error, LibraryBusy))
if stage == 'release_check' and not mutated:
status = 'BLOCKED'
return finish(status, str(error), owner_safe)
finally:
if mutex is not None:
if acquired:
@@ -350,16 +429,38 @@ def run_management(request, *, adapter=None, verifier=None, idle_check=None, pro
uninitialize_com(initialized)
def run_management_batch(requests, *, stop=None, progress=None, **kwargs):
def run_management_batch(requests, *, stop=None, progress=None, item_executor=None, **kwargs):
results = []
retained_libraries = set()
for request in requests:
if stop is not None and stop.is_set():
break
result = run_management(request, progress=progress, **kwargs)
key = str(request.library).casefold()
if key in retained_libraries:
result = ManagementResult(request.operation, request.library, request.name, request.partition,
request.source, request.auto_verify, 'complete', 'SKIPPED', True,
'同库原生会话尚未释放,未启动新的写入')
results.append(result)
if progress:
progress({'stage': result.status, 'block': result.name, 'detail': result.detail})
continue
def action():
result = run_management(request, progress=progress, stop=stop, **kwargs)
if not result.safe_to_continue:
from .operation_lifecycle import recheck_failed_release
result = recheck_failed_release(request.library, request.name, result)
return result
from .operation_queue import QueueCancelled
try:
result = item_executor(request, action) if item_executor is not None else action()
except QueueCancelled as error:
results.append(ManagementResult(request.operation, request.library, request.name, request.partition,
request.source, request.auto_verify, 'complete', 'CANCELLED', True, str(error)))
break
results.append(result)
if progress:
progress({'stage': result.status, 'block': result.name, 'detail': result.detail,
'evidence_path': str(result.evidence_path or '')})
if not result.safe_to_continue:
break
retained_libraries.add(key)
return results
+130 -211
View File
@@ -236,12 +236,13 @@ def pcb_processes():
def is_reusable_circuit_pcb(exe, command, expected_pcb):
"""Match the ReusableCircuit session for this CBB by argument, not install layout."""
from .failed_session import _command_args
exe = os.path.normcase(str(exe or ''))
command = os.path.normcase(str(command or ''))
args = _command_args(str(command or ''))
expected = os.path.normcase(os.path.abspath(str(expected_pcb)))
return (os.path.basename(exe) == 'expeditionpcb.exe'
and '-reusablecircuit' in command
and expected in command)
and '-reusablecircuit' in {arg.casefold() for arg in args[1:]}
and expected in {os.path.normcase(os.path.abspath(arg)) for arg in args[1:]})
def select_reusable_circuit_pcb(candidates):
@@ -636,7 +637,7 @@ def execute_job(job, emit):
pcb_log_offset = snapshot_log(pcb_log)
integration_log_dir = Path(job['pcb']).parent / 'LogFiles'
integration_offsets = {}
for name in ('ForwardAnnotation.txt', 'MsgBoxResponse.log', 'PartPkg.log'):
for name in (() if job.get('diagnostic_only') else ('ForwardAnnotation.txt', 'MsgBoxResponse.log', 'PartPkg.log')):
path = integration_log_dir / name
try:
baseline = path.read_bytes()
@@ -663,6 +664,8 @@ def execute_job(job, emit):
# Inspect every source before returning a generic server/packager
# failure: either can be a consequence of a lost library session.
startup = pcb_log_reports_missing_server(job['pcb'], offset=pcb_log_offset)
if job.get('diagnostic_only'):
return startup
integration = pcb_log_reports_integration_error(job['pcb'], offsets=integration_offsets)
if 'ICDB_REPAIR_REQUIRED' in startup:
return startup
@@ -692,6 +695,8 @@ def execute_job(job, emit):
reference = match[2]
if not reference or (config.parent/reference).resolve() != library:
raise ValueError(f'DIAGNOSTIC_LIBRARY_MISMATCH: {config}: {reference}')
from .diagnostic_open import snapshot_diagnostic_logs
job['diagnostic_log_baseline'] = snapshot_diagnostic_logs(job)
else:
ensure_pcb_central_library(
job['pcb'],
@@ -708,6 +713,7 @@ def execute_job(job, emit):
# The successful background runs use the normal ReusableCircuit
# launcher. Keep the experimental Library Manager menu route opt-in
# for diagnostics only; it is not the production trigger on EE7.9.
pcb_started_ns = time.time_ns()
if bool(job.get('native_lm_verify', False)):
from .library_manager_verify import launch_library_verify
emit(stage='lm_launch', library=job['lmc_file'], block=job['block_name'])
@@ -749,46 +755,51 @@ def execute_job(job, emit):
parent_pid=worker_parent_pid or lm_pid, command=worker_command,
binding='commandline' if is_reusable_circuit_pcb(
worker_executable, worker_command, job['pcb']) else 'launcher_child')
try:
from .mentor_paths import discover_sdd_home
home = discover_sdd_home(job.get('mentor_bin'))
wdir = os.environ.get('WDIR', '').split(';')[0].strip()
if wdir or home:
log_root = Path(os.path.expandvars(wdir)) if wdir else home/'standard'
(root/'diagnostics/native-client-context.json').write_text(json.dumps({
'native_pid': worker_pid, 'started_ns': pcb_started_ns,
'client_log_root': str(log_root/'iCDB/Client/Logs'),
'database': str(Path(job['project']).parent/'database')},
ensure_ascii=False), encoding='utf-8')
except (OSError, ValueError):
pass
break
time.sleep(.2)
if worker_pid is None:
raise RuntimeError('ReusableCircuit common 启动后未观察到 wg PCB 子进程')
app, doc, pi = wait_for_target_document(
job,
lambda: win32com.client.GetActiveObject('MGCPCB.Application'),
process_alive=lambda: any(candidate_pid in {lm_pid, worker_pid, worker_parent_pid}
for candidate_pid, _, _ in pcb_processes()),
emit=emit,
error_reader=native_error,
)
identity_verified = True
if job.get('diagnostic_only'):
# The diagnostic opens only to establish current native evidence.
# Do not refresh integration, save/commit Cert, or edit PRJ paths.
if detail := native_error():
raise RuntimeError(detail)
pi = doc = app = None
gc.collect()
from .failed_session import close_failed_document
diagnostic_close_attempted = True
closed = close_failed_document(job, emit=emit)
if closed.get('status') != 'CLOSED':
raise RuntimeError(f'DIAGNOSTIC_CLOSE_UNCONFIRMED: {closed}')
release_deadline = time.monotonic() + 30
while True:
released = cleanup_verify_processes(lm_pid, worker_pid, worker_parent_pid,
expected_pcb=job['pcb'], baseline_pids=pcb_before, emit=emit)
if (released and released.get('safe_to_continue')
and not unreleased_addresses(job['lmc_file'], job['project'], job['pcb'])):
break
if time.monotonic() >= release_deadline:
raise RuntimeError('DIAGNOSTIC_CLOSE_UNCONFIRMED: 原生会话或数据库仍占用')
time.sleep(.2)
if detail := native_error():
# Global PCB COM can wait on another design's modal window.
# Opening checks use only the newly bound process and fresh logs.
from .diagnostic_open import check_target_window
def diagnostic_emit(**event):
nonlocal diagnostic_close_attempted
if event.get('stage') == 'diagnostic_wm_close':
diagnostic_close_attempted = True
emit(**event)
checked = check_target_window(job, worker_pid, emit=diagnostic_emit,
error_reader=native_error, launcher_process=launcher_process)
released = cleanup_verify_processes(lm_pid, worker_pid, worker_parent_pid,
expected_pcb=job['pcb'], baseline_pids=pcb_before, emit=emit)
if (checked.get('status') != 'PASS' or not checked.get('native_processes_exited')
or not released or not released.get('safe_to_continue')):
raise RuntimeError(f'DIAGNOSTIC_CLOSE_UNCONFIRMED: {checked}; {released}')
from .diagnostic_open import _read_bounded
if detail := _read_bounded(native_error):
raise RuntimeError(detail)
completed = True
emit(stage='diagnostic_complete', status='PASS')
return {'status': 'PASS', 'stdout': 'ICDB 启动检查通过;未执行集成或保存,已关闭工程。'}
return {'status': 'PASS', 'stdout': 'ICDB 启动检查通过;工程及启动器已退出。'}
from .pcb_bridge import ProcessWatch
with ProcessWatch((lm_pid, worker_pid, worker_parent_pid)) as watch:
app, doc, pi = wait_for_target_document(
job, lambda: win32com.client.GetActiveObject('MGCPCB.Application'),
process_alive=watch.alive, emit=emit, error_reader=native_error)
identity_verified = True
# Do not acquire MGCPCBAutomationLicensing from this 64-bit Python
# worker. EE7.9 registers that automation class for the 32-bit host;
# the old, working path acquires the token inside the SysWOW64
@@ -810,15 +821,16 @@ def execute_job(job, emit):
script = Path(__file__).with_name('save_verified_block.vbs')
cscript = Path(os.environ.get('WINDIR', r'C:\Windows'))/'SysWOW64'/'cscript.exe'
from .desktop_worker import launch_save_helper
helper_gate = root/'start-save.signal'
with (root/'save.log').open('w', encoding='utf-8') as log:
helper = launch_save_helper([str(cscript), '//nologo', str(script), job['project'],
Path(job['pcb']).name, str(job.get('cancel_signal') or '')],
Path(job['pcb']).name, str(job.get('cancel_signal') or ''), str(helper_gate)],
start_gate=helper_gate, identity_path=root/'worker-state.json',
operation_id=job.get('operation_id', root.name),
stdout=log, stderr=log, creationflags=getattr(subprocess, 'CREATE_NO_WINDOW', 0))
emit(stage='save_helper', helper_pid=helper.pid)
last_markers = ()
while helper.poll() is None:
if job.get('cancel_signal') and Path(job['cancel_signal']).exists():
raise RuntimeError('VERIFY_CANCELLED: 已停止等待保存客户端')
try:
current = tuple(line.strip() for line in (root/'save.log').read_text(encoding='utf-8', errors='replace').splitlines() if line.strip())
except OSError:
@@ -826,14 +838,18 @@ def execute_job(job, emit):
if current != last_markers:
last_markers = current
from .runtime_timing import parse_save_progress
emit(stage='save_progress', **parse_save_progress('\n'.join(current)))
emit(stage='save_progress', helper_exited=False, **parse_save_progress('\n'.join(current)))
time.sleep(.2)
code = helper.returncode
save_log = (root/'save.log').read_bytes().decode('utf-8', errors='replace')
from .runtime_timing import parse_save_progress
emit(stage='save_progress', **parse_save_progress(save_log))
emit(stage='save_progress', helper_exited=True, **parse_save_progress(save_log))
markers = set(save_log.splitlines())
if (code not in (None, 0)) or not save_log_has_core_markers(save_log):
from .operation_lifecycle import classify_save_state
write_facts = classify_save_state(save_log.splitlines(), helper_exited=True, native_released=None)
if ((code not in (None, 0)) or not save_log_has_core_markers(save_log)
or write_facts['save_state'] != 'COMMITTED' or write_facts['close_state'] != 'CLOSED'
or write_facts['interrupted']):
raise RuntimeError(f'保存/关闭未完成 ({code}): {save_log}')
if 'QUIT_OK' not in markers:
emit(stage='save_quit_recovered', detail='Save/Close 已确认,忽略卡住的 app.Quit。')
@@ -876,9 +892,12 @@ def execute_job(job, emit):
pass
if helper is not None and helper.poll() is None:
emit(stage='save_helper_retained', helper_pid=helper.pid)
elif helper is not None:
# A failed/truncated Save result never authorizes discard-close.
emit(stage='save_result_retained', detail='保存或关闭结果不完整,保留原生会话供核对')
elif diagnostic_close_attempted:
# A timed-out close helper may still own an in-flight native Close.
# Observe release below, but never start a second close helper.
# The process-bound checker already sent its sole WM_CLOSE.
# Preserve an unknown prompt; never fall back to global COM Close.
pass
elif not completed and identity_verified:
try:
@@ -909,19 +928,6 @@ def execute_job(job, emit):
lm_session.close()
def _terminate_pid(pid, *, timeout=20):
"""Terminate a validated automation helper PID, never its native children."""
try:
subprocess.run(
[str(Path(os.environ.get('WINDIR', r'C:\Windows'))/'System32'/'taskkill.exe'), '/PID', str(int(pid)), '/F'],
capture_output=True,
timeout=timeout,
creationflags=getattr(subprocess, 'CREATE_NO_WINDOW', 0),
)
except Exception:
pass
def cleanup_verify_processes(lm_pid, worker_pid, worker_parent_pid=None, *,
expected_pcb=None, baseline_pids=(), emit=None):
"""Observe native process release without interrupting database writes.
@@ -983,6 +989,22 @@ def worker_main(request_path):
root = Path(job['root'])
def emit(**event):
from .session_record import update_worker_state
phase = {'save_start': 'NATIVE_CALL', 'integration_action': 'NATIVE_CALL',
'integration_action_complete': 'POSTCALL', 'saved_and_closed': 'RELEASE_CHECK',
'complete': 'FINISHED', 'failed': 'UNCONFIRMED'}.get(event.get('stage'))
identities = [(role, event.get(key)) for role, key in
(('native', 'pcb_pid'), ('native', 'lm_pid'), ('helper', 'helper_pid'))
if event.get(key)]
facts = None
if event.get('stage') == 'save_progress':
from .operation_lifecycle import classify_save_state
facts = classify_save_state(event.get('markers', ()),
helper_exited=event.get('helper_exited'), native_released=None)
phase = ('POSTCALL' if event.get('helper_exited') is True else
'CLOSING' if 'CLOSE_START' in event.get('markers', ()) else 'NATIVE_CALL')
if phase or identities:
update_worker_state(job, phase or 'UNCONFIRMED', identities=identities, write_outcome=facts)
event.update(time=time.strftime('%Y-%m-%d %H:%M:%S'), block=job['block_name'],
monotonic_seconds=time.monotonic())
with (root/'events.jsonl').open('a', encoding='utf-8') as stream:
@@ -990,15 +1012,24 @@ def worker_main(request_path):
try:
if job.get('supervisor_pid'):
from .session_record import update_worker_state
update_worker_state(job, 'PRECALL')
from .desktop_worker import start_supervisor_watchdog
start_supervisor_watchdog(job['supervisor_pid'], cancel_path=job.get('cancel_signal'), stop_clients=True)
start_supervisor_watchdog(job['supervisor_pid'], cancel_path=job.get('cancel_signal'), stop_clients=True,
supervisor_identity=job.get('supervisor_identity'),
session_record_path=job.get('session_record_path'), operation_id=job.get('operation_id'))
start_gate = Path(job.get('start_gate', '')) if job.get('start_gate') else None
if start_gate is not None:
gate_deadline = time.monotonic() + 15
from .defaults import WORKER_START_GATE_TIMEOUT_SECONDS
gate_deadline = time.monotonic() + WORKER_START_GATE_TIMEOUT_SECONDS
while not start_gate.is_file():
if job.get('cancel_signal') and Path(job['cancel_signal']).exists():
raise RuntimeError('VERIFY_CANCELLED: 启动已取消,未启动 PCB')
if time.monotonic() >= gate_deadline:
raise RuntimeError('DIALOG_AGENT_START_TIMEOUT: 弹窗代理未就绪,未启动 PCB')
time.sleep(.05)
if job.get('cancel_signal') and Path(job['cancel_signal']).exists():
raise RuntimeError('VERIFY_CANCELLED: 启动已取消,未启动 PCB')
outcome = execute_job(job, emit)
except Exception as error:
traceback.print_exc()
@@ -1434,170 +1465,58 @@ def close_library_reader(pid, desktop, *, timeout_seconds=5):
return False
def read_library_flag(library, name, *, timeout_seconds=90, log_path=None):
"""Return (flag, detail) for one CBB by reading its tree icon.
Read-only: Library Manager is started on a private desktop, the tree is
walked with messages, the icon index is read, and the process is killed.
The cursor is never moved and no menu is opened.
"""
import win32con
import win32process
import win32service
executable = library_manager_path()
if not executable.is_file():
return FLAG_UNKNOWN, f'找不到 Library Manager: {executable}'
def _existing_library_tree_images(library, names, *, timeout_seconds):
"""Observe a matching, enabled LM; never launch, select, answer or close it."""
from pywinauto import Desktop
from pywinauto.controls.common_controls import TreeViewWrapper
deadline = time.monotonic() + timeout_seconds
desktop_name = 'CbbFlagRead_' + os.urandom(5).hex()
desktop = win32service.CreateDesktop(desktop_name, 0, win32con.GENERIC_ALL, None)
handle = None
pid = None
handled = set()
log = log_path or Path(library).parent/'LogFiles'/'cbb_flag_read_dialogs.jsonl'
library = Path(library)
result = {name: None for name in names}
try:
startup = win32process.STARTUPINFO()
startup.lpDesktop = 'winsta0\\' + desktop_name
handle, thread, pid, _ = win32process.CreateProcess(
str(executable), subprocess.list2cmdline([str(executable), str(library)]),
None, None, False, win32con.CREATE_NO_WINDOW, None, None, startup)
thread.Close()
target = None
while time.monotonic() < deadline:
target = _private_library_window(desktop, library)
if target is not None:
break
handle_known_dialogs(pid, handled, log, desktop=desktop)
time.sleep(.3)
if target is None:
return FLAG_UNKNOWN, 'Library Manager 窗口未就绪(状态未读取)'
trees = [w for w in target.descendants() if 'TreeView' in w.class_name()]
matches = [window for window in Desktop(backend='win32').windows()
if window.window_text().startswith('Library Manager:')
and same_path(window.window_text().split(':', 1)[1].strip(), library)]
if len(matches) != 1:
return result, '未找到唯一匹配库的已打开 Library Manager,状态无法确认'
target = matches[0]
if not target.is_enabled():
return result, 'Library Manager 有原生提示或正在操作,状态无法确认'
trees = [window for window in target.descendants() if 'TreeView' in window.class_name()]
if len(trees) != 1:
return FLAG_UNKNOWN, '导航树不唯一(状态未读取)'
return result, '导航树不唯一,状态无法确认'
tree = TreeViewWrapper(trees[0].handle)
item = tree.get_item(['Central Libraries', Path(library).name, 'Reusable Blocks', 'Logical-Physical', name], exact=True)
item.ensure_visible()
index = _tree_image(item, tree)
from .library_reader import filesystem_evidence_map, infer_tree_statuses, reconcile_statuses
evidence = filesystem_evidence_map(library, (name,))
filesystem = {name: evidence[name]['status']}
calibrated = infer_tree_statuses({name: index}, filesystem, (name,))
status = reconcile_statuses(calibrated, filesystem, (name,))[name]
flag = FLAG_GREEN if status == 'READY' else FLAG_NOT_GREEN if status == 'PENDING' else FLAG_UNKNOWN
detail = f'图标索引 {index};持久化证据={evidence.get(name, {})};校准状态={status}'
if flag == FLAG_UNKNOWN:
detail += ';未取得足够的持久化证据与图标 ID 对应关系,未强行判定'
return flag, detail
except Exception as error:
return FLAG_UNKNOWN, f'状态读取失败: {error}'
finally:
released = pid is None or close_library_reader(pid, desktop)
if handle is not None:
deadline = time.monotonic() + timeout_seconds
for name in names:
if time.monotonic() >= deadline or not target.is_enabled():
break
try:
handle.Close()
item = tree.get_item(['Central Libraries', library.name, 'Reusable Blocks',
'Logical-Physical', name], exact=True)
result[name] = _tree_image(item, tree)
except Exception:
pass
try:
desktop.CloseDesktop()
except Exception:
pass
return result, '已观察匹配库的导航树图标;图标本身不是权威状态码'
except Exception as error:
return result, f'状态读取失败: {error}'
if not released:
raise RuntimeError(f'LIBRARY_RELEASE_UNCONFIRMED: pid={pid}; desktop={desktop_name}')
def read_library_flag(library, name, *, timeout_seconds=90, log_path=None):
library = Path(library)
images, detail = _existing_library_tree_images(library, (name,), timeout_seconds=timeout_seconds)
if images[name] is None:
return FLAG_UNKNOWN, detail
from .library_reader import filesystem_evidence_map, infer_tree_statuses, reconcile_statuses
evidence = filesystem_evidence_map(library, (name,))
filesystem = {name: evidence[name]['status']}
calibrated = infer_tree_statuses(images, filesystem, (name,))
status = reconcile_statuses(calibrated, filesystem, (name,))[name]
flag = FLAG_GREEN if status == 'READY' else FLAG_NOT_GREEN if status == 'PENDING' else FLAG_UNKNOWN
return flag, f'{detail};图标索引 {images[name]};校准状态={status}'
def read_library_tree_image_map(library, names, *, timeout_seconds=45, log_path=None):
"""Read all CBB tree icon resource IDs in one private LM session.
This is deliberately read-only. It avoids the previous one-process-per-
CBB implementation, while keeping the LM window off the user's desktop.
The returned integer is only an observation. It is logged with each CBB
and calibrated against independent persisted Cert evidence because Mentor
resource IDs vary by installation. Context-menu labels are not used.
"""
import win32con
import win32process
import win32service
from pywinauto import Desktop
from pywinauto.controls.common_controls import TreeViewWrapper
names = tuple(names)
result = {name: None for name in names}
executable = library_manager_path()
if not executable.is_file():
return result
desktop_name = 'CbbStatusRead_' + os.urandom(5).hex()
desktop = win32service.CreateDesktop(desktop_name, 0, win32con.GENERIC_ALL, None)
handle = None
pid = None
handled = set()
log = log_path or Path(library).parent/'LogFiles'/'cbb_status_read_dialogs.jsonl'
try:
startup = win32process.STARTUPINFO()
startup.lpDesktop = 'winsta0\\' + desktop_name
handle, thread, pid, _ = win32process.CreateProcess(
str(executable), subprocess.list2cmdline([str(executable), str(library)]),
None, None, False, win32con.CREATE_NO_WINDOW, None, None, startup)
handle.Close(); handle = None
thread.Close()
deadline = time.monotonic() + timeout_seconds
target = None
while time.monotonic() < deadline:
target = _private_library_window(desktop, library)
if target is not None:
break
handle_known_dialogs(pid, handled, log, desktop=desktop)
time.sleep(.25)
if target is None:
return result
trees = [w for w in target.descendants() if 'TreeView' in w.class_name()]
if len(trees) != 1:
return result
tree = TreeViewWrapper(trees[0].handle)
root_name = Path(library).name
details = {}
for name in names:
if time.monotonic() >= deadline:
logger.warning('library_tree_image_probe_timeout path=%s processed=%s total=%s',
library, len(details), len(names))
break
try:
item = tree.get_item(['Central Libraries', root_name, 'Reusable Blocks', 'Logical-Physical', name], exact=True)
details[name] = _tree_image_details(item, tree)
result[name] = details[name]['image_id']
except Exception:
result[name] = None
logger.info('library_tree_images path=%s names=%s histogram=%s', library, len(names),
dict(Counter(str(value) for value in result.values() if value is not None)))
for name in names:
if name in details:
logger.info('library_tree_image_item name=%s image_id=%s selected_image_id=%s state_image_id=%s image_color=%s selected_image_color=%s state_image_color=%s',
name, details[name]['image_id'], details[name]['selected_image_id'],
details[name]['state_image_id'], details[name]['image_color'],
details[name]['selected_image_color'], details[name]['state_image_color'])
else:
logger.warning('library_tree_image_item_missing name=%s reason=tree_path_lookup_failed', name)
return result
finally:
released = pid is None or close_library_reader(pid, desktop)
if handle is not None:
try:
handle.Close()
except Exception:
pass
try:
desktop.CloseDesktop()
except Exception:
pass
if not released:
raise RuntimeError(f'LIBRARY_RELEASE_UNCONFIRMED: pid={pid}; desktop={desktop_name}')
library = Path(library)
images, _ = _existing_library_tree_images(library, tuple(names), timeout_seconds=timeout_seconds)
return images
def read_library_status_map(library, names, *, timeout_seconds=45, log_path=None,
@@ -265,14 +265,8 @@ def validate_export_output(kind, code, report, log):
def _process_snapshot():
import win32com.client
service = win32com.client.GetObject('winmgmts:')
try:
return [{'pid': int(p.ProcessId), 'parent': int(p.ParentProcessId or 0),
'name': str(p.Name), 'executable': str(p.ExecutablePath or '')}
for p in service.ExecQuery('SELECT ProcessId, ParentProcessId, Name, ExecutablePath FROM Win32_Process')]
finally:
service = None
from .owner_probe import query_process_snapshot
return query_process_snapshot(timeout_seconds=10)
def _database_released(project_root):
@@ -291,13 +285,69 @@ def _database_released(project_root):
return True
def _process_birth(value):
from datetime import datetime, timezone, timedelta
if not value:
return None
if isinstance(value, datetime):
return value.astimezone(timezone.utc)
if re.fullmatch(r'\d{14}\.\d{6}[+-]\d{3}', value):
zone = timezone(timedelta(minutes=int(value[21:])))
return datetime.strptime(value[:21], '%Y%m%d%H%M%S.%f').replace(tzinfo=zone).astimezone(timezone.utc)
try:
parsed = datetime.fromisoformat(value)
if parsed.tzinfo is not None:
return parsed.astimezone(timezone.utc)
except ValueError:
pass
return value
def _export_descendants(snapshot, tracked, existing):
"""A historical parent PID is not evidence of current parentage."""
from datetime import datetime
rows = {row['pid']: row for row in snapshot}
def matches(row):
birth = _process_birth(row.get('created'))
return tracked[row['pid']] is None or birth is None or tracked[row['pid']] == birth
changed = True
while changed:
changed = False
for row in snapshot:
if row['pid'] in tracked or row['parent'] not in tracked:
continue
birth = _process_birth(row.get('created'))
if (row['pid'], birth) in existing:
continue
parent = rows.get(row['parent'])
parent_birth = tracked[row['parent']]
if parent is not None and not matches(parent):
continue
if isinstance(birth, datetime) and isinstance(parent_birth, datetime) and birth < parent_birth:
continue
tracked[row['pid']] = birth
changed = True
return [row for row in snapshot if row['pid'] in tracked and matches(row)]
def _run_exports(project, root, mentor_home, timeout_seconds):
"""Observe exact exporter/descendant handles; never kill a database owner."""
import win32api, win32con, win32event, pywintypes
import win32api, win32con, win32event, win32process, pywintypes
root = Path(root)
_write_json(root/'export-launch.json', {'started': False})
before = _process_snapshot()
if any(p['name'].casefold() in _OWNER_NAMES for p in before):
try:
_write_json(root/'export-launch.json', {'started': False})
except OSError as error:
return {'success': False, 'safe_to_continue': True, 'native_started': False,
'detail': str(error)}
diagnostic_warnings = []
try:
before = _process_snapshot()
except Exception as error:
return {'success': False, 'safe_to_continue': True, 'native_started': False,
'detail': str(error)}
from .backup_recovery import _native_scope_match
if any(p['name'].casefold() in _OWNER_NAMES and _native_scope_match(
p.get('command', ''), roots=(project.parent,)) is True for p in before):
return {'success': False, 'safe_to_continue': False, 'detail': 'An existing iCDB owner prevents isolated identity export.'}
if not _database_released(project.parent):
return {'success': False, 'safe_to_continue': False, 'detail': 'Analysis copy has unreleased native addresses or DB handles.'}
@@ -318,31 +368,48 @@ def _run_exports(project, root, mentor_home, timeout_seconds):
for kind, command, report, log in commands:
process = None
handles = {}
identities = {}
released = False
record = {'command': command, 'kind': kind, 'success': False, 'safe_to_continue': False}
launch_attempted = False
record = {'command': command, 'kind': kind, 'success': False,
'safe_to_continue': False, 'native_started': False}
try:
deadline = time.monotonic()+float(timeout_seconds)
with (root/(kind+'-stdout.txt')).open('wb') as output:
_write_json(root/'export-launch.json', {'started': True})
launch_attempted = True
record['native_started'] = None
process = subprocess.Popen(command, stdout=output, stderr=subprocess.STDOUT,
creationflags=subprocess.CREATE_NO_WINDOW)
record['native_started'] = True
record['pid'] = process.pid
descendants = {process.pid}
handle = getattr(process, '_handle', None)
birth = _process_birth(win32process.GetProcessTimes(handle)['CreationTime']) if handle is not None else None
descendants = {process.pid: birth}
existing = {(p['pid'], _process_birth(p.get('created'))) for p in before}
if birth is not None:
record['root_created'] = birth.isoformat()
_write_json(root/(kind+'-process.json'), record)
while time.monotonic() < deadline:
snapshot = _process_snapshot()
changed = True
while changed:
previous = len(descendants)
descendants.update(p['pid'] for p in snapshot if p['parent'] in descendants)
changed = len(descendants) != previous
owners = [p for p in snapshot if p['pid'] in descendants or p['name'].casefold() in _OWNER_NAMES]
owned = {p['pid'] for p in _export_descendants(snapshot, descendants, existing)}
owners = [p for p in snapshot if p['pid'] in owned or (
p['name'].casefold() in _OWNER_NAMES and _native_scope_match(
p.get('command', ''), roots=(project.parent,)) is True)]
for owner in owners:
if owner['pid'] not in handles:
try: handles[owner['pid']] = win32api.OpenProcess(win32con.SYNCHRONIZE, False, owner['pid'])
except pywintypes.error as error:
if error.winerror != 87: raise
record['observed_pids'] = sorted(descendants | handles.keys())
if error.winerror not in (5, 87): raise
# A dying process can deny OpenProcess. Keep it in
# owners until a fresh snapshot proves it absent.
record.setdefault('process_observation_errors', {})[str(owner['pid'])] = {
'winerror': error.winerror, 'detail': str(error)}
record['observed_pids'] = sorted(descendants.keys() | handles.keys())
for owner in owners:
identities.setdefault(owner['pid'],
{key: owner.get(key, '') for key in ('pid', 'created', 'executable')})
record['process_identities'] = list(identities.values())
if (process.poll() is not None and not owners
and all(win32event.WaitForSingleObject(h, 0) == win32con.WAIT_OBJECT_0 for h in handles.values())):
released = _database_released(project.parent)
@@ -355,15 +422,37 @@ def _run_exports(project, root, mentor_home, timeout_seconds):
validate_export_output(kind, process.returncode, report, log)
record.update(success=True, detail='Native export complete and owners released.')
except Exception as error:
creation_rejected = (isinstance(error, (FileNotFoundError, PermissionError))
or getattr(error, 'winerror', None) in (2, 3, 5, 193, 206, 740))
if process is None and (not launch_attempted or creation_rejected):
# Only a known CreateProcess rejection proves that no native owner started.
released = True
record['native_started'] = False
_write_json(root/'export-launch.json', {'started': False})
record.update(detail=str(error), safe_to_continue=released)
return record
finally:
if process is not None and not released:
_retained_exports.append((process, handles))
import json
owner_path = root/'export-owner.json'
try:
owner = json.loads(owner_path.read_text('utf-8')) if owner_path.is_file() else {}
except (ValueError, OSError):
owner = {} # Preserve unknown ownership; never lose the live handles.
_retained_exports.append((process, handles, {
'library': owner.get('library'), 'project': str(project), 'analysis_root': str(root)}))
else:
for handle in handles.values(): handle.Close()
_write_json(root/(kind+'-process.json'), record)
return {'success': True, 'safe_to_continue': True, 'detail': 'All three native exports completed and released.'}
try:
_write_json(root/(kind+'-process.json'), record)
except OSError as error:
if not released:
raise
warning = 'Export diagnostic write failed: '+str(error)
diagnostic_warnings.append(warning)
record['detail'] = (record.get('detail', '')+'\n'+warning).strip()
return {'success': True, 'safe_to_continue': True,
'detail': '\n'.join(['All three native exports completed and released.', *diagnostic_warnings])}
def capture_identity(source, evidence_root, *, expected_library, owner_library, mentor_home,
@@ -374,18 +463,22 @@ def capture_identity(source, evidence_root, *, expected_library, owner_library,
from .scoped_manifest import copy_verified
from .native_export_guard import begin_export_analysis, finish_export_analysis
import win32event, win32con
source, pcb, top = source_project(source)
root = Path(evidence_root).absolute()
library = Path(owner_library)
lease = LibraryLock(library)
if owned_snapshot(lease.path) is None:
raise ValueError('Native identity capture requires a current-thread transaction library lease.')
if (root.resolve().is_relative_to(source.parent) or source.parent.is_relative_to(root.resolve())
or root.resolve().is_relative_to(library.parent.resolve())):
raise ValueError('Native identity evidence must be independent of source and library.')
check_exporters(mentor_home)
root.mkdir(parents=True, exist_ok=False)
mutex = win32event.CreateMutex(None, False, r'Local\CbbRefreshVerify795')
root = Path(evidence_root)
try:
root = root.absolute()
source, pcb, top = source_project(source)
library = Path(owner_library)
lease = LibraryLock(library)
if owned_snapshot(lease.path) is None:
raise ValueError('Native identity capture requires a current-thread transaction library lease.')
if (root.resolve().is_relative_to(source.parent) or source.parent.is_relative_to(root.resolve())
or root.resolve().is_relative_to(library.parent.resolve())):
raise ValueError('Native identity evidence must be independent of source and library.')
check_exporters(mentor_home)
root.mkdir(parents=True, exist_ok=False)
mutex = win32event.CreateMutex(None, False, r'Local\CbbRefreshVerify795')
except Exception as error:
return IdentityOutcome(None, True, str(error), root)
acquired = False
owner_safe = True
outcome = IdentityOutcome(None, True, 'Identity capture did not start.', root)
@@ -403,10 +496,11 @@ def capture_identity(source, evidence_root, *, expected_library, owner_library,
export_token = begin_export_analysis(library, root, clone)
owner_safe = False
native = _run_exports(clone, root, mentor_home, timeout_seconds)
_write_json(root/'native-outcome.json', native)
if native.get('safe_to_continue') is True:
finish_export_analysis(library, export_token)
finish_export_analysis(library, export_token,
native_not_started=native.get('native_started') is False)
owner_safe = True
_write_json(root/'native-outcome.json', native)
if not owner_safe or not native.get('success'):
outcome = IdentityOutcome(None, owner_safe, native.get('detail', 'Native export failed.'), root)
else:
@@ -423,6 +517,10 @@ def capture_identity(source, evidence_root, *, expected_library, owner_library,
finally:
if acquired: win32event.ReleaseMutex(mutex)
mutex.Close()
_write_json(root/'identity.json', {'identity': outcome.identity, 'safe_to_continue': outcome.safe_to_continue,
'detail': outcome.detail, 'policy': POLICY})
try:
_write_json(root/'identity.json', {'identity': outcome.identity, 'safe_to_continue': outcome.safe_to_continue,
'detail': outcome.detail, 'policy': POLICY})
except OSError as error:
outcome = IdentityOutcome(outcome.identity, outcome.safe_to_continue,
outcome.detail+'\nIdentity summary write failed: '+str(error), root)
return outcome
+96 -15
View File
@@ -1,20 +1,62 @@
"""Export-copy ownership bound to the existing per-library lease checkpoint."""
from contextlib import contextmanager, ExitStack
import json
import logging
from pathlib import Path
from uuid import uuid4
from .management import _write_json
def assert_retained_exports_idle():
def _preexisting_export_pids(record):
from datetime import datetime
from .native_engineering_identity import _process_birth
identities = record.get('process_identities', [])
root = next((p for p in identities if p.get('pid') == record.get('pid')), {})
birth = _process_birth(record.get('root_created') or root.get('created'))
if not isinstance(birth, datetime):
return set()
return {p['pid'] for p in identities if p.get('pid') != record.get('pid')
and isinstance(_process_birth(p.get('created')), datetime)
and _process_birth(p['created']) < birth}
def assert_retained_exports_idle(library=None):
import win32con, win32event
from .native_engineering_identity import _retained_exports
from .library_lock import LibraryBusy
for process, handles in _retained_exports:
if process.poll() is None or any(win32event.WaitForSingleObject(handle, 0) != win32con.WAIT_OBJECT_0
for handle in handles.values()):
raise LibraryBusy('Native identity exporter or descendant has not released.')
for entry in _retained_exports:
process, handles = entry[:2]
scope = entry[2] if len(entry) > 2 else {}
if (library is not None and scope.get('library')
and Path(scope['library']).resolve() != Path(library).resolve()):
continue
excluded = set()
exclusion_error = ''
if scope.get('analysis_root') and getattr(process, 'pid', None):
evidence_path = Path(scope['analysis_root'])
try:
matched = False
for path in Path(scope['analysis_root']).glob('*-process.json'):
evidence_path = path
record = json.loads(path.read_text('utf-8'))
if not isinstance(record, dict):
raise ValueError('Export process evidence must be a JSON object')
if record.get('pid') == process.pid:
matched = True
excluded.update(_preexisting_export_pids(record))
if not matched:
raise ValueError(f'No process evidence matches exporter PID {process.pid}')
except (OSError, ValueError, KeyError, TypeError) as error:
excluded.clear()
exclusion_error = f'exclusion_unavailable: {evidence_path}: {error}'
logging.getLogger(__name__).warning('%s; library=%s; pid=%s',
exclusion_error, scope.get('library'), process.pid)
if process.poll() is None or any(pid not in excluded
and win32event.WaitForSingleObject(handle, 0) != win32con.WAIT_OBJECT_0
for pid, handle in handles.items()):
raise LibraryBusy('Native identity exporter or descendant has not released.'
+ (' '+exclusion_error if exclusion_error else ''))
def begin_export_analysis(library, root, project):
@@ -40,26 +82,33 @@ def begin_export_analysis(library, root, project):
@contextmanager
def prove_export_copies_released(library):
def prove_export_copies_released(library, *, allow_unknown_checkpoint=False, unstarted_token=''):
"""Fresh process and exclusive-copy proof, including after interpreter exit."""
import win32file, win32con, pywintypes
from .library_lock import blocker_path, checkpoint_record, _supervisor_alive, LibraryBusy
from .native_engineering_identity import _process_snapshot, _OWNER_NAMES
from .scoped_manifest import scan_tree
assert_retained_exports_idle()
library = Path(library).resolve()
assert_retained_exports_idle(library)
marker = blocker_path(library)
checkpoint = checkpoint_record(marker)
if marker.exists() and checkpoint is None:
raise LibraryBusy('Invalid library checkpoint cannot release export ownership.')
if not allow_unknown_checkpoint:
raise LibraryBusy('Invalid library checkpoint cannot release export ownership.')
# Only the explicit release path supplies full-library/native proof and
# archives the unchanged metadata while its coordination handle is held.
assert_retained_exports_idle()
if any(p['name'].casefold() in _OWNER_NAMES for p in _process_snapshot()):
raise LibraryBusy('Unknown export checkpoint still has a possible native identity owner.')
yield []
return
if checkpoint is None or 'export_analysis' not in checkpoint:
yield []
return
records = checkpoint['export_analysis']
if not isinstance(records, list) or not records:
raise LibraryBusy('Invalid export-analysis ownership record.')
owners = [p for p in _process_snapshot() if p['name'].casefold() in _OWNER_NAMES]
if owners: raise LibraryBusy(f'Native identity owners remain active: {owners}')
snapshot = None
copies = []
with ExitStack() as stack:
for record in records:
@@ -78,15 +127,46 @@ def prove_export_copies_released(library):
processes = [root/(kind+'-process.json') for kind in ('ascii', 'csv', 'edif')
if (root/(kind+'-process.json')).exists()]
if not processes:
launch = json.loads((root/'export-launch.json').read_text(encoding='utf-8'))
if launch != {'started': False}:
launch_path = root/'export-launch.json'
launch = json.loads(launch_path.read_text('utf-8')) if launch_path.exists() else None
unstarted_current_lease = False
if unstarted_token and record['token'] == unstarted_token:
# In-memory no-start proof may clear its own guard even
# when creating the first launch record failed.
from .management_transaction import current_lease
current = current_lease(library)
unstarted_current_lease = (record['lease_id'] == current['lease_id']
and record['supervisor_pid'] == current['supervisor_pid'])
if launch != {'started': False} and not (launch is None and unstarted_current_lease):
raise ValueError('Export launch has no confirmed process record.')
else:
if snapshot is None:
snapshot = _process_snapshot()
from .backup_recovery import _native_scope_match
if any(p['name'].casefold() in _OWNER_NAMES and _native_scope_match(
p.get('command', ''), roots=(project.parent,)) is True for p in snapshot):
raise LibraryBusy('A native identity owner is bound to this analysis copy.')
for path in processes:
process = json.loads(path.read_text(encoding='utf-8'))
if ('pid' not in process and process.get('safe_to_continue') is True
and process.get('native_started') is False
and json.loads((root/'export-launch.json').read_text('utf-8')) == {'started': False}):
continue # Failed CreateProcess: no native ownership was acquired.
pids = process.get('observed_pids', []) + [process['pid']]
if any(not isinstance(pid, int) or pid <= 0 for pid in pids):
raise ValueError('Invalid export process identity.')
if any(_supervisor_alive(pid) for pid in pids):
identities = {p['pid']: p for p in process.get('process_identities', [])}
excluded = _preexisting_export_pids(process)
def original_alive(pid):
if pid in excluded:
return False
recorded = identities.get(pid, {})
current = next((p for p in snapshot if p['pid'] == pid), None)
if recorded.get('created') and current and current.get('created'):
return (recorded['created'] == current['created']
and recorded.get('executable') == current.get('executable'))
return any(p['pid'] == pid for p in snapshot) or _supervisor_alive(pid)
if any(original_alive(pid) for pid in pids):
raise LibraryBusy('A recorded export process or descendant remains alive.')
entries = scan_tree(root/'project')
if any(path.name.casefold() == 'saddress.adr' for path in entries):
@@ -105,9 +185,10 @@ def prove_export_copies_released(library):
yield copies
def finish_export_analysis(library, token):
def finish_export_analysis(library, token, *, native_not_started=False):
from .library_lock import _guard, blocker_path, checkpoint_record, LibraryBusy
with _guard, prove_export_copies_released(library):
with _guard, prove_export_copies_released(library,
unstarted_token=token if native_not_started else ''):
marker = blocker_path(library)
checkpoint = checkpoint_record(marker)
records = checkpoint.get('export_analysis', []) if checkpoint else []
+265 -68
View File
@@ -21,6 +21,12 @@ from .scoped_manifest import hash_file
from .network_path import absolute_preserving_drive
_retained_workers = []
class NativeOperationError(RuntimeError):
def __init__(self, detail, *, safe_to_continue):
super().__init__(detail)
self.safe_to_continue = safe_to_continue
# Coordinator-owned Create/Add/conflict/update/remove/backup-reimport evidence:
# diagnostics/issue2-native-crud-20261103/engine-e2e/cycle-result.json
_PRODUCTION_ADD_VALIDATED = True
@@ -62,6 +68,40 @@ class NativeManagementAdapter:
self.timeout_seconds = timeout_seconds
self.mapped_drive = ''
self.mapped_remote_root = ''
self._pending_workers = {}
self.stop = None
self.session_progress = None
def bind_stop(self, stop):
self.stop = stop
def release_confirmed(self, library):
"""Worker exit/result files do not replace independent resource proof."""
key = str(Path(library).resolve()).casefold()
pending = self._pending_workers.get(key)
if pending is None:
return False
worker, evidence = pending
if worker.poll() is None:
return False
try:
if (evidence/'session.json').is_file():
from .session_record import reconcile_session_record
proof = reconcile_session_record(evidence/'session.json', evidence_root=evidence)
if proof['status'] != 'RELEASED':
return False
else:
from .library_lock import release_retained_library
proof = release_retained_library(library, evidence/'release-proof')
if proof.get('status') != 'RELEASE_CONFIRMED':
return False
except Exception:
return False
worker.close()
if worker in _retained_workers:
_retained_workers.remove(worker)
del self._pending_workers[key]
return True
def bind_library_mapping(self, drive, remote_root):
self.mapped_drive = drive or ''
@@ -105,14 +145,18 @@ class NativeManagementAdapter:
}
def _worker(self, job, evidence):
evidence = Path(evidence)
evidence.mkdir(parents=True, exist_ok=False)
desktop = 'CbbManagement_' + uuid4().hex[:12]
timeout = float(job.get('timeout_seconds', self.timeout_seconds))
job = {**job, 'desktop': desktop, 'mentor_home': str(self.home or ''),
'mapped_drive': job.get('mapped_drive', self.mapped_drive),
'mapped_remote_root': job.get('mapped_remote_root', self.mapped_remote_root),
'timeout_seconds': timeout, 'supervisor_pid': os.getpid()}
try:
evidence = Path(evidence)
evidence.mkdir(parents=True, exist_ok=False)
desktop = 'CbbManagement_' + uuid4().hex[:12]
timeout = float(job.get('timeout_seconds', self.timeout_seconds))
job = {**job, 'desktop': desktop, 'mentor_home': str(self.home or ''),
'mapped_drive': job.get('mapped_drive', self.mapped_drive),
'mapped_remote_root': job.get('mapped_remote_root', self.mapped_remote_root),
'timeout_seconds': timeout, 'supervisor_pid': os.getpid()}
except Exception as error:
error.safe_to_continue = True
raise
if job['operation'] == 'add':
import win32con, win32event
mutex = win32event.CreateMutex(None, False, r'Local\CbbRefreshVerify795')
@@ -158,7 +202,30 @@ class NativeManagementAdapter:
return self._launch_worker(job, evidence)
def _launch_worker(self, job, evidence):
state = {'attempted': False, 'worker': None, 'released': False}
try:
return self._launch_worker_observed(job, evidence, state)
except Exception as error:
proof = getattr(error, 'worker_launch_release', None)
no_launch = (isinstance(proof, dict) and proof.get('safe_to_continue') is True
and proof.get('process_created') is False and proof.get('worker_exited') is True
and proof.get('remaining_pids') == [])
error.safe_to_continue = (not state['attempted'] or state['released']
or (state['worker'] is None and no_launch))
if state['worker'] is not None and not error.safe_to_continue:
if state['worker'] not in _retained_workers:
_retained_workers.append(state['worker'])
self._pending_workers[str(Path(job['library']).resolve()).casefold()] = (state['worker'], evidence)
raise
def _launch_worker_observed(self, job, evidence, state):
from .desktop_worker import DesktopWorker
from .session_record import process_identity
job = {**job, 'cancel_signal': str(evidence/'cancel-management.signal'),
'start_gate': str(evidence/'start-management.signal'),
'session_record_path': str(evidence/'session.json'), 'operation_id': uuid4().hex,
'supervisor_pid': os.getpid(), 'supervisor_identity': process_identity(os.getpid()),
'block': job.get('name')}
request_path = evidence/'request.json'
desktop, timeout = job['desktop'], job['timeout_seconds']
deadline = time.monotonic() + timeout
@@ -168,13 +235,31 @@ class NativeManagementAdapter:
_write_json(request_path, job)
command = ([sys.executable] if getattr(sys, 'frozen', False)
else [sys.executable, '-m', 'cbb_refresh_verify'])
state['attempted'] = True
worker = (self.worker_factory or DesktopWorker)(command + ['--management-worker', str(request_path)], desktop)
state['worker'] = worker
worker.cbb_library = str(Path(job['library']).resolve())
if hasattr(worker, 'bind_cancellation'):
worker.bind_cancellation(job['cancel_signal'])
if hasattr(worker, 'bind_session'):
worker.bind_session(job)
Path(job['start_gate']).touch()
while worker.poll() is None and time.monotonic() < deadline:
if self.stop is not None and self.stop.is_set():
Path(job['cancel_signal']).touch()
break
time.sleep(.1)
if worker.poll() is None:
# DesktopWorker.close() requests a kill without a cancellation file.
# Keep this exact worker alive until its native call returns normally.
_retained_workers.append(worker)
self._pending_workers[str(Path(job['library']).resolve()).casefold()] = (worker, evidence)
Path(job['cancel_signal']).touch()
if getattr(worker, 'session_record_path', None):
from .session_record import handoff_session
handoff_session(worker.session_record_path, evidence_root=evidence, process=worker,
progress=(lambda value: self.session_progress({'stage': 'session_observation', **value}))
if self.session_progress is not None else None)
if add_job:
(evidence/'cancel-native-add.signal').touch()
return {'success': False, 'safe_to_continue': False,
@@ -183,6 +268,7 @@ class NativeManagementAdapter:
result = json.loads((evidence/'result.json').read_text(encoding='utf-8'))
if not isinstance(result.get('safe_to_continue'), bool):
raise ValueError('原生释放记录缺失')
state['released'] = result['safe_to_continue']
return result
except (OSError, ValueError) as error:
return {'success': False, 'safe_to_continue': False, 'detail': str(error)}
@@ -192,7 +278,8 @@ class NativeManagementAdapter:
def catalog(self, library, evidence):
result = self._worker({'operation': 'catalog', 'library': str(absolute_preserving_drive(library))}, evidence)
if not result.get('success') or not result.get('safe_to_continue'):
raise RuntimeError(result.get('detail', '原生目录复读未确认,禁止文件夹回退。'))
raise NativeOperationError(result.get('detail', '原生目录复读未确认,禁止文件夹回退。'),
safe_to_continue=result.get('safe_to_continue') is True)
return Catalog(Path(result['library']), int(result['flow']),
tuple(CatalogEntry(**entry) for entry in result['entries']), result['safe_to_continue'])
@@ -228,48 +315,65 @@ class NativeManagementAdapter:
from .management_transaction import current_lease, prepare_transaction
from .management import source_project
from .backup_recovery import _align_library
if not self.capabilities()['add'].supported:
raise ValueError(self.capabilities()['add'].reason)
try:
if not self.capabilities()['add'].supported:
raise ValueError(self.capabilities()['add'].reason)
validate_native_copy_paths(request.source, request.library, request.name)
except ValueError as error:
current_lease(request.library)
original_assets = asset_identity(original_source, project_library(original_source))
staged, pcb, _ = source_project(request.source, request.top_level_block)
if staged.parent != Path(root).resolve()/'source':
raise ValueError('Only the engine staged source can be aligned for native Add.')
_align_library(staged, pcb, request.library)
if asset_identity(staged, request.library) != original_assets:
raise ValueError('Staged engineering assets differ from the original source.')
except Exception as error:
return request, IdentityOutcome(None, True, str(error), Path(root))
current_lease(request.library)
original_assets = asset_identity(original_source, project_library(original_source))
staged, pcb, _ = source_project(request.source, request.top_level_block)
if staged.parent != Path(root).resolve()/'source':
raise ValueError('Only the engine staged source can be aligned for native Add.')
_align_library(staged, pcb, request.library)
if asset_identity(staged, request.library) != original_assets:
raise ValueError('Staged engineering assets differ from the original source.')
proof = capture_identity(staged, Path(root)/'source-identity', expected_library=request.library,
owner_library=request.library, mentor_home=self.home, timeout_seconds=request.timeout_seconds)
if proof.identity is None or not proof.safe_to_continue:
return request, proof
token = prepare_transaction(request, root, proof.identity, mentor_home=self.home,
fixture=getattr(self, '_fixture_authority', None))
try:
token = prepare_transaction(request, root, proof.identity, mentor_home=self.home,
fixture=getattr(self, '_fixture_authority', None))
except Exception as error:
return request, IdentityOutcome(None, proof.safe_to_continue, str(error), Path(root))
return replace(request, transaction_token=token), proof
def _add_transaction(self, request, evidence):
from contextlib import contextmanager
from .management_transaction import validate_transaction
from .library_lock import LibraryLock
from .native_engineering_identity import capture_identity, project_library
from .management import source_project
from .backup_recovery import _align_library, _idle_check
job = {'operation': 'add', 'authority_kind': 'management', 'transaction_nonce': request.transaction_token,
'library': str(absolute_preserving_drive(request.library)), 'source': str(request.source),
'name': request.name, 'partition': request.partition, 'top_level_block': request.top_level_block,
'source_identity': engineering_identity(request.source), 'supervisor_pid': os.getpid(),
'mentor_home': str(self.home), 'timeout_seconds': request.timeout_seconds,
'mapped_drive': request.mapped_drive, 'mapped_remote_root': request.mapped_remote_root}
prepared = validate_transaction(job, evidence.parent)
owner_safe = False
with LibraryLock(request.library, safe_to_release=lambda: owner_safe):
try:
job = {'operation': 'add', 'authority_kind': 'management', 'transaction_nonce': request.transaction_token,
'library': str(absolute_preserving_drive(request.library)), 'source': str(request.source),
'name': request.name, 'partition': request.partition, 'top_level_block': request.top_level_block,
'source_identity': engineering_identity(request.source), 'supervisor_pid': os.getpid(),
'mentor_home': str(self.home), 'timeout_seconds': request.timeout_seconds,
'mapped_drive': request.mapped_drive, 'mapped_remote_root': request.mapped_remote_root}
prepared = validate_transaction(job, evidence.parent)
except Exception as error:
error.safe_to_continue = True
raise
owner_safe = True
@contextmanager
def preserve_release():
nonlocal owner_safe
try:
yield
except Exception as error:
owner_safe = getattr(error, 'safe_to_continue', owner_safe) is True
error.safe_to_continue = owner_safe
raise
with LibraryLock(request.library, safe_to_release=lambda: owner_safe), preserve_release():
owner_safe = False
result = self._worker(job, evidence)
owner_safe = result.get('safe_to_continue') is True
if not owner_safe or not result.get('success'):
return NativeOutcome(False, owner_safe, result.get('detail', 'Native Add failed.'), evidence_path=evidence)
owner_safe = False
# The native editor may leave PCB/pcb.prj relative to the staged
# location. Correct only documented library-reference tokens after
# its owner has exited, then reopen the native catalog independently.
@@ -277,12 +381,10 @@ class NativeManagementAdapter:
work = Path(request.library).parent/'ReusableBlocksLP'/request.name/'Work'
projects = list(work.glob('*.prj'))
if len(projects) != 1:
owner_safe = False
return NativeOutcome(True, False, 'Registered engineering project is not unique.',
return NativeOutcome(True, owner_safe, 'Registered engineering project is not unique.',
identity_matches=False, evidence_path=evidence)
target, pcb, _ = source_project(projects[0], request.top_level_block)
_idle_check(request.library, target, pcb)
owner_safe = False
from .maintenance_entry import maintenance_project_files
project_files = maintenance_project_files(target, pcb)
before_refs = {str(p): hash_file(p) for p in project_files}
@@ -291,13 +393,13 @@ class NativeManagementAdapter:
raise ValueError('Registered project references do not resolve to the exact target library.')
_write_json(evidence/'reference-alignment.json', {'before': before_refs,
'after': {str(p): hash_file(p) for p in project_files}, 'library': job['library']})
owner_safe = False
catalog = self.catalog(request.library, evidence/'reopen-aligned')
owner_safe = catalog.safe_to_continue
entries = [entry for entry in catalog.entries if entry.name == request.name and entry.partition == request.partition]
if (not owner_safe or Path(catalog.library).resolve() != Path(request.library).resolve()
or catalog.flow != 1 or len(entries) != 1):
owner_safe = False
return NativeOutcome(True, False, 'Aligned target did not reopen with the exact catalog identity.',
return NativeOutcome(True, owner_safe, 'Aligned target did not reopen with the exact catalog identity.',
identity_matches=False, evidence_path=evidence)
owner_safe = False
proof = capture_identity(target, evidence/'registered-identity', expected_library=request.library,
@@ -615,13 +717,43 @@ def _set_probe_field(hwnd, control_id, expected, root):
def _probe_add_dialog(app, job, desktop, owned_pids, root, *, library=None):
try:
return _probe_add_dialog_impl(app, job, desktop, owned_pids, root, library=library)
except Exception as error:
if job.get('_native_add_pending') and str(error).startswith('NATIVE_ADD_FAILED:'):
try:
failure = json.loads((Path(root)/'native-add-output.json').read_text('utf-8')).get('failure', {})
if failure.get('acknowledged') is True:
job['_native_add_pending'] = False
from .session_record import update_worker_state
update_worker_state(job, 'POSTCALL')
except (OSError, ValueError, AttributeError):
pass
raise
finally:
import gc
gc.collect() # Release UIA wrappers before closing native child dialogs.
_close_probe_dialogs(desktop, owned_pids)
if not job.get('_native_add_pending'):
_close_probe_dialogs(desktop, owned_pids)
def _require_management_continuation(job, root):
"""Check before a new native call; never abort a call already on the stack."""
from .library_lock import _supervisor_alive
cancelled = job.get('cancel_signal') and Path(job['cancel_signal']).exists()
supervisor = job.get('supervisor_pid')
if cancelled or supervisor is None or not _supervisor_alive(supervisor):
raise RuntimeError('MANAGEMENT_CANCELLED: 操作已停止或监督器已退出,未提交新的原生调用')
if job.get('supervisor_identity'):
from .session_record import process_identity, IDENTITY_KEYS
actual = process_identity(supervisor)
if any(actual[k] != job['supervisor_identity'].get(k) for k in IDENTITY_KEYS):
raise RuntimeError('MANAGEMENT_CANCELLED: 监督器身份已变化,未提交新的原生调用')
def _require_add_supervisor(job, root):
try:
_require_management_continuation(job, root)
except RuntimeError as error:
raise RuntimeError(f'NATIVE_ADD_CANCELLED: {error}') from error
from .library_lock import _supervisor_alive
deadline = job.get('native_add_deadline')
if ((Path(root)/'cancel-native-add.signal').exists()
@@ -666,7 +798,8 @@ def _probe_add_dialog_impl(app, job, desktop, owned_pids, root, *, library=None)
pythoncom.PumpWaitingMessages(); time.sleep(.1)
raise TimeoutError(f'原生窗口未出现:{title}')
_require_add_supervisor(job, root)
app.OnReusableBlocksEditor()
from .session_record import dispatch_native
dispatch_native(job, lambda: app.OnReusableBlocksEditor())
_require_add_supervisor(job, root)
editor = await_window('Reusable Blocks')
tabs = Desktop(backend='uia').window(handle=editor).descendants(control_type='TabItem')
@@ -703,7 +836,11 @@ def _probe_add_dialog_impl(app, job, desktop, owned_pids, root, *, library=None)
_write_json(root/'native-add-log-baseline.json', log_baseline)
observed_failures = {}
_require_add_supervisor(job, root)
win32gui.PostMessage(ok, win32con.BM_CLICK, 0, 0)
from .session_record import dispatch_native, update_worker_state
def submit_add():
job['_native_add_pending'] = True
win32gui.PostMessage(ok, win32con.BM_CLICK, 0, 0)
dispatch_native(job, submit_add, asynchronous=True, required=True)
success = None
while time.monotonic() < deadline:
pythoncom.PumpWaitingMessages()
@@ -733,6 +870,8 @@ def _probe_add_dialog_impl(app, job, desktop, owned_pids, root, *, library=None)
break
if not success:
raise TimeoutError('未取得原生新增成功消息;保留副本与私有桌面证据。')
job['_native_add_pending'] = False
update_worker_state(job, 'POSTCALL')
def run_isolated_probe(source, root, name, *, mentor_home=None, replacement_source=None):
@@ -906,17 +1045,21 @@ def _catalog_native(job, root):
def _catalog_com(job, root):
import pythoncom, win32com.client, win32con, win32gui, win32process
from .mentor_adapter import _process_details
if job['operation'] == 'add':
if job.get('authority_kind') == 'management':
from .management_transaction import validate_transaction
validate_transaction(job, root.parent)
else:
validate_owned_fixture_add(job, root)
desktop, name = _current_desktop()
if name != job['desktop'] or not name.startswith('CbbManagement_'):
raise RuntimeError('拒绝在非本次私有桌面读取原生目录。')
if _process_details('LibraryManager.exe'):
raise RuntimeError('已有 Library Manager 运行,请关闭后重试。')
try:
if job['operation'] == 'add':
if job.get('authority_kind') == 'management':
from .management_transaction import validate_transaction
validate_transaction(job, root.parent)
else:
validate_owned_fixture_add(job, root)
desktop, name = _current_desktop()
if name != job['desktop'] or not name.startswith('CbbManagement_'):
raise RuntimeError('拒绝在非本次私有桌面读取原生目录。')
if _process_details('LibraryManager.exe'):
raise RuntimeError('已有 Library Manager 运行,请关闭后重试。')
except Exception as error:
error.safe_to_continue = True
raise
stop = threading.Event()
refused = []
observed = []
@@ -964,10 +1107,16 @@ def _catalog_com(job, root):
result = {'success': False, 'safe_to_continue': False}
watcher = None
try:
app = win32com.client.DispatchEx('LibraryManager.Application')
_require_management_continuation(job, root)
from .session_record import dispatch_native, update_worker_state
# Keep the launch pending until its exact native identity is recorded.
app = dispatch_native(job, lambda: win32com.client.DispatchEx('LibraryManager.Application'),
asynchronous=True)
app.Visible = True
processes = _process_details('LibraryManager.exe')
owned_pids.update(int(p[0]) for p in processes)
from .session_record import update_worker_state
update_worker_state(job, 'POSTCALL', identities=[('native', pid) for pid in owned_pids])
if len(owned_pids) != 1:
raise RuntimeError('无法唯一确认本次 Library Manager 进程。')
expected_binary = Path(job['mentor_home'])/'lm/win32/bin/LibraryManager.exe'
@@ -977,12 +1126,16 @@ def _catalog_com(job, root):
watcher.start()
if job['operation'] in ('probe_create_add', 'probe_create'):
target = prepare_probe_library(job, root)
library = app.CreateLibrary(str(target), 1)
_require_management_continuation(job, root)
library = dispatch_native(job, lambda: app.CreateLibrary(str(target), 1), asynchronous=True)
_write_json(root.parent/'probe-owner.json', {'library': str(target), 'token': job['probe_token']})
else:
if job['operation'] == 'probe_readd':
validate_probe_readd(job, root)
library = app.OpenLibrary(str(absolute_preserving_drive(job['library'])))
_require_management_continuation(job, root)
library = dispatch_native(job, lambda: app.OpenLibrary(str(absolute_preserving_drive(job['library']))),
asynchronous=True)
update_worker_state(job, 'POSTCALL')
if refused:
raise RuntimeError('目录读取需要升级中心库或选择流程;已拒绝,目标未获准自动升级。')
if library is None or Path(str(library.FullName)).resolve() != Path(job['library']).resolve():
@@ -991,9 +1144,10 @@ def _catalog_com(job, root):
if int(library.FlowType) != 1:
raise RuntimeError('Native Add requires the DxDesigner Logical-Physical library flow.')
_probe_add_dialog(app, job, desktop, owned_pids, root, library=library)
library.Close()
dispatch_native(job, lambda: library.Close(), phase='CLOSING')
library = None
library = app.OpenLibrary(str(absolute_preserving_drive(job['library'])))
_require_management_continuation(job, root)
library = dispatch_native(job, lambda: app.OpenLibrary(str(absolute_preserving_drive(job['library']))))
if (library is None or Path(str(library.FullName)).resolve() != Path(job['library']).resolve()
or int(library.FlowType) != 1):
raise RuntimeError('Native Add reopen did not confirm the exact library and flow.')
@@ -1033,14 +1187,16 @@ def _catalog_com(job, root):
result['detail'] = str(error)
result['exception'] = _exception_details(error)
finally:
if library is not None:
try: library.Close()
if library is not None and not job.get('_native_add_pending'):
try:
dispatch_native(job, lambda: library.Close(), phase='CLOSING')
except Exception as error:
result['close_detail'] = str(error)
result['close_exception'] = _exception_details(error)
library = None
if app is not None:
try: app.Quit()
if app is not None and not job.get('_native_add_pending'):
try:
dispatch_native(job, lambda: app.Quit(), phase='CLOSING')
except Exception as error:
result['quit_detail'] = str(error)
result['quit_exception'] = _exception_details(error)
@@ -1059,11 +1215,27 @@ def management_worker_main(request_path):
request_path = Path(request_path)
root = request_path.parent
job = json.loads(request_path.read_text(encoding='utf-8'))
result = {'success': False, 'safe_to_continue': False}
result = {'success': False, 'safe_to_continue': True}
try:
_, desktop = _current_desktop()
if desktop != job['desktop'] or not desktop.startswith('CbbManagement_'):
raise RuntimeError('原生管理操作仅允许在本次私有桌面执行。')
_require_management_continuation(job, root)
if job.get('cancel_signal'):
from .session_record import update_worker_state
update_worker_state(job, 'PRECALL')
from .desktop_worker import start_supervisor_watchdog
start_supervisor_watchdog(job['supervisor_pid'], cancel_path=job['cancel_signal'],
supervisor_identity=job.get('supervisor_identity'),
session_record_path=job.get('session_record_path'), operation_id=job.get('operation_id'))
if job.get('start_gate'):
deadline = time.monotonic() + 30
while not Path(job['start_gate']).is_file():
_require_management_continuation(job, root)
if time.monotonic() >= deadline:
raise RuntimeError('MANAGEMENT_CANCELLED: 启动门未放行,未启动原生操作')
time.sleep(.05)
_require_management_continuation(job, root)
if job.get('mapped_drive') and job.get('mapped_remote_root'):
from .network_path import ensure_mapped_drive_visible
ensure_mapped_drive_visible(job['mapped_drive'], job['mapped_remote_root'])
@@ -1077,6 +1249,7 @@ def management_worker_main(request_path):
raise ValueError('Native Add requires Logical-Physical.')
if engineering_identity(job['source']) != job.get('source_identity'):
raise ValueError('Staged source engineering identity changed before native Add.')
result['safe_to_continue'] = False
result = _catalog_native(job, root)
elif job['operation'] == 'remove':
adapter = NativeManagementAdapter(mentor_home=job['mentor_home'])
@@ -1090,23 +1263,47 @@ def management_worker_main(request_path):
command = [str(adapter.home/'lm/win32/bin/CleanWGLib.exe'), str(job['library']), 'F4', str(input_path)]
_write_json(root/'command.json', command)
with (root/'native-output.txt').open('wb') as output:
process = subprocess.Popen(command, stdout=output, stderr=subprocess.STDOUT,
creationflags=subprocess.CREATE_NO_WINDOW)
_write_json(root/'native-process.json', {'pid': process.pid, 'command': command})
_require_management_continuation(job, root)
from .session_record import dispatch_native, update_worker_state
result['safe_to_continue'] = False
try:
process = dispatch_native(job, lambda: subprocess.Popen(command, stdout=output,
stderr=subprocess.STDOUT, creationflags=subprocess.CREATE_NO_WINDOW), asynchronous=True)
except (FileNotFoundError, PermissionError):
result['safe_to_continue'] = True
raise
update_worker_state(job, 'NATIVE_CALL', identities=[('native', process.pid)], continuation=True)
try:
_write_json(root/'native-process.json', {'pid': process.pid, 'command': command})
except OSError:
pass # Keep supervising the native owner even when evidence storage fails.
# Never terminate this native database owner on timeout.
try:
code = process.wait(timeout=float(job['timeout_seconds']))
except subprocess.TimeoutExpired:
result.update(detail=f'TIMEOUT: CleanWGLib PID {process.pid} 尚未释放')
else:
text = (root/'native-output.txt').read_bytes().decode('mbcs', errors='replace')
result.update(success=removal_output_ok(code, text, job['name'], job['partition']),
safe_to_continue=True, detail=text, exit_code=code)
while True:
try:
code = process.wait(timeout=1)
break
except subprocess.TimeoutExpired:
continue
result['wait_exceeded'] = True
update_worker_state(job, 'POSTCALL')
result.update(safe_to_continue=True, exit_code=code)
text = (root/'native-output.txt').read_bytes().decode('mbcs', errors='replace')
result.update(success=removal_output_ok(code, text, job['name'], job['partition']),
detail=text)
else:
raise ValueError('未验证的原生管理操作')
except Exception as error:
if getattr(error, 'safe_to_continue', None) is True:
result['safe_to_continue'] = True
result['detail'] = str(error)
result['exception'] = _exception_details(error)
finally:
from .session_record import update_worker_state
_write_json(root/'result.json', result)
# FINISHED permits the orphan watchdog to exit immediately.
update_worker_state(job, 'FINISHED' if result.get('safe_to_continue') is True else 'UNCONFIRMED')
return 0 if result['success'] and result['safe_to_continue'] else 2
+2 -2
View File
@@ -124,7 +124,7 @@ def backup_project(target, destination):
handle.Close()
def attempt_recovery(request, diagnosis, evidence_root, *, progress=None):
def attempt_recovery(request, diagnosis, evidence_root, *, progress=None, stop=None):
root = Path(evidence_root)/'native-repair'
root.mkdir(exist_ok=True)
marker = root/'attempt.json'
@@ -152,7 +152,7 @@ def attempt_recovery(request, diagnosis, evidence_root, *, progress=None):
raise ValueError('REPAIR_DATABASE_IN_USE: 目标的保存/关闭助手尚未退出')
callback = (lambda event: progress({'block': request.block_name, **event})) if progress else None
record = run_native_repair(target, root, mentor_bin=request.mentor_bin,
timeout_seconds=max(60, request.timeout_seconds), progress=callback)
timeout_seconds=max(60, request.timeout_seconds), progress=callback, stop=stop)
native_status = record.get('status')
record['release'] = repair_release_state(request, target, record)
if native_status == 'REPAIRED' and record['release']['safe_to_continue'] is not True:
+101 -16
View File
@@ -9,6 +9,20 @@ import ntpath
import os
from pathlib import Path
from typing import Protocol
from functools import wraps
from threading import RLock
from weakref import WeakValueDictionary
_retained_path_leases = {}
_live_path_leases = WeakValueDictionary()
_path_lease_guard = RLock()
def _lease_locked(method):
@wraps(method)
def locked(*args, **kwargs):
with _path_lease_guard: return method(*args, **kwargs)
return locked
class NetworkPathError(RuntimeError):
@@ -45,23 +59,35 @@ class WindowsWNetBackend:
def __init__(self) -> None:
self._mpr = ctypes.WinDLL("mpr", use_last_error=True)
self._kernel32 = ctypes.WinDLL("kernel32", use_last_error=True)
self._unavailable_drives = set()
def mapping(self, drive: str) -> str | None:
size = wintypes.DWORD(32768)
buffer = ctypes.create_unicode_buffer(size.value)
error = self._mpr.WNetGetConnectionW(drive, buffer, ctypes.byref(size))
if error == 0:
return buffer.value
if error == self.ERROR_NOT_CONNECTED:
return None
raise OSError(error, f"WNetGetConnectionW({drive})")
def mappings(self) -> dict[str, str]:
result: dict[str, str] = {}
self._unavailable_drives = set()
for code in range(ord("D"), ord("Z") + 1):
drive = f"{chr(code)}:"
size = wintypes.DWORD(32768)
buffer = ctypes.create_unicode_buffer(size.value)
error = self._mpr.WNetGetConnectionW(drive, buffer, ctypes.byref(size))
if error == 0:
result[drive] = buffer.value
elif error != self.ERROR_NOT_CONNECTED:
raise OSError(error, f"WNetGetConnectionW({drive})")
try:
remote = self.mapping(drive)
except OSError:
self._unavailable_drives.add(drive)
continue
if remote:
result[drive] = remote
return result
def used_drives(self) -> set[str]:
mask = int(self._kernel32.GetLogicalDrives())
return {
return getattr(self, '_unavailable_drives', set()) | {
f"{chr(ord('A') + index)}:"
for index in range(26)
if mask & (1 << index)
@@ -88,6 +114,14 @@ def _same_remote(left: str, right: str) -> bool:
return ntpath.normcase(ntpath.normpath(left)) == ntpath.normcase(ntpath.normpath(right))
def _mapping(backend, drive):
direct = getattr(backend, 'mapping', None)
if callable(direct):
return direct(drive)
return next((value for key, value in backend.mappings().items()
if str(key).upper().rstrip('\\/') == drive.upper()), None)
def ensure_mapped_drive_visible(drive: str, remote_root: str, *,
backend: WNetBackend | None = None) -> dict[str, object]:
"""Ensure a worker process can see the mapped drive used by Mentor.
@@ -108,13 +142,11 @@ def ensure_mapped_drive_visible(drive: str, remote_root: str, *,
normalized_drive = raw_drive.upper()
backend = backend or WindowsWNetBackend()
try:
mappings = backend.mappings()
current = _mapping(backend, normalized_drive)
except OSError as error:
raise NetworkPathError(
f"NETWORK_UNAVAILABLE: 无法读取网络盘映射 {normalized_drive}: {error}"
) from error
current = next((value for key, value in mappings.items()
if str(key).upper().rstrip('\\/') == normalized_drive), None)
if current:
if _same_remote(current, raw_remote):
return {'drive': normalized_drive, 'remote_root': current, 'created': False}
@@ -180,7 +212,11 @@ class LibraryPathLease:
_closed: bool = False
_retained_for_recovery: bool = False
def __post_init__(self):
with _path_lease_guard: _live_path_leases[id(self)] = self
@classmethod
@_lease_locked
def acquire(cls, path, *, backend: WNetBackend | None = None) -> "LibraryPathLease":
raw = _normalize_input_path(path)
if not raw.startswith("\\\\"):
@@ -189,8 +225,7 @@ class LibraryPathLease:
if drive:
try:
backend = backend or WindowsWNetBackend()
remote = next((value for key, value in backend.mappings().items()
if str(key).upper().rstrip('\\/') == drive.upper()), '')
remote = _mapping(backend, drive.upper()) or ''
except (OSError, AttributeError):
remote = ''
return cls(local, local, drive.upper(), remote, False, backend)
@@ -225,10 +260,31 @@ class LibraryPathLease:
f"NETWORK_UNAVAILABLE: 无法挂载中心库 {remote_root}: {error}"
) from error
@_lease_locked
def retain_for_recovery(self) -> None:
"""Keep the Mentor drive available to a retained native session."""
self._retained_for_recovery = True
_retained_path_leases[id(self)] = self
@_lease_locked
def release_after_recovery(self, proof) -> bool:
"""Restore normal mapping lifetime after this library's release proof."""
if not isinstance(proof, dict) or proof.get('status') != 'RELEASE_CONFIRMED':
return False
candidates = [str(path) for path in (self.display_path, self.mentor_path)
if str(path).casefold().endswith('.lmc')]
drive, tail = ntpath.splitdrive(str(self.mentor_path))
if (self.remote_root and drive.upper() == self.drive.upper()
and str(self.mentor_path).casefold().endswith('.lmc')):
candidates.append(self.remote_root + tail)
library = str(proof.get('library') or '')
if not any(_same_remote(library, candidate) for candidate in candidates):
return False
self._retained_for_recovery = False
_retained_path_leases.pop(id(self), None)
return True
@_lease_locked
def transfer_ownership(self, other) -> None:
"""Keep a shared mapping alive for the currently loaded library."""
if (isinstance(other, LibraryPathLease) and not other._closed
@@ -237,16 +293,28 @@ class LibraryPathLease:
other.created, other._backend = True, self._backend
self.created = False
@_lease_locked
def close(self) -> None:
if self._closed:
return
self._closed = True
if self._retained_for_recovery or not self.created or self._backend is None or not self.drive:
if self._retained_for_recovery:
return
if not self.created or self._backend is None or not self.drive:
self._closed = True
return
for live in tuple(_live_path_leases.values()):
same_backend = (live._backend is self._backend or
isinstance(self._backend, WindowsWNetBackend) and isinstance(live._backend, WindowsWNetBackend))
if live is not self and not live._closed and same_backend:
self.transfer_ownership(live)
if not self.created:
self._closed = True
return
try:
current = self._backend.mappings().get(self.drive)
current = _mapping(self._backend, self.drive)
if current and _same_remote(current, self.remote_root):
self._backend.cancel_mapping(self.drive)
self._closed = True
except OSError as error:
raise NetworkPathError(
f"NETWORK_UNAVAILABLE: 释放临时盘符 {self.drive} 失败: {error}"
@@ -257,3 +325,20 @@ class LibraryPathLease:
def __exit__(self, exc_type, exc_value, traceback) -> None:
self.close()
def release_retained_path_leases(proof, *, current_lease=None) -> int:
"""Recover matching leases even after the UI dropped its old reference."""
released = 0
for lease in tuple(_retained_path_leases.values()):
if not lease.release_after_recovery(proof):
continue
if lease is not current_lease:
lease.transfer_ownership(current_lease)
try:
lease.close()
except NetworkPathError:
_retained_path_leases[id(lease)] = lease
raise
released += 1
return released
@@ -0,0 +1,101 @@
"""Facts from one helper invocation; process exit is not a Save result."""
from collections.abc import Sequence
def recheck_failed_release(library, block, result):
"""Refresh resource facts after a late exit; keep the original business failure."""
if result.safe_to_continue or not result.evidence_path:
return result
from pathlib import Path
from dataclasses import replace
import json
from .session_record import read_session_record, inspect_session_record
from .library_lock import release_retained_library
evidence = Path(result.evidence_path)
try:
session = Path(getattr(result, 'session_record_path', None) or evidence.with_name('session.json'))
if session.is_file():
record = read_session_record(session, evidence_root=evidence.parent)
if record.get('block') != block or Path(record.get('library', '')).resolve() != Path(library).resolve():
return result
if inspect_session_record(session, evidence_root=evidence.parent).get('status') == 'ACTIVE':
return result
else:
record = json.loads(evidence.read_text('utf-8'))
if (record.get('name') != block or record.get('operation') not in ('add', 'update', 'remove')
or Path(record.get('library', '')).resolve() != Path(library).resolve()):
return result
bound = []
for phase in ('before', 'remove', 'after-remove', 'add', 'after-add'):
candidate = evidence.parent/phase/'session.json'
if candidate.is_file():
owned = read_session_record(candidate, evidence_root=evidence.parent)
if Path(owned.get('library', '')).resolve() != Path(library).resolve():
return result
if inspect_session_record(candidate, evidence_root=evidence.parent).get('status') == 'ACTIVE':
return result
bound.append(candidate)
if not bound:
return result # A generic timeout without a bound owner is not an exit fact.
proof = release_retained_library(library, evidence.parent/'late-release', block_name=block)
if proof.get('status') != 'RELEASE_CONFIRMED' or Path(proof.get('library', '')).resolve() != Path(library).resolve():
return result
evidence.with_name('late-release-result.json').write_text(json.dumps(dict(
business_status=result.status, safe_to_continue=True, release_proof=proof),
ensure_ascii=False, default=str, indent=2), 'utf-8')
return replace(result, safe_to_continue=True, release_proof=proof)
except Exception:
return result
def classify_save_state(markers: Sequence[str], *, helper_exited: bool | None,
native_released: bool | None, forced: bool = False) -> dict:
names = {line.strip().split('|', 1)[0] for line in markers}
save_results = names & {'SAVE_OK', 'SAVE_REJECTED', 'SAVE_RESULT_UNKNOWN',
'CANCELLED_BEFORE_SAVE'}
conflicting_save = (len(save_results) > 1 or
('CANCELLED_BEFORE_SAVE' in names and 'SAVE_START' in names))
conflicting_close = {'CLOSE_OK', 'CLOSE_FAILED'} <= names
save_state = 'UNKNOWN'
if not conflicting_save:
for marker, state in (('SAVE_OK', 'COMMITTED'), ('SAVE_REJECTED', 'REJECTED'),
('CANCELLED_BEFORE_SAVE', 'NOT_STARTED')):
if marker in names:
save_state = state
close_state = 'UNKNOWN'
if not conflicting_close:
if 'CLOSE_OK' in names:
close_state = 'CLOSED'
elif 'CLOSE_FAILED' in names:
close_state = 'FAILED'
resource_state = ('RELEASED' if native_released is True and helper_exited is True else
'RETAINED' if native_released is False or helper_exited is False else
'UNKNOWN')
reasons = []
if conflicting_save or conflicting_close:
reasons.append('本轮返回标记矛盾')
if save_state == 'UNKNOWN':
reasons.append('保存返回结果无法确认')
if close_state == 'FAILED':
reasons.append('原生关闭失败')
if 'POST_SAVE_FAILED' in names:
reasons.append('保存后核对失败')
if forced:
reasons.append('会话被中断,退出不代表保存完成')
interrupted = (forced or conflicting_save or conflicting_close or
'SAVE_RESULT_UNKNOWN' in names or
(helper_exited is True and 'SAVE_START' in names and save_state == 'UNKNOWN'))
return dict(save_state=save_state, close_state=close_state,
helper_state='EXITED' if helper_exited is True else
'RUNNING' if helper_exited is False else 'UNKNOWN',
resource_state=resource_state, interrupted=interrupted,
detail=';'.join(reasons))
def stop_disposition(*, call_started: bool | None, call_returned: bool | None,
identity_confirmed: bool) -> str:
if not identity_confirmed or call_started is None:
return 'OBSERVE_UNCONFIRMED'
if call_started is False:
return 'CANCEL_BEFORE_CALL' if call_returned is False else 'OBSERVE_UNCONFIRMED'
return 'CLEANUP_RETURNED' if call_returned is True else 'WAIT_FOR_RETURN'
+49
View File
@@ -0,0 +1,49 @@
"""Cooperative FIFO admission for this application's native operations."""
from threading import Condition
class QueueCancelled(RuntimeError):
pass
class NativeQueue:
def __init__(self):
self.condition = Condition()
self.pending = []
self.active = False
def wake(self):
with self.condition: self.condition.notify_all()
def run(self, action, *, stop=None, waiting=None, ready=lambda: True, wait_state=None):
ticket = object()
acquired, notified = False, False
previous_state = None
with self.condition: self.pending.append((ticket, ready))
try:
while not acquired:
with self.condition:
if stop is not None and stop.is_set():
raise QueueCancelled('已取消排队,未启动原生操作。')
first = next((token for token, available in self.pending if available()), None)
if not self.active and first is ticket:
self.pending = [entry for entry in self.pending if entry[0] is not ticket]
self.active = acquired = True
else:
self.condition.wait(.1)
if not acquired and waiting is not None:
try:
current_state = wait_state() if wait_state is not None else None
if not notified or current_state != previous_state:
waiting()
notified, previous_state = True, current_state
except Exception:
pass
if stop is not None and stop.is_set():
raise QueueCancelled('已取消排队,未启动原生操作。')
return action()
finally:
with self.condition:
self.pending = [entry for entry in self.pending if entry[0] is not ticket]
if acquired: self.active = False
self.condition.notify_all()
+37 -10
View File
@@ -11,9 +11,20 @@ def query_owners(names, *, timeout_seconds=10):
names = tuple(names)
if not names or any(not re.fullmatch(r'[\w.-]+\.exe', name, re.I) for name in names):
raise ValueError('OWNER_QUERY_INVALID')
return _query_processes(names, timeout_seconds=timeout_seconds)
def query_process_snapshot(*, timeout_seconds=10):
"""Bound a complete process tree; retain unknown executable descendants."""
return _query_processes(None, timeout_seconds=timeout_seconds)
def _query_processes(names, *, timeout_seconds):
full_snapshot = names is None
request = {'snapshot_version': 1} if full_snapshot else {'names': names}
with tempfile.TemporaryDirectory(prefix='cbb-owner-probe-') as directory:
job = Path(directory)/'request.json'
job.write_text(json.dumps({'names': names}), encoding='utf-8')
job.write_text(json.dumps(request), encoding='utf-8')
command = ([sys.executable] if getattr(sys, 'frozen', False) else
[sys.executable, '-m', 'cbb_refresh_verify']) + ['--owner-probe', str(job)]
try:
@@ -27,13 +38,20 @@ def query_owners(names, *, timeout_seconds=10):
try:
record = json.loads(job.with_name('result.json').read_text('utf-8'))
rows = record['owners']
if result.returncode != 0 or record['names'] != list(names) or not isinstance(rows, list):
identity_matches = (type(record.get('snapshot_version')) is int and record['snapshot_version'] == 1
if full_snapshot else record.get('names') == list(names))
if result.returncode != 0 or not identity_matches or not isinstance(rows, list):
raise ValueError('invalid result')
seen = set()
for row in rows:
if (type(row['pid']) is not int or row['pid'] <= 0 or row['name'].casefold() not in {n.casefold() for n in names}
if (type(row['pid']) is not int or row['pid'] < (0 if full_snapshot else 1)
or row['pid'] in seen or not isinstance(row['name'], str) or not row['name']
or not full_snapshot and row['name'].casefold() not in {n.casefold() for n in names}
or full_snapshot and not isinstance(row['created'], str)
or not isinstance(row['command'], str) or not isinstance(row['executable'], str)
or type(row['parent']) is not int or row['parent'] < 0):
raise ValueError('invalid owner')
seen.add(row['pid'])
return rows
except (OSError, ValueError, KeyError, TypeError, AttributeError) as error:
raise RuntimeError('OWNER_QUERY_UNCONFIRMED: 进程查询未返回有效证据') from error
@@ -43,19 +61,28 @@ def owner_probe_main(job):
import pythoncom
import win32com.client
job = Path(job)
names = json.loads(job.read_text('utf-8'))['names']
if not names or any(not re.fullmatch(r'[\w.-]+\.exe', name, re.I) for name in names):
return 1
request = json.loads(job.read_text('utf-8'))
full_snapshot = request == {'snapshot_version': 1} and type(request.get('snapshot_version')) is int
names = request.get('names')
if not full_snapshot:
if (not isinstance(names, list) or not names
or any(not isinstance(name, str) or not re.fullmatch(r'[\w.-]+\.exe', name, re.I) for name in names)):
return 1
pythoncom.CoInitialize()
try:
service = win32com.client.GetObject('winmgmts:')
query = 'SELECT ProcessId,Name,CommandLine,ExecutablePath,ParentProcessId FROM Win32_Process WHERE ' + ' OR '.join(
f"Name='{name}'" for name in names)
query = ('SELECT ProcessId,Name,CommandLine,ExecutablePath,ParentProcessId' +
(',CreationDate' if full_snapshot else '') + ' FROM Win32_Process')
if not full_snapshot:
query += ' WHERE ' + ' OR '.join(f"Name='{name}'" for name in names)
rows = [{'pid': int(item.ProcessId), 'name': str(item.Name),
'command': str(item.CommandLine or ''), 'executable': str(item.ExecutablePath or ''),
'parent': int(item.ParentProcessId or 0)} for item in service.ExecQuery(query)]
'parent': int(item.ParentProcessId or 0),
**({'created': str(item.CreationDate or '')} if full_snapshot else {})}
for item in service.ExecQuery(query)]
service = None
job.with_name('result.json').write_text(json.dumps({'names': names, 'owners': rows}), encoding='utf-8')
response = {'snapshot_version': 1} if full_snapshot else {'names': names}
job.with_name('result.json').write_text(json.dumps({**response, 'owners': rows}), encoding='utf-8')
return 0
finally:
pythoncom.CoUninitialize()
+29 -15
View File
@@ -91,6 +91,35 @@ def require_document_identity(document, job):
return integration
class ProcessWatch:
"""Hold this task's native handles; repeated COM attach polls never run WMI."""
def __init__(self, pids):
self.pids = set(int(pid) for pid in pids if pid)
self.handles = None
def __enter__(self):
return self
def alive(self):
import pywintypes, win32api, win32con, win32event
if self.handles is None:
self.handles = []
try:
for pid in self.pids:
try:
self.handles.append(win32api.OpenProcess(win32con.SYNCHRONIZE, False, pid))
except pywintypes.error as error:
if error.winerror != 87: raise
except Exception:
self.__exit__(None, None, None)
raise
return any(win32event.WaitForSingleObject(h, 0) != win32con.WAIT_OBJECT_0 for h in self.handles)
def __exit__(self, *_):
for handle in self.handles or (): handle.Close()
self.handles = []
def wait_for_target_document(job, app_provider, *, process_alive, emit=lambda **event: None,
error_reader=None, sleep=time.sleep):
"""Wait for the exact launched document; the parent process owns timeout."""
@@ -131,18 +160,3 @@ def wait_for_target_document(job, app_provider, *, process_alive, emit=lambda **
last_observation = observation
sleep(.25)
raise PcbBridgeError('PCB_EXITED_BEFORE_COM', 'PCB 在 COM 文档就绪前退出')
def execution_outcome(*, save_complete: bool, pcb_exited: bool,
fatal_error: str = '') -> dict[str, object]:
if fatal_error:
code, _, detail = str(fatal_error).partition(':')
return {'status': 'FAIL', 'error_code': code.strip() or 'FATAL_DIALOG',
'stderr': detail.strip() or str(fatal_error), 'safe_to_continue': True}
if not save_complete:
return {'status': 'FAIL', 'error_code': 'SAVE_NOT_CONFIRMED',
'stderr': '保存/关闭未完成', 'safe_to_continue': True}
if not pcb_exited:
return {'status': 'FAIL', 'error_code': 'PCB_EXIT_NOT_CONFIRMED',
'stderr': '保存后未确认 PCB 退出', 'safe_to_continue': True}
return {'status': 'PASS', 'stdout': 'Project Integration 已同步;保存、关闭和 PCB 退出均已确认。'}
+154
View File
@@ -0,0 +1,154 @@
"""Explicit user answers to the owned PCB's exact read/write question."""
from contextlib import contextmanager, ExitStack
from pathlib import Path
from threading import Lock
import time
from .post_repair_consent import exact_prompt
def _prompt_identity(snapshot):
# UIA adds chrome/reorders controls and omits Win32 caption accelerator marks.
return tuple(sorted((c.handle, c.control_id, c.class_name,
c.name.replace('&', '') if c.class_name == 'Button' else c.name)
for c in snapshot.controls if c.handle and c.handle != snapshot.hwnd))
class OpenChoice:
def __init__(self, block, library):
self.block, self.library = block, library
self.decision = None
self._lock = Lock()
def choose(self, decision):
if decision not in ('continue', 'skip', 'cancel'):
raise ValueError('无效的打开工程选择')
with self._lock:
if self.decision is None:
self.decision = decision
def validate_owner(job, target_pid, launcher_pid, worker_pid):
from .owner_probe import query_process_snapshot
from .session_record import process_identity
from .mentor_adapter import is_reusable_circuit_pcb
rows = {r['pid']: r for r in query_process_snapshot()}
target, launcher = rows.get(target_pid), rows.get(launcher_pid)
if (not target or not launcher or target['parent'] != launcher_pid or launcher['parent'] != worker_pid
or not str(job.get('desktop', '')).startswith('CbbVerify_')
or job.get('interactive_input') or job.get('native_lm_verify') or job.get('diagnostic_only')):
raise ValueError('PCB_OPEN_OWNER_UNCONFIRMED')
for row in (target, launcher):
if not is_reusable_circuit_pcb(row['executable'], row['command'], job['pcb']):
raise ValueError('PCB_OPEN_TARGET_CHANGED')
return {'target': process_identity(target_pid), 'launcher': process_identity(launcher_pid)}
@contextmanager
def resource_guard(job, allowed_pids):
import win32con, win32file
from .icdb_ownership import database_directories, unreleased_addresses
from .owner_probe import query_process_snapshot
from .library_lock import _owner_is_blocking
from .verify_runner import preflight, VerifyRequest
library = Path(job['lmc_file'])
project, pcb = preflight(VerifyRequest(library, job['block_name'], Path(job['root'])))
if project.resolve() != Path(job['project']).resolve() or pcb.resolve() != Path(job['pcb']).resolve():
raise ValueError('PCB_OPEN_PROJECT_CHANGED')
roots = (project.parent, library.parent/'SymbolLibs')
def check():
if unreleased_addresses(library, project, pcb):
raise ValueError('目标仍有数据库地址记录,暂不能确认读写打开。')
for row in query_process_snapshot():
if row['pid'] not in allowed_pids and _owner_is_blocking(row, library, roots):
raise ValueError(f'目标还有其他会话,PID={row["pid"]};未确认读写打开。')
check()
paths = [library, pcb.parent/'Layout/LayoutDB.lyt',
*(d/'icdb.dat' for d in database_directories(library, project, pcb) if (d/'icdb.dat').is_file())]
with ExitStack() as stack:
for path in dict.fromkeys(paths):
handle = win32file.CreateFile(str(path), win32con.GENERIC_READ, 0, None,
win32con.OPEN_EXISTING, win32con.FILE_ATTRIBUTE_NORMAL, None)
stack.callback(handle.Close)
check()
yield
class OpenController:
def __init__(self, job):
self.job = job
self.choice = None
self.original = self.owner = None
self.sent = False
self.sent_at = 0.0
def handle(self, snapshot, snapshots, *, target_pid, launcher_pid, worker_pid, show, perform):
from .dialog_supervisor import DialogPollResult
from .management import _write_json
try:
if not exact_prompt(snapshot) or snapshot.pid != target_pid or len(snapshots) != 1:
raise ValueError('PCB_OPEN_PROMPT_CHANGED')
owner = validate_owner(self.job, target_pid, launcher_pid, worker_pid)
if self.choice is None:
self.owner, self.original = owner, snapshot
self.choice = OpenChoice(self.job['block_name'], self.job['lmc_file'])
show(self.choice)
if (owner != self.owner or snapshot.hwnd != self.original.hwnd
or _prompt_identity(snapshot) != _prompt_identity(self.original)):
raise ValueError('PCB_OPEN_IDENTITY_CHANGED')
if self.sent:
if time.monotonic()-self.sent_at < 2:
return DialogPollResult(observed_snapshots=tuple(snapshots))
raise ValueError('PCB_OPEN_ANSWER_NOT_COMPLETED')
if self.choice.decision is None:
return DialogPollResult(activity=True, observed_snapshots=tuple(snapshots))
control = 6 if self.choice.decision == 'continue' else 7
action = {'action': 'user_confirmed_pcb_open', 'channels': ['win32_button_click'],
'target': {'control_id': control}, 'desktop': self.job['desktop'],
'pid': target_pid, 'owner': owner, 'choice': self.choice.decision}
def send():
self.sent, self.sent_at = True, time.monotonic()
_write_json(Path(self.job['root'])/'pcb-open-user-choice.json', dict(
choice=self.choice.decision, project=self.job['project'], pcb=self.job['pcb'],
library=self.job['lmc_file'], owner=owner, snapshot=snapshot.to_record()))
return perform(snapshot, action, snapshot.phase)
if control == 6:
allowed = {target_pid, launcher_pid, worker_pid, self.job.get('supervisor_pid')}
with resource_guard(self.job, allowed):
result = send()
else:
result = send()
error = result.error_code
if not error and result.accepted_controls != (control,):
error = 'PCB_OPEN_ANSWER_FAILED'
if not error and control == 7:
error = 'PCB_OPEN_SKIPPED' if self.choice.decision == 'skip' else 'USER_CANCELLED'
return DialogPollResult(activity=True, action_sent=True, snapshot=snapshot,
error_code=error, detail=result.detail or ('用户跳过本项。' if control == 7 else ''),
accepted_controls=result.accepted_controls, channel=result.channel,
action=action['action'], observed_snapshots=tuple(snapshots))
except Exception as error:
return DialogPollResult(error_code='DATABASE_RECOVERY_REQUIRED', detail=str(error),
snapshot=snapshot, observed_snapshots=tuple(snapshots))
def validate_action_window(snapshot, action):
from dataclasses import replace
import win32gui, win32process
from .native_repair_ui import _current_desktop_name
from .dialog_models import ControlSnapshot
from .session_record import process_identity
if (not exact_prompt(snapshot) or action.get('target', {}).get('control_id') not in (6, 7)
or snapshot.pid != action.get('pid') or _current_desktop_name() != action.get('desktop')
or not str(action.get('desktop', '')).startswith('CbbVerify_')
or win32process.GetWindowThreadProcessId(snapshot.hwnd)[1] != snapshot.pid
or process_identity(snapshot.pid) != action.get('owner', {}).get('target')):
raise ValueError('PCB_OPEN_LIVE_OWNER_CHANGED')
children = []
win32gui.EnumChildWindows(snapshot.hwnd, lambda h, _: children.append(int(h)), None)
controls = tuple(ControlSnapshot(h, win32gui.GetDlgCtrlID(h) & 0xffff,
win32gui.GetClassName(h), win32gui.GetWindowText(h), enabled=bool(win32gui.IsWindowEnabled(h))) for h in children)
if not exact_prompt(replace(snapshot, controls=controls)) or any(
c.handle != next(x.handle for x in snapshot.controls if x.control_id == c.control_id)
for c in controls if c.class_name == 'Button'):
raise ValueError('PCB_OPEN_LIVE_PROMPT_CHANGED')
+23 -9
View File
@@ -37,10 +37,11 @@ def _choice(snapshot):
class RestoreController:
def __init__(self, job):
def __init__(self, job, *, user_open=False):
from .post_repair_consent import _validate_grant
self.job = job
self.grant = _validate_grant(job)
self.grant = None if user_open else _validate_grant(job)
self.owner = None
self.prompt = None
self.original = None
self.selection_sent = self.confirm_sent = self.failed = False
@@ -60,13 +61,20 @@ class RestoreController:
or snapshot.pid != self.original.pid
or _candidate_key(choice['candidates']) != _candidate_key(self.prompt.candidates)):
raise ValueError('PCB_VERSION_CANDIDATES_CHANGED')
pre = json.loads((Path(self.job['root'])/'post-repair-prelaunch.json').read_text(encoding='utf-8'))
if (pre.get('nonce') != self.grant['nonce'] or pre.get('desktop') != self.job['desktop']
or pre.get('worker_pid') != worker_pid or pre.get('supervisor_pid') != self.job['supervisor_pid']
or pre.get('addresses') != []
or not all(consent._same(pre[key], self.grant[key]) for key in ('project', 'pcb', 'library'))):
raise ValueError('PCB_VERSION_PRELAUNCH_PROOF_INVALID')
consent._validate_current_processes(self.job, pre, target_pid, launcher_pid, worker_pid, companion_pids)
if self.grant is None:
from .pcb_open_consent import validate_owner
owner = validate_owner(self.job, target_pid, launcher_pid, worker_pid)
if self.owner is not None and owner != self.owner:
raise ValueError('PCB_VERSION_OWNER_CHANGED')
self.owner = owner
else:
pre = json.loads((Path(self.job['root'])/'post-repair-prelaunch.json').read_text(encoding='utf-8'))
if (pre.get('nonce') != self.grant['nonce'] or pre.get('desktop') != self.job['desktop']
or pre.get('worker_pid') != worker_pid or pre.get('supervisor_pid') != self.job['supervisor_pid']
or pre.get('addresses') != []
or not all(consent._same(pre[key], self.grant[key]) for key in ('project', 'pcb', 'library'))):
raise ValueError('PCB_VERSION_PRELAUNCH_PROOF_INVALID')
consent._validate_current_processes(self.job, pre, target_pid, launcher_pid, worker_pid, companion_pids)
if self.prompt is None:
self.original = snapshot
self.prompt = RestoreChoice(self.job['block_name'], self.job['lmc_file'], choice['candidates'])
@@ -91,6 +99,8 @@ class RestoreController:
action = {'action': 'user_selected_pcb_version', 'channels': ['win32_button_click'],
'target': {'control_id': control}, 'desktop': self.job['desktop'], 'pid': target_pid,
'choice': selected['kind'], 'candidates': list(self.prompt.candidates)}
if self.owner is not None:
action['owner'] = self.owner
self.selection_sent = True
self.confirm_sent = confirming
self.sent_at = time.monotonic()
@@ -119,6 +129,10 @@ def validate_action_window(snapshot, action):
import win32gui, win32process
from .native_repair_ui import _current_desktop_name
from .dialog_models import ControlSnapshot
if action.get('owner') is not None:
from .session_record import process_identity
if process_identity(snapshot.pid) != action['owner']['target']:
raise ValueError('PCB_VERSION_LIVE_PROCESS_CHANGED')
if (not str(action.get('desktop', '')).startswith('CbbVerify_')
or _current_desktop_name() != action.get('desktop') or snapshot.pid != action.get('pid')
or win32process.GetWindowThreadProcessId(snapshot.hwnd)[1] != snapshot.pid
+784
View File
@@ -0,0 +1,784 @@
"""Preview and cooperatively close the exact native sessions a user confirms."""
from contextlib import contextmanager
import json
import hashlib
import os
from pathlib import Path
import re
import time
from uuid import uuid4
from .owner_probe import query_process_snapshot
from .native_engineering_identity import _OWNER_NAMES
_NAMES = {'expeditionpcb.exe', 'librarymanager.exe', 'icdbprojectbackup.exe',
'viewdraw.exe', 'icdbnetserver.exe', 'icdbserver.exe',
'reuseblockimporter.exe', 'cleanwglib.exe', 'cscript.exe', 'wscript.exe'} | _OWNER_NAMES
_WORKER_NAMES = {'python.exe', 'pythonw.exe', 'cbbrefreshverify.exe'}
_WORKER_FLAGS = ('--verify-worker', '--repair-worker', '--management-worker', '--native-backup-worker')
_FORCE_CLIENT_NAMES = {'expeditionpcb.exe', 'librarymanager.exe', 'viewdraw.exe', 'icdbprojectbackup.exe'}
def force_block_reason(row):
"""A library path proves scope, never exclusive ownership of a server."""
if row.get('runtime_phase') in ('NATIVE_CALL', 'CLOSING', 'UNCONFIRMED'):
return '本工具的保存或原生调用尚未确认返回,请等待或查看原生提示'
if row.get('scope_confirmed') is not True:
return '进程归属未确认,不允许强制结束'
name = str(row.get('name', '')).casefold()
if name not in _FORCE_CLIENT_NAMES:
return '服务器或其他受保护进程仅供观察,不允许强制结束'
if (not row.get('created_handle') or not row.get('image')
or Path(row['image']).name.casefold() != name):
return '客户端身份未确认,不允许强制结束'
if (row.get('force_user_confirmed') is not True or not row.get('owner_sid')
or not isinstance(row.get('session_id'), int)):
return '未确认属于当前用户和 Windows 会话,不允许强制结束'
activity = row.get('native_activity', {})
if activity.get('status') in ('ACTIVE', 'UNCONFIRMED'):
return activity.get('detail', '原生操作状态未确认')+';请查看进展或请求原生取消后重新检查'
if name == 'icdbprojectbackup.exe' and activity.get('status') != 'IDLE':
return '备份/恢复/修复状态未确认,请重新检查'
return ''
def native_activity(name, windows, *, incomplete=None):
if incomplete or any(w['class'] == 'UNCONFIRMED' for w in windows):
return {'status': 'UNCONFIRMED', 'detail': '原生窗口读取不完整'}
if any((name.casefold() == 'icdbprojectbackup.exe' and not w['enabled'])
or any(c['class'].casefold() == 'msctls_progress32' for c in w['children']) for w in windows):
return {'status': 'ACTIVE', 'detail': '原生窗口显示操作进行中'}
if name.casefold() == 'icdbprojectbackup.exe':
main = [w for w in windows if w['class'].startswith('Afx:')
and w['title'].startswith('iCDB Project Backup')]
output = main[0].get('output') if len(main) == 1 else None
if not isinstance(output, str) or not output.strip():
return {'status': 'UNCONFIRMED', 'detail': '无法确认备份/恢复/修复输出'}
active = list(re.finditer(r'(?i)repairing|restoring|backing up|creating (?:a )?backup', output))
if active and max(output.rfind('Backup information loaded correctly.'),
output.rfind('Repair project error [')) < active[-1].start():
return {'status': 'ACTIVE', 'detail': '原生备份/恢复/修复尚未确认结束', 'output': output}
return {'status': 'IDLE', 'detail': '未观察到进行中的原生操作'}
@contextmanager
def _process_handle(pid, *, terminate=False):
import win32api, win32con
handle = win32api.OpenProcess(win32con.SYNCHRONIZE | 0x1000 | (1 if terminate else 0), False, int(pid))
try:
yield handle
finally:
handle.Close()
def _handle_identity(handle):
import ctypes
from ctypes import wintypes
import win32process
kernel = ctypes.WinDLL('kernel32', use_last_error=True)
kernel.QueryFullProcessImageNameW.argtypes = [wintypes.HANDLE, wintypes.DWORD,
wintypes.LPWSTR, ctypes.POINTER(wintypes.DWORD)]
kernel.QueryFullProcessImageNameW.restype = wintypes.BOOL
buffer = ctypes.create_unicode_buffer(32768)
length = wintypes.DWORD(len(buffer))
if not kernel.QueryFullProcessImageNameW(int(handle), 0, buffer, ctypes.byref(length)):
raise ctypes.WinError(ctypes.get_last_error())
return {'created_handle': win32process.GetProcessTimes(handle)['CreationTime'].isoformat(),
'image': buffer.value}
def _process_identity(row):
with _process_handle(row['pid']) as handle:
identity = _handle_identity(handle)
try:
identity.update(_force_ownership(handle, row['pid']))
except Exception as error:
identity.update(force_user_confirmed=False, owner_error=str(error))
return identity
def _force_ownership(handle, pid):
import win32api, win32con, win32process, win32security, win32ts
if win32process.GetProcessId(handle) != pid:
raise ValueError('进程身份已变化')
def owner(process):
token = win32security.OpenProcessToken(process, win32con.TOKEN_QUERY)
try:
return win32security.ConvertSidToStringSid(
win32security.GetTokenInformation(token, win32security.TokenUser)[0])
finally:
token.Close()
sid = owner(handle)
session = win32ts.ProcessIdToSessionId(pid)
return {'owner_sid': sid, 'session_id': session, 'force_user_confirmed':
sid == owner(win32api.GetCurrentProcess())
and session == win32ts.ProcessIdToSessionId(os.getpid())}
class _WindowSnapshot(dict):
def __init__(self, pids):
super().__init__((pid, []) for pid in pids)
self.incomplete = {}
def _process_threads(pids):
import ctypes
from ctypes import wintypes
class ThreadEntry(ctypes.Structure):
_fields_ = [('size', wintypes.DWORD), ('usage', wintypes.DWORD), ('tid', wintypes.DWORD),
('pid', wintypes.DWORD), ('base', wintypes.LONG), ('delta', wintypes.LONG),
('flags', wintypes.DWORD)]
kernel = ctypes.WinDLL('kernel32', use_last_error=True)
kernel.CreateToolhelp32Snapshot.argtypes = [wintypes.DWORD, wintypes.DWORD]
kernel.CreateToolhelp32Snapshot.restype = wintypes.HANDLE
kernel.CloseHandle.argtypes = [wintypes.HANDLE]
for name in ('Thread32First', 'Thread32Next'):
function = getattr(kernel, name)
function.argtypes = [wintypes.HANDLE, ctypes.POINTER(ThreadEntry)]
function.restype = wintypes.BOOL
snapshot = kernel.CreateToolhelp32Snapshot(4, 0) # TH32CS_SNAPTHREAD
if snapshot == ctypes.c_void_p(-1).value:
raise ctypes.WinError(ctypes.get_last_error())
result = {pid: [] for pid in pids}
try:
entry = ThreadEntry(); entry.size = ctypes.sizeof(entry)
more = kernel.Thread32First(snapshot, ctypes.byref(entry))
while more:
if entry.pid in result:
result[entry.pid].append(int(entry.tid))
entry.size = ctypes.sizeof(entry)
more = kernel.Thread32Next(snapshot, ctypes.byref(entry))
if ctypes.get_last_error() != 18: # ERROR_NO_MORE_FILES
raise ctypes.WinError(ctypes.get_last_error())
finally:
kernel.CloseHandle(snapshot)
return result
def _thread_desktop_name(tid):
import ctypes
from ctypes import wintypes
user = ctypes.WinDLL('user32', use_last_error=True)
user.GetThreadDesktop.argtypes = [wintypes.DWORD]
user.GetThreadDesktop.restype = wintypes.HANDLE
user.GetUserObjectInformationW.argtypes = [wintypes.HANDLE, ctypes.c_int, wintypes.LPVOID,
wintypes.DWORD, ctypes.POINTER(wintypes.DWORD)]
user.GetUserObjectInformationW.restype = wintypes.BOOL
ctypes.set_last_error(0)
desktop = user.GetThreadDesktop(tid)
if not desktop:
if error := ctypes.get_last_error():
raise ctypes.WinError(error)
return None # A non-GUI thread need not have an assigned desktop.
# GetThreadDesktop returns a borrowed handle; never CloseDesktop it.
name, needed = ctypes.create_unicode_buffer(512), wintypes.DWORD()
if not user.GetUserObjectInformationW(desktop, 2, name, ctypes.sizeof(name), ctypes.byref(needed)):
raise ctypes.WinError(ctypes.get_last_error())
return name.value
def _window_error(snapshot, pid):
return getattr(snapshot, 'incomplete', {}).get(pid)
def _desktop_names():
import ctypes
from ctypes import wintypes
user = ctypes.WinDLL('user32', use_last_error=True)
callback_type = ctypes.WINFUNCTYPE(wintypes.BOOL, wintypes.LPWSTR, wintypes.LPARAM)
user.GetProcessWindowStation.restype = wintypes.HANDLE
user.EnumDesktopsW.argtypes = [wintypes.HANDLE, callback_type, wintypes.LPARAM]
names = []
@callback_type
def collect(name, _):
names.append(name)
return len(names) <= 128
ctypes.set_last_error(0)
if not user.EnumDesktopsW(user.GetProcessWindowStation(), collect, 0) or len(names) > 128:
raise OSError('无法完整枚举当前 Windows 窗口站桌面')
return names
@contextmanager
def _window_desktop(name, *, write=False):
"""Keep the desktop accessible while reading or messaging its HWNDs."""
if not name:
yield None
return
import ctypes
from ctypes import wintypes
user = ctypes.WinDLL('user32', use_last_error=True)
user.OpenDesktopW.argtypes = [wintypes.LPCWSTR, wintypes.DWORD, wintypes.BOOL, wintypes.DWORD]
user.OpenDesktopW.restype = wintypes.HANDLE
user.CloseDesktop.argtypes = [wintypes.HANDLE]
desktop = user.OpenDesktopW(name, 0, False, 1 | (128 if write else 0))
if not desktop:
raise ctypes.WinError(ctypes.get_last_error())
try:
# OpenDesktop alone does not map a foreign desktop's HWNDs into this
# process. Enumerate while its handle stays open before any ownership
# checks or messages; otherwise GetWindowThreadProcessId returns 0.
_desktop_windows(desktop)
yield desktop
finally:
user.CloseDesktop(desktop)
def _desktop_windows(desktop):
import ctypes
from ctypes import wintypes
user = ctypes.WinDLL('user32', use_last_error=True)
callback_type = ctypes.WINFUNCTYPE(wintypes.BOOL, wintypes.HWND, wintypes.LPARAM)
user.EnumDesktopWindows.argtypes = [wintypes.HANDLE, callback_type, wintypes.LPARAM]
windows = []
@callback_type
def collect(hwnd, _):
windows.append(int(hwnd))
return len(windows) <= 4096
ctypes.set_last_error(0)
complete = user.EnumDesktopWindows(desktop, collect, 0)
error = ctypes.get_last_error()
# Windows also returns zero with ERROR_SUCCESS for an empty desktop.
if (not complete and (error or windows)) or len(windows) > 4096:
raise OSError('无法完整读取原生桌面窗口')
return windows
def _post_window_message(window, hwnd, pid, message):
import win32gui, win32process
with _window_desktop(window.get('desktop', ''), write=True):
if win32process.GetWindowThreadProcessId(hwnd)[1] != pid:
raise ValueError('窗口归属已变化,未发送请求')
win32gui.PostMessage(hwnd, message, 0, 0)
def _process_windows(pids):
result = _WindowSnapshot(pids)
if not pids: return result
import win32api, win32con, win32gui, win32process
from .native_repair_ui import _native_output
observed = set()
def capture(hwnd, _, *, pid, desktop_name):
if win32process.GetWindowThreadProcessId(hwnd)[1] != pid:
raise ValueError('窗口所属进程已变化')
if hwnd in observed or not win32gui.IsWindowVisible(hwnd):
return
if desktop_name is None:
raise ValueError('窗口所属桌面无法读取')
observed.add(hwnd)
children, documents = [], []
def child(handle, _):
kind = win32gui.GetClassName(handle)
if kind == 'MDIClient' and win32gui.GetWindow(handle, win32con.GW_CHILD):
documents.append(handle)
if win32gui.IsWindowVisible(handle):
children.append({'hwnd': int(handle), 'class': kind,
'id': win32gui.GetDlgCtrlID(handle) & 0xffff,
'text': win32gui.GetWindowText(handle), 'enabled': bool(win32gui.IsWindowEnabled(handle))})
win32api.SetLastError(0)
win32gui.EnumChildWindows(hwnd, child, None)
window = {'hwnd': int(hwnd), 'title': win32gui.GetWindowText(hwnd),
'desktop': desktop_name,
'class': win32gui.GetClassName(hwnd), 'enabled': bool(win32gui.IsWindowEnabled(hwnd)),
'children': children, 'documents': documents, 'rect': list(win32gui.GetWindowRect(hwnd))}
if any(c['class'].casefold().startswith('richedit') for c in children):
window['output'] = _native_output([window])
result[pid].append(window)
try:
threads = _process_threads(pids)
except Exception as error:
result.incomplete = {pid: str(error)[:500] for pid in pids}
return result
desktops, unresolved = {}, set()
for pid in pids:
if not threads.get(pid):
result.incomplete[pid] = '进程线程已退出或暂时无法读取'
for tid in threads.get(pid, ()):
try:
desktop = _thread_desktop_name(tid)
if desktop:
desktops.setdefault(desktop, set()).add(pid)
except Exception as error:
unresolved.add(pid)
if pid not in unresolved and not any(pid in members for members in desktops.values()):
result.incomplete.setdefault(pid, '窗口所属桌面暂时无法确认')
if unresolved:
# Native PCB threads can deny GetThreadDesktop even when their named
# desktop is readable. Complete a bounded station scan for those PIDs.
try:
names = _desktop_names()
if not names: raise OSError('当前窗口站未返回桌面')
for name in names:
desktops.setdefault(name, set()).update(unresolved)
except Exception as error:
for pid in unresolved:
result.incomplete.setdefault(pid, str(error)[:500])
# Filter by PID before reading window data, including on fallback desktops.
for name, members in desktops.items():
try:
with _window_desktop(name) as desktop:
for hwnd in _desktop_windows(desktop):
pid = None
try:
pid = win32process.GetWindowThreadProcessId(hwnd)[1]
if pid in members:
capture(hwnd, None, pid=pid, desktop_name=name)
except Exception as error:
for affected in (members if pid is None else {pid} & members):
result.incomplete.setdefault(affected, str(error)[:500])
except Exception as error:
for pid in members:
result.incomplete.setdefault(pid, str(error)[:500])
return result
def _project_name(row, library):
from .failed_session import _command_args
for arg in _command_args(row['command'])[1:]:
try:
relative = Path(arg.split('=', 1)[-1]).relative_to(library.parent/'ReusableBlocksLP')
if len(relative.parts) >= 2 and relative.parts[1].casefold() == 'work':
return relative.parts[0]
except ValueError:
continue
return '当前库关联会话'
def close_target(name, windows, *, incomplete=None):
"""No save/restore approval and no close against an unknown modal/window."""
if incomplete:
return None, '该会话窗口读取不完整,请重新检查:'+str(incomplete)
main = [w for w in windows if w['class'].startswith('Afx:') and (
name.casefold() == 'icdbprojectbackup.exe' and w['title'].startswith('iCDB Project Backup')
or name.casefold() == 'librarymanager.exe' and w['title'].startswith('Library Manager')
or name.casefold() == 'expeditionpcb.exe' and 'Expedition PCB' in w['title'])]
if len(main) != 1:
return None, '未确认空闲主窗口;有工程时请先在原生窗口保存并关闭'
window = main[0]
for other in windows:
if other is window: continue
benign = (not other['title'] and other['class'] in
('UAC_InputIndicatorOverlayWnd', 'UAC Input Indicator'))
benign = benign or (other['class'] == 'Internet Explorer_Hidden' and not other['title']
and not other.get('children') and other.get('rect', [1, 1, 2, 2])[0:2] == other.get('rect', [1, 1, 2, 2])[2:4])
if not benign: return None, '存在原生弹窗,请先处理该窗口后重新检查'
if not window['enabled']:
return None, '原生程序正在工作,请稍后重新检查'
if any(c['class'].casefold() == 'msctls_progress32' for c in window['children']):
return None, '原生操作尚未结束'
if name.casefold() == 'icdbprojectbackup.exe':
output = window.get('output')
if not isinstance(output, str): return None, '原生输出尚无法确认,请重新检查'
active = list(re.finditer(r'(?i)repairing|restoring|backing up|creating (?:a )?backup', output))
last_load = output.rfind('Backup information loaded correctly.')
failure = output.rfind('Repair project error [')
if active and max(last_load, failure) < active[-1].start():
return None, '备份、恢复或修复需要在原生窗口确认结束后再释放'
return window, ('将正常关闭工程;未保存时等待确认' if window.get('documents') else '可正常关闭')
def _save_question(windows, *, incomplete=None):
if incomplete:
return None
for window in windows:
text = '\n'.join(c['text'] for c in window['children'] if c['class'] == 'Static')
buttons = {c['id']: c for c in window['children'] if c['class'] == 'Button' and c['enabled']}
if (window['class'] == '#32770' and 'Expedition' in window['title']
and re.search(r'\bsave\b', text, re.I) and re.search(r'changes|modified', text, re.I)
and set(buttons) == {6, 7, 2}):
return {'hwnd': window['hwnd'], 'text': text, 'buttons': buttons,
'desktop': window.get('desktop', '')}
return None
def _cancel_question(name, windows, *, incomplete=None):
if (name.casefold() != 'icdbprojectbackup.exe'
or native_activity(name, windows, incomplete=incomplete)['status'] != 'ACTIVE'):
return None
candidates = []
for window in windows:
if not (window['class'].startswith('Afx:') or window['class'] == '#32770'):
continue
if not window['title'].startswith('iCDB Project Backup') or not window['enabled']:
continue
for button in window['children']:
if (button['class'] == 'Button' and button.get('id') == 2 and button.get('enabled')
and button.get('text', '').replace('&', '').strip() in ('Cancel', '取消')):
candidates.append({'hwnd': window['hwnd'], 'button': button,
'title': window['title'],
'desktop': window.get('desktop', '')})
return candidates[0] if len(candidates) == 1 else None
def inspect_residual_sessions(library, *, block_names=()):
from .library_lock import _owner_matches_scope, blocker_path, checkpoint_record, has_retained_library
library = Path(library).resolve()
if library.suffix.casefold() != '.lmc':
raise ValueError('请选择中心库 .lmc 文件')
if not library.is_file():
raise ValueError('中心库路径当前不可见,请检查文件是否存在及网络盘映射后重新检查')
blocks = tuple(dict.fromkeys(str(name) for name in block_names))
if any(not name or name in ('.', '..') or any(c in name for c in '\\/:*?"<>|') for name in blocks):
raise ValueError('残留检查的工程名称无效')
roots = tuple(library.parent/'ReusableBlocksLP'/b/'Work' for b in blocks) + (library.parent/'SymbolLibs',)
if not blocks: roots = (library.parent,)
candidates = []
for row in query_process_snapshot():
if row['pid'] == os.getpid(): continue
native = row['name'].casefold() in _NAMES
worker = (row['name'].casefold() in _WORKER_NAMES
and any(flag in row['command'].casefold() for flag in _WORKER_FLAGS))
if not (native or worker): continue
scope = _owner_matches_scope(row, library, roots)
if scope is not False:
candidates.append({**row, 'scope_confirmed': scope is True})
windows = _process_windows({r['pid'] for r in candidates})
from .session_record import active_write_records
runtime_writes = active_write_records(library)
sessions = []
for row in candidates:
item = dict(row)
item['library'] = str(library)
if runtime_writes and row['scope_confirmed']:
item['runtime_phase'] = runtime_writes[0]['phase']
item['block_names'] = list(blocks)
error = _window_error(windows, row['pid'])
item.update(windows_incomplete=bool(error), window_read_error=error)
try:
item.update(_process_identity(row))
target, detail = close_target(row['name'], windows.get(row['pid'], []), incomplete=error) if row['scope_confirmed'] else (
None, '归属未确认,仅作观察;将检查当前库实际占用。')
item.update(can_close=target is not None, detail=detail)
except Exception as identity_error:
item.update(can_close=False, detail=f'进程已变化或暂时无法确认:{identity_error}')
item['project'] = _project_name(row, library) if row['scope_confirmed'] else '归属未确认'
item['prompt'] = _save_question(windows.get(row['pid'], []), incomplete=error) if row['scope_confirmed'] else None
observed = windows.get(row['pid'], [])
item['windows'] = observed
item['desktops'] = sorted({w.get('desktop', '') for w in observed if w.get('desktop')})
item['native_activity'] = native_activity(row['name'], observed, incomplete=error)
item['cancel_prompt'] = (_cancel_question(row['name'], observed, incomplete=error)
if row['scope_confirmed'] and item.get('force_user_confirmed') is True else None)
if any(name.startswith('Cbb') for name in item['desktops']) and not item['can_close']:
item['detail'] += ';窗口在后台,可查看原生提示并使用已提供的保存/取消操作'
item['force_block_reason'] = force_block_reason(item)
item['force_allowed'] = not item['force_block_reason']
if item['force_block_reason']:
item['detail'] += ';'+item['force_block_reason']
sessions.append(item)
previous = []
marker = blocker_path(library)
for path in (marker, marker.with_suffix('.pending')):
if path.exists():
record = checkpoint_record(path)
previous.append({'path': str(path), 'recognized': record is not None,
'status': record.get('status', 'UNRECOGNIZED') if record else 'UNRECOGNIZED',
'time': record.get('time', '') if record else '',
'pid': record.get('pid') if record else None})
if has_retained_library(library):
previous.append({'path': str(library), 'status': 'RETAINED_IN_MEMORY', 'time': '',
'pid': os.getpid(), 'kind': 'in_memory'})
return {'library': str(library), 'block_names': list(blocks), 'sessions': sessions,
'previous_records': previous}
def _request_close(row, *, stop=None):
import win32con, win32event
with _process_handle(row['pid']) as handle:
if win32event.WaitForSingleObject(handle, 0) == win32con.WAIT_OBJECT_0:
return {'release_status': 'RELEASED', 'detail': '已退出'}
identity = _handle_identity(handle)
if identity != {key: row.get(key) for key in ('created_handle', 'image')}:
return {'release_status': 'IDENTITY_CHANGED', 'detail': '进程身份已变化,未关闭'}
snapshot = _process_windows({row['pid']})
windows, error = snapshot.get(row['pid'], []), _window_error(snapshot, row['pid'])
if error:
return {'release_status': 'WAITING', 'detail': '该会话窗口读取不完整,请重新检查:'+error}
if row['name'].casefold() == 'expeditionpcb.exe':
from .failed_session import _startup_snapshot, _cancel_target
for window in windows:
if window['class'] != '#32770': continue
with _window_desktop(window.get('desktop', '')):
target = _cancel_target(_startup_snapshot(window['hwnd'], row['pid']))
if target:
_post_window_message(window, target[1].handle, row['pid'], win32con.BM_CLICK)
deadline = time.monotonic()+3
while time.monotonic() < deadline:
if win32event.WaitForSingleObject(handle, 100) == win32con.WAIT_OBJECT_0:
return {'release_status': 'RELEASED', 'detail': '已取消启动并退出'}
if stop is not None and stop.is_set():
return {'release_status': 'CANCELLED', 'detail': '已停止释放'}
snapshot = _process_windows({row['pid']})
windows, error = snapshot.get(row['pid'], []), _window_error(snapshot, row['pid'])
if error:
break
if close_target(row['name'], windows)[0] is not None: break
break
target, detail = close_target(row['name'], windows, incomplete=error)
if target is None: return {'release_status': 'WAITING', 'detail': detail}
if stop is not None and stop.is_set():
return {'release_status': 'CANCELLED', 'detail': '已停止释放'}
_post_window_message(target, target['hwnd'], row['pid'], win32con.WM_CLOSE)
deadline = time.monotonic()+3
while time.monotonic() < deadline:
if win32event.WaitForSingleObject(handle, 100) == win32con.WAIT_OBJECT_0:
return {'release_status': 'RELEASED', 'detail': '已正常退出'}
if stop is not None and stop.is_set(): break
return {'release_status': 'WAITING', 'detail': '已请求正常关闭;请处理原生窗口,完成后重新检查'}
def _force_close(row, *, stop=None):
"""Separate UI consent permits ending only this exact, scoped process handle."""
import win32api, win32con, win32event
if stop is not None and stop.is_set():
return {'release_status': 'CANCELLED', 'detail': '已停止释放'}
if row.get('library'):
from .session_record import active_write_records
if active_write_records(row['library']):
return {'release_status': 'WAITING', 'detail': '保存或原生调用尚未确认返回,未强制结束'}
if reason := force_block_reason(row):
return {'release_status': 'WAITING', 'detail': reason}
with _process_handle(row['pid'], terminate=True) as handle:
if win32event.WaitForSingleObject(handle, 0) == win32con.WAIT_OBJECT_0:
return {'release_status': 'RELEASED', 'detail': '已退出'}
if _handle_identity(handle) != {k: row.get(k) for k in ('created_handle', 'image')}:
return {'release_status': 'IDENTITY_CHANGED', 'detail': '进程身份已变化,未结束'}
ownership = _force_ownership(handle, row['pid'])
if (not ownership['force_user_confirmed'] or
any(ownership[k] != row.get(k) for k in ('owner_sid', 'session_id'))):
return {'release_status': 'IDENTITY_CHANGED', 'detail': '用户或 Windows 会话已变化,未结束'}
snapshot = _process_windows({row['pid']})
activity = native_activity(row['name'], snapshot.get(row['pid'], []),
incomplete=_window_error(snapshot, row['pid']))
if reason := force_block_reason({**row, 'native_activity': activity}):
return {'release_status': 'WAITING', 'detail': reason, 'native_activity': activity}
if stop is not None and stop.is_set():
return {'release_status': 'CANCELLED', 'detail': '已停止释放,未强制结束'}
win32api.TerminateProcess(handle, 1)
exited = win32event.WaitForSingleObject(handle, 3000) == win32con.WAIT_OBJECT_0
return {'release_status': 'RELEASED' if exited else 'WAITING',
'detail': '已强制结束客户端,工程待检查,未执行修复' if exited else '已请求结束客户端,等待退出'}
def _finish_library_release(library, evidence):
from .library_lock import release_retained_library
return release_retained_library(library, evidence)
def _request_native_cancel(row, *, stop=None):
"""Only click the exact currently visible Cancel control the user confirmed."""
import win32con
if row.get('scope_confirmed') is not True:
raise ValueError('取消目标归属未确认,请重新检查')
fresh = next((r for r in query_process_snapshot() if r['pid'] == row['pid']), None)
if fresh is None or any(fresh.get(k) != row.get(k) for k in ('created', 'command', 'executable', 'name')):
raise ValueError('取消目标身份已变化,请重新检查')
with _process_handle(row['pid']) as handle:
if _handle_identity(handle) != {k: row.get(k) for k in ('created_handle', 'image')}:
raise ValueError('取消目标身份已变化,请重新检查')
owner = _force_ownership(handle, row['pid'])
if not owner['force_user_confirmed'] or any(owner[k] != row.get(k) for k in ('owner_sid', 'session_id')):
raise ValueError('取消目标用户/会话已变化')
snapshot = _process_windows({row['pid']})
prompt = _cancel_question(row['name'], snapshot.get(row['pid'], []),
incomplete=_window_error(snapshot, row['pid']))
if not prompt or prompt != row.get('cancel_prompt'):
raise ValueError('原生取消按钮已变化,请重新检查')
button = prompt['button']
if stop is not None and stop.is_set():
return {'release_status': 'CANCELLED', 'detail': '已停止,未请求原生取消'}
_post_window_message(prompt, button['hwnd'], row['pid'], win32con.BM_CLICK)
return {'release_status': 'WAITING', 'detail': '已请求原生取消,需重新检查实际结果;未强制结束'}
def request_native_cancel(row, library, evidence_dir, *, stop=None):
# Reuse serialized admission, consent and incremental evidence for cancellation.
plan = {'library': str(library), 'sessions': [row], 'block_names': row.get('block_names', [])}
return release_residual_sessions(plan, evidence_dir, stop=stop, cancel_native=True)
def _session_evidence(row):
"""Keep the full preview in memory, but bound repeatedly written Output."""
value = dict(row)
if 'native_activity' in value:
value['native_activity'] = {k: v for k, v in value['native_activity'].items() if k != 'output'}
if 'windows' in value:
value['windows'] = []
for window in row['windows']:
saved = {k: window[k] for k in ('hwnd', 'title', 'class', 'desktop', 'enabled') if k in window}
output = window.get('output')
if isinstance(output, str):
data = output.encode('utf-8')
saved.update(output_excerpt=data[:4096].decode('utf-8', errors='ignore'),
output_bytes=len(data), output_sha256=hashlib.sha256(data).hexdigest(),
output_truncated=len(data) > 4096)
value['windows'].append(saved)
return value
def release_targets_complete(rows):
return all(row.get('release_status') == 'RELEASED' or (
row.get('scope_confirmed') is False and row.get('release_status') == 'OBSERVED') for row in rows)
def _release_residual_sessions(plan, evidence_dir, *, stop=None, force=False, cancel_native=False):
library = Path(plan['library']).resolve()
root = Path(evidence_dir).resolve()/('sessions-'+uuid4().hex[:12])
if root.is_relative_to(library.parent) or library.parent.is_relative_to(root):
raise ValueError('释放记录需位于中心库之外')
root.mkdir(parents=True)
result = {'status': 'PARTIAL', 'library': str(library), 'sessions': [
{**row, 'release_status': 'NOT_STARTED', 'detail': '尚未执行',
'consent_allowed': not force_block_reason(row) if force else row.get('scope_confirmed') is True}
for row in plan['sessions']], 'force': force, 'cancel_native': cancel_native,
'evidence_path': str(root/'result.json'), 'previous_records': plan.get('previous_records', [])}
def save():
from .management import _write_json
try:
result['report_saved'] = True
_write_json(root/'result.json', {**result, 'sessions': [_session_evidence(row) for row in result['sessions']]})
return True
except OSError as error:
result.update(status='PARTIAL', report_saved=False, evidence_error=str(error))
return False
if not save():
return result # No native actions without the initial checkpoint.
if force or cancel_native:
consent_path = root/'consent.json'
consent = {'version': 1, 'mode': 'cancel_native' if cancel_native else 'force_document_clients', 'library': str(library),
'time_ns': time.time_ns(), 'sessions': [
{**_session_evidence(row), 'force_allowed': not force_block_reason(row),
'cancel_allowed': bool(row.get('cancel_prompt')) and row.get('scope_confirmed') is True,
'force_block_reason': force_block_reason(row)} for row in plan['sessions']]}
try:
with consent_path.open('x', encoding='utf-8') as stream:
json.dump(consent, stream, ensure_ascii=False, indent=2)
stream.flush(); os.fsync(stream.fileno())
except OSError as error:
result['detail'] = f'确认记录保存失败,未执行:{error}'
save()
raise
result['consent_path'] = str(consent_path)
try:
fresh = {r['pid']: r for r in query_process_snapshot()}
blocks = plan.get('block_names', ())
roots = (tuple(library.parent/'ReusableBlocksLP'/b/'Work' for b in blocks)
+ (library.parent/'SymbolLibs',)) if blocks else (library.parent,)
# Close native children first; keep output rows in the original preview order.
ordered = sorted(result['sessions'], key=lambda r: r['name'].casefold() in _WORKER_NAMES)
for row in ordered:
current = fresh.get(row['pid'])
outcome = {'release_status': 'RELEASED', 'detail': '已退出'}
if stop is not None and stop.is_set():
outcome = {'release_status': 'CANCELLED', 'detail': '已停止释放'}
elif row.get('scope_confirmed') is False:
from .library_lock import _owner_is_blocking
outcome = ({'release_status': 'WAITING', 'detail': '当前归属已变化,请重新检查确认;未关闭'}
if current is not None and _owner_is_blocking(current, library, roots) else
{'release_status': 'OBSERVED', 'detail': '仅观察,未关闭;由目标资源检查确认是否影响当前库'})
elif current is not None:
from .library_lock import _owner_matches_scope
scoped = _owner_matches_scope(current, library, roots) is True
if row.get('scope_confirmed') is not True or not scoped or row['pid'] == os.getpid():
outcome = {'release_status': 'WAITING', 'detail': '未在确认名单中证明属于当前库,未关闭;请重新检查'}
elif not row.get('created') or any(current.get(key) != row.get(key)
for key in ('created', 'executable', 'command', 'name')):
outcome = {'release_status': 'IDENTITY_CHANGED', 'detail': '确认后进程身份已变化,未关闭'}
elif force and (reason := force_block_reason(row)):
outcome = {'release_status': 'WAITING', 'detail': reason, 'force_allowed': False}
else:
row.update(release_status='REQUESTING', detail='正在请求原生操作,结果待确认')
if not save():
return result
try:
outcome = (_request_native_cancel(row, stop=stop) if cancel_native else
(_force_close if force else _request_close)(row, stop=stop))
except Exception as error:
outcome = {'release_status': 'WAITING', 'detail': str(error)}
row.update(outcome)
if not save():
return result # Preserve completed facts; stop further actions on storage failure.
after = {r['pid']: r for r in query_process_snapshot()}
remaining_windows = _process_windows({r['pid'] for r in result['sessions']
if r['release_status'] == 'WAITING' and r['pid'] in after})
for row in result['sessions']:
if row['release_status'] == 'WAITING' and row['pid'] not in after:
row.update(release_status='RELEASED', detail='已退出')
elif row['release_status'] == 'WAITING':
windows = remaining_windows.get(row['pid'], [])
error = _window_error(remaining_windows, row['pid'])
row.update(prompt=_save_question(windows, incomplete=error), windows=windows,
windows_incomplete=bool(error), window_read_error=error,
cancel_prompt=(_cancel_question(row['name'], windows, incomplete=error)
if row.get('force_user_confirmed') is True else None),
native_activity=native_activity(row['name'], windows, incomplete=error))
if release_targets_complete(result['sessions']) and not (stop and stop.is_set()):
result['proof'] = _finish_library_release(library, root/'library-proof')
proof = result['proof']
if (proof.get('status') != 'RELEASE_CONFIRMED'
or Path(proof.get('library', '')).resolve() != library):
raise ValueError('释放证明与当前库不一致')
result['status'] = 'RELEASE_CONFIRMED'
from .session_record import record_session_release
errors = record_session_release(proof, plan.get('runtime_sessions', []))
if errors:
result['evidence_error'] = ';'.join(errors)
except BaseException as error:
result.update(status='PARTIAL', detail=f'{type(error).__name__}: {error}')
if not isinstance(error, Exception):
raise
finally:
save()
return result
def release_residual_sessions(plan, evidence_dir, *, stop=None, force=False, cancel_native=False):
import win32con, win32event
mutex = win32event.CreateMutex(None, False, r'Local\CbbRefreshVerify795')
acquired = False
try:
deadline = time.monotonic()+15
while time.monotonic() < deadline and not (stop and stop.is_set()):
acquired = win32event.WaitForSingleObject(mutex, 0) in (win32con.WAIT_OBJECT_0, win32con.WAIT_ABANDONED)
if acquired: break
time.sleep(.2)
if not acquired:
return {'status': 'PARTIAL', 'library': plan['library'], 'sessions': [
{**r, 'release_status': 'WAITING', 'detail': '其它批次尚在执行;请停止该批次后重试'}
for r in plan['sessions']], 'detail': '未取得执行通道,未关闭任何进程。'}
return _release_residual_sessions(plan, evidence_dir, stop=stop, force=force, cancel_native=cancel_native)
finally:
if acquired: win32event.ReleaseMutex(mutex)
mutex.Close()
def answer_save_question(row, choice):
"""Only the exact displayed save question receives an explicit user answer."""
import win32con, win32event, win32gui
if choice not in (6, 7, 2): raise ValueError('无效的保存选择')
mutex = win32event.CreateMutex(None, False, r'Local\CbbRefreshVerify795')
acquired = win32event.WaitForSingleObject(mutex, 0) in (win32con.WAIT_OBJECT_0, win32con.WAIT_ABANDONED)
if not acquired:
mutex.Close()
raise ValueError('其它批次正在执行,请稍后处理保存提示')
try:
_answer_save_question(row, choice)
finally:
win32event.ReleaseMutex(mutex)
mutex.Close()
def _answer_save_question(row, choice):
import win32con
fresh = next((r for r in query_process_snapshot() if r['pid'] == row['pid']), None)
if fresh is None or any(fresh.get(k) != row.get(k) for k in ('created', 'command', 'executable')):
raise ValueError('进程已变化,请重新检查')
with _process_handle(row['pid']) as handle:
if _handle_identity(handle) != {k: row.get(k) for k in ('created_handle', 'image')}:
raise ValueError('进程已变化,请重新检查')
snapshot = _process_windows({row['pid']})
prompt = _save_question(snapshot.get(row['pid'], []), incomplete=_window_error(snapshot, row['pid']))
if not prompt or prompt != row.get('prompt'):
raise ValueError('原生提示已变化,请重新检查')
button = prompt['buttons'][choice]
_post_window_message(prompt, button['hwnd'], row['pid'], win32con.BM_CLICK)
+5
View File
@@ -58,6 +58,11 @@ class RunTimings:
elif stage == 'identity_verified':
self.end('pcb_startup', now)
self.begin('integration', now)
elif stage == 'diagnostic_native_loaded_candidate':
self.end('pcb_startup', now)
self.begin('diagnostic_close', now)
elif stage == 'diagnostic_complete':
self.end('diagnostic_close', now)
elif stage == 'integration_action':
key = ACTION_KEYS.get(event.get('action'))
if key:
+80 -9
View File
@@ -1,5 +1,8 @@
Option Explicit
Dim app, doc, pi, licensing, token, expected, attempt, stable, measuredStart, operationOK, remaining
Dim nativeError, nativeDescription
WaitForStartGate
CheckCancellation
expected = WScript.Arguments(0)
Set app = GetObject(, "MGCPCB.Application")
Set doc = app.ActiveDocument
@@ -17,40 +20,89 @@ token = licensing.GetToken(doc.Validate(0))
doc.Validate token
LogElapsed "validate", measuredStart
WScript.Echo "LICENSE_OK"
CheckCancellation
WScript.Echo "SAVE_START"
CheckCancellation
measuredStart = Timer
On Error Resume Next
operationOK = doc.Save()
nativeError = Err.Number
nativeDescription = Err.Description
Err.Clear
On Error GoTo 0
LogElapsed "save", measuredStart
If Not operationOK Then WScript.Quit 4
If nativeError <> 0 Then
WScript.Echo "SAVE_RESULT_UNKNOWN|" & CStr(nativeError) & "|" & nativeDescription
WScript.Quit 11
End If
If Not operationOK Then
WScript.Echo "SAVE_REJECTED"
WScript.Quit 4
End If
WScript.Echo "SAVE_OK"
CheckCancellation
measuredStart = Timer
On Error Resume Next
operationOK = doc.IsSaved()
nativeError = Err.Number
nativeDescription = Err.Description
Err.Clear
On Error GoTo 0
LogElapsed "is_saved", measuredStart
If Not operationOK Then WScript.Quit 5
If nativeError <> 0 Or Not operationOK Then
WScript.Echo "POST_SAVE_FAILED|" & CStr(nativeError) & "|" & nativeDescription
WScript.Quit 5
End If
WScript.Echo "IS_SAVED"
stable = False
WScript.Echo "POST_SAVE_CHECK"
measuredStart = Timer
For attempt = 1 To 50
On Error Resume Next
Set pi = doc.ProjectIntegration
If pi.ForwardAnnotationStatus = 2 And pi.SynchCESStatus = 2 And pi.LoadCESStatus = 2 And pi.BackAnnotationStatus = 2 Then
stable = (pi.ForwardAnnotationStatus = 2 And pi.SynchCESStatus = 2 And pi.LoadCESStatus = 2 And pi.BackAnnotationStatus = 2)
nativeError = Err.Number
nativeDescription = Err.Description
Err.Clear
On Error GoTo 0
If nativeError <> 0 Then
WScript.Echo "POST_SAVE_FAILED|" & CStr(nativeError) & "|" & nativeDescription
WScript.Quit 7
End If
If stable Then
stable = True
Exit For
End If
WScript.Sleep 200
Next
LogElapsed "post_save_check", measuredStart
If Not stable Then WScript.Quit 7
If Not stable Then
WScript.Echo "POST_SAVE_FAILED|integration_not_synchronized"
WScript.Quit 7
End If
WScript.Echo "POST_SAVE_SYNCHRONIZED"
If Not doc.IsSaved() Then WScript.Quit 8
On Error Resume Next
operationOK = doc.IsSaved() And LCase(pi.ProjectFile) = LCase(expected)
If WScript.Arguments.Count > 1 Then operationOK = operationOK And LCase(doc.Name) = LCase(WScript.Arguments(1))
nativeError = Err.Number
nativeDescription = Err.Description
Err.Clear
On Error GoTo 0
If nativeError <> 0 Or Not operationOK Then
WScript.Echo "POST_SAVE_FAILED|" & CStr(nativeError) & "|identity_or_saved_state_unconfirmed|" & nativeDescription
WScript.Quit 8
End If
WScript.Echo "CLOSE_START"
CheckCancellation
measuredStart = Timer
On Error Resume Next
operationOK = doc.Close(True)
nativeError = Err.Number
nativeDescription = Err.Description
Err.Clear
On Error GoTo 0
LogElapsed "close", measuredStart
If Not operationOK Then WScript.Quit 6
If nativeError <> 0 Or Not operationOK Then
WScript.Echo "CLOSE_FAILED|" & CStr(nativeError) & "|" & nativeDescription
WScript.Quit 6
End If
WScript.Echo "CLOSE_OK"
Set pi = Nothing
Set doc = Nothing
@@ -75,11 +127,30 @@ Sub CheckCancellation()
If Len(cancelPath) = 0 Then Exit Sub
Set fso = CreateObject("Scripting.FileSystemObject")
If fso.FileExists(cancelPath) Then
WScript.Echo "CANCELLED_BEFORE_SAVE"
WScript.Echo "VERIFY_CANCELLED"
WScript.Quit 9
End If
End Sub
Sub WaitForStartGate()
Dim fso, started, elapsed
If WScript.Arguments.Count < 4 Then Exit Sub
Set fso = CreateObject("Scripting.FileSystemObject")
started = Timer
Do
CheckCancellation
If fso.FileExists(WScript.Arguments(3)) Then Exit Sub
elapsed = Timer - started
If elapsed < 0 Then elapsed = elapsed + 86400
If elapsed >= 30 Then
WScript.Echo "CANCELLED_BEFORE_SAVE|start_gate_timeout"
WScript.Quit 9
End If
WScript.Sleep 50
Loop
End Sub
Sub LogElapsed(phaseName, started)
Dim elapsed
elapsed = Timer - started
+59 -11
View File
@@ -15,6 +15,15 @@ import stat
MANIFEST_VERSION = 1
MANIFEST_RULE_VERSION = 2
_ROOT_RUNTIME_PREFIX = 'work.before-restore-'
_ROOT_RUNTIME_PATH = ('projectbackup', 'backups')
_NATIVE_RUNTIME_DIRECTORIES = ('log', 'dbcheck', 'dbrestore')
_NATIVE_SERVER_DIRECTORY = 'cdbsvr'
_PROJECT_PCB_DIRECTORY = 'pcb'
_PROJECT_DATABASE_PREFIXES = (('database',), (_PROJECT_PCB_DIRECTORY, 'database'))
_NONPRIMARY_FILENAMES = ('saddress.adr', 'saddress_.adr')
_NONPRIMARY_FILENAME_PREFIXES = ('~$',)
_NONPRIMARY_DIRECTORIES = frozenset({
'log', 'logs', 'temp', 'tmp', '__pycache__', 'backup', 'backups',
'output', 'outputs', 'report', 'reports', 'gerber', 'gerbers', 'odb++',
@@ -29,22 +38,21 @@ def _root_runtime(directories):
"""Known runtime/history children of a library or project workspace root."""
return bool(directories and (
directories[0] in _NONPRIMARY_DIRECTORIES
or directories[0].startswith('work.before-restore-')
or directories[:2] == ('projectbackup', 'backups')))
or directories[0].startswith(_ROOT_RUNTIME_PREFIX)
or directories[:len(_ROOT_RUNTIME_PATH)] == _ROOT_RUNTIME_PATH))
def _native_runtime(directories, database_prefix):
runtime_prefix = database_prefix + ('cdbsvr',)
runtime_prefix = database_prefix + (_NATIVE_SERVER_DIRECTORY,)
return (directories[:len(runtime_prefix)] == runtime_prefix
and len(directories) > len(runtime_prefix)
and directories[len(runtime_prefix)] in ('log', 'dbcheck', 'dbrestore'))
and directories[len(runtime_prefix)] in _NATIVE_RUNTIME_DIRECTORIES)
def _project_runtime(directories):
return (_root_runtime(directories)
or (directories[:1] == ('pcb',) and _root_runtime(directories[1:]))
or _native_runtime(directories, ('database',))
or _native_runtime(directories, ('pcb', 'database')))
or (directories[:1] == (_PROJECT_PCB_DIRECTORY,) and _root_runtime(directories[1:]))
or any(_native_runtime(directories, prefix) for prefix in _PROJECT_DATABASE_PREFIXES))
@dataclass(frozen=True)
@@ -85,7 +93,8 @@ class ManifestScope:
"""
directories = parts if directory else parts[:-1]
if not directory and (Path(parts[-1]).suffix in _NONPRIMARY_SUFFIXES
or parts[-1] == 'saddress.adr' or parts[-1].startswith('~$')):
or parts[-1] in _NONPRIMARY_FILENAMES
or parts[-1].startswith(_NONPRIMARY_FILENAME_PREFIXES)):
return True
if self.kind == 'project':
return _project_runtime(directories)
@@ -123,6 +132,36 @@ class ManifestScope:
return 'shared_dependency'
def comparison_rules(scope):
"""Describe semantic exclusions using the classifier's rule constants."""
project_positions = ['project root or its PCB child']
project_native = ' or '.join('/'.join(prefix + (_NATIVE_SERVER_DIRECTORY,))
for prefix in _PROJECT_DATABASE_PREFIXES)
native_positions = [project_native + ' under project root']
if scope.kind == 'library':
project_positions = ['library root', 'selected CBB Work/Cert root or its PCB child']
native_positions = [
project_native + ' under selected CBB Work/Cert',
'SymbolLibs/<partition>/' + _NATIVE_SERVER_DIRECTORY,
'<shared dependency>/' + _NATIVE_SERVER_DIRECTORY,
]
return {
'rule_version': MANIFEST_RULE_VERSION,
'excluded_categories': sorted(_NONPRIMARY_DIRECTORIES),
'excluded_suffixes': sorted(_NONPRIMARY_SUFFIXES),
'runtime_exclusions': {
'directory_name_positions': project_positions,
'directory_prefixes': [_ROOT_RUNTIME_PREFIX],
'relative_directories': ['/'.join(_ROOT_RUNTIME_PATH)],
'native_directory_names': sorted(_NATIVE_RUNTIME_DIRECTORIES),
'native_directory_positions': native_positions,
'file_names': list(_NONPRIMARY_FILENAMES),
'file_name_prefixes': list(_NONPRIMARY_FILENAME_PREFIXES),
'coordination_checkpoints': 'recognized CBB coordination checkpoints',
},
}
def hash_file(path):
digest = hashlib.sha256()
with Path(path).open('rb') as stream:
@@ -271,7 +310,10 @@ def copy_verified(source, destination, *, scope=None, progress=None, hash_functi
Native icdb.dat/LMC files are held with exclusive read handles for the copy.
Their expected SHA256 comes from the source read stream, independently of
hashing destination bytes. Failures retain the partial destination evidence.
hashing destination bytes. Unlocked primary files are also hashed while
streaming the copy, then rehashed at source to detect even same-size/mtime
edits. Every primary destination is independently hashed. Failures retain
the partial destination evidence.
Progress receives one dict with stage, hashed_files/bytes and copied_files.
"""
import win32con
@@ -329,9 +371,15 @@ def copy_verified(source, destination, *, scope=None, progress=None, hash_functi
shutil.copystat(path, copied)
if stream_hash is not None:
digest = stream_hash.hexdigest()
elif label is not None:
stream_hash = hashlib.sha256()
with path.open('rb') as incoming, copied.open('wb') as outgoing:
for data in iter(lambda: incoming.read(1024 * 1024), b''):
stream_hash.update(data)
outgoing.write(data)
shutil.copystat(path, copied)
digest = stream_hash.hexdigest()
else:
if label is not None:
digest = hash_function(path)
shutil.copy2(path, copied)
before, after = entries[path], path.stat()
if (before.st_size, before.st_mtime_ns) != (after.st_size, after.st_mtime_ns):
+477
View File
@@ -0,0 +1,477 @@
"""Bounded, observation-only journal for native calls that outlive a task."""
from contextlib import ExitStack
from datetime import datetime, timezone
import json
import logging
import os
from pathlib import Path
import stat
import threading
from uuid import uuid4
PROTOCOL = 'cbb-runtime-session'
MAX_BYTES = 64 * 1024
IDENTITY_KEYS = ('pid', 'created_handle', 'image', 'owner_sid', 'session_id')
_observers = {}
_observer_guard = threading.Lock()
_retained_libraries = {}
_retained_results = {}
def _library_key(library):
return os.path.normcase(os.path.normpath(str(library)))
def library_available(library):
"""Cached queue predicate: no file, process or network queries."""
key = _library_key(library)
with _observer_guard:
return not any(key in aliases for aliases in _retained_libraries.values())
def retained_session_snapshot(library):
key = _library_key(library)
with _observer_guard:
return {path: (aliases, _retained_results.get(path, {}).get('block'),
_retained_results.get(path, {}).get('operation_id'))
for path, aliases in _retained_libraries.items() if key in aliases}
def reconciled_retained_sessions(snapshot, *, block_name=None):
with _observer_guard:
return [{'path': path, 'block_name': block, 'operation_id': operation_id}
for path, (aliases, block, operation_id) in snapshot.items()
if _retained_libraries.get(path) is aliases
and _retained_results.get(path, {}).get('operation_id') == operation_id
and (block_name is None or block == block_name)]
def acknowledge_library_release(proof, snapshot, *, block_name=None):
"""Called under fresh proof handles; never clear a newer handoff."""
if proof.get('status') != 'RELEASE_CONFIRMED':
return
key = _library_key(proof['library'])
with _observer_guard:
records = [dict(path=path, operation_id=operation_id)
for path, (aliases, block, operation_id) in snapshot.items()
if key in aliases and _retained_libraries.get(path) is aliases
and _retained_results.get(path, {}).get('operation_id') == operation_id
and (block_name is None or block == block_name)]
record_session_release(proof, records)
with _observer_guard:
for path, (aliases, block, operation_id) in snapshot.items():
if (key in aliases and _retained_libraries.get(path) is aliases
and _retained_results.get(path, {}).get('operation_id') == operation_id
and (block_name is None or block == block_name)):
_retained_libraries.pop(path, None)
_retained_results.pop(path, None)
def record_session_release(proof, records):
"""Persist a past proof only for the pre-proof operation, never replay it."""
errors = []
if proof.get('status') != 'RELEASE_CONFIRMED':
return errors
library = Path(proof['library']).resolve()
for record in records:
try:
path = Path(record['path'])
session = read_session_record(path, evidence_root=path.parent)
if (session['operation_id'] != record.get('operation_id')
or Path(session['library']).resolve() != library):
continue
write_session_record(path.with_name('observation-release-'+uuid4().hex+'.json'), {
'operation_id': session['operation_id'], 'library': str(library),
'status': 'RELEASED', 'release_proof': proof})
except (OSError, ValueError, KeyError) as error:
errors.append(str(error))
logging.getLogger(__name__).warning('Unable to record session release', exc_info=True)
return errors
def retained_session_states(library):
key = _library_key(library)
with _observer_guard:
return [dict(_retained_results[path], path=path) for path, aliases in _retained_libraries.items()
if key in aliases and path in _retained_results]
def library_wait_reason(library):
"""Cached explanation for queue admission; never probe disk or native clients."""
key = _library_key(library)
with _observer_guard:
retained = [path for path, aliases in _retained_libraries.items() if key in aliases]
names = sorted({_retained_results.get(path, {}).get('block') or '当前库' for path in retained})
return '该库有残留会话('+', '.join(names)+'),请查看并处理。' if names else ''
def register_worker_session(worker, job):
"""The launcher must persist identity before opening the worker's gate."""
path = Path(job['session_record_path'])
write_session_record(path, {k: job.get(k) for k in (
'operation_id', 'operation', 'library', 'block', 'desktop', 'cancel_signal')}
| {'processes': [{'role': 'worker', **process_identity(worker.pid)}]})
worker.session_record_path = path
worker.cbb_library = job['library']
def update_worker_state(job, phase, *, identities=(), write_outcome=None, required=False, continuation=False):
"""required controls I/O; only existing same-phase calls may continue recording after supervisor loss."""
if not job.get('session_record_path'):
if required:
raise ValueError('原生写入缺少会话记录位置,未提交新的调用')
return False
from .desktop_worker import worker_phase
with worker_phase(job['operation_id'], phase, continuation=continuation):
path = Path(job['session_record_path']).with_name('worker-state.json')
try:
state = ({'operation_id': job['operation_id'], 'processes': []} if not path.exists()
else read_session_record(path, evidence_root=path.parent))
if state['operation_id'] != job['operation_id']:
raise ValueError('工作状态操作身份不一致')
for role, pid in identities:
if pid and not any(p['pid'] == pid for p in state['processes']):
state['processes'].append({'role': role, **process_identity(pid)})
state['phase'] = phase
if write_outcome is not None:
state['write_outcome'] = write_outcome
write_session_record(path, state)
return True
except Exception:
if required:
raise
# An unavailable journal cannot interrupt an already executing COM call.
logging.getLogger(__name__).warning('Worker session journal unavailable', exc_info=True)
return False
def dispatch_native(job, call, *, phase='NATIVE_CALL', asynchronous=False, required=False):
"""Reserve the native call before dispatch; asynchronous results require an explicit acknowledgment."""
update_worker_state(job, phase, required=required)
try:
result = call()
except BaseException:
update_worker_state(job, 'UNCONFIRMED')
raise
if not asynchronous:
update_worker_state(job, 'POSTCALL')
return result
def process_identity(pid):
from .residual_sessions import _process_handle, _handle_identity, _force_ownership
with _process_handle(pid) as handle:
identity = _handle_identity(handle)
identity.update(_force_ownership(handle, pid))
return {'pid': int(pid), **{k: identity[k] for k in IDENTITY_KEYS if k != 'pid'}}
def write_session_record(path, record):
path = Path(path)
value = {'protocol': PROTOCOL, 'version': 1, **record,
'recorded_at': datetime.now(timezone.utc).isoformat()}
data = json.dumps(value, ensure_ascii=False, default=str, indent=2).encode('utf-8')
if len(data) > MAX_BYTES:
raise ValueError('会话记录超过大小限制')
pending = path.with_name(path.name + '.' + uuid4().hex + '.pending')
try:
with pending.open('xb') as stream:
stream.write(data)
stream.flush()
os.fsync(stream.fileno())
os.replace(pending, path)
finally:
pending.unlink(missing_ok=True)
def read_session_record(path, *, evidence_root):
path, root = Path(path).absolute(), Path(evidence_root).absolute()
if not path.is_relative_to(root) or path == root:
raise ValueError('会话记录位于受控证据目录之外')
# Check every component before resolve, which would hide a junction.
for item in (path, *path.parents):
attrs = item.lstat()
if getattr(attrs, 'st_file_attributes', 0) & stat.FILE_ATTRIBUTE_REPARSE_POINT:
raise ValueError('会话记录路径包含重解析点')
with path.open('rb') as stream:
data = stream.read(MAX_BYTES + 1)
if len(data) > MAX_BYTES:
raise ValueError('会话记录超过大小限制')
value = json.loads(data)
if (not isinstance(value, dict) or value.get('protocol') != PROTOCOL or
type(value.get('version')) is not int or value['version'] != 1 or
not isinstance(value.get('operation_id'), str) or not value['operation_id']):
raise ValueError('会话记录协议无效')
return value
def _session(path, root):
value = read_session_record(path, evidence_root=root)
if (not isinstance(value.get('library'), str) or not Path(value['library']).is_absolute()
or Path(value['library']).suffix.casefold() != '.lmc'):
raise ValueError('会话库身份无效')
state_path = Path(path).with_name('worker-state.json')
state = {}
if state_path.exists():
state = read_session_record(state_path, evidence_root=root)
if state['operation_id'] != value['operation_id']:
raise ValueError('工作进程状态不属于本次操作')
processes = value.get('processes', []) + state.get('processes', [])
if not processes:
raise ValueError('会话缺少精确进程身份')
for process in processes:
if (not isinstance(process, dict) or any(k not in process for k in IDENTITY_KEYS)
or type(process['pid']) is not int or process['pid'] <= 0):
raise ValueError('会话进程身份不完整')
return value, state, processes
def _result(status, detail='', *, session=None, state=None, proof=None):
return {'status': status, 'detail': detail, 'library': (session or {}).get('library'),
'block': (session or {}).get('block'),
'operation_id': (session or {}).get('operation_id'),
'phase': (state or {}).get('phase', 'UNCONFIRMED'), 'release_proof': proof}
def _reconcile(path, root, handles, stack, *, release=True):
import pywintypes, win32api, win32con, win32event
from .residual_sessions import _handle_identity, _force_ownership
session, state, processes = _session(path, root)
active = False
for identity in processes:
if identity.get('role') == 'supervisor':
continue
key = tuple(identity[k] for k in IDENTITY_KEYS)
changed_key = ('identity_changed', *key)
if changed_key in handles:
return _result('IDENTITY_CHANGED', '记录中的进程身份已变化,仅供观察', session=session, state=state)
if key not in handles:
try:
handle = win32api.OpenProcess(win32con.SYNCHRONIZE | 0x1000, False, identity['pid'])
except pywintypes.error as error:
if error.winerror != 87: # Access denied is not an exit observation.
raise
handles[key] = None
else:
stack.callback(handle.Close)
if win32event.WaitForSingleObject(handle, 0) == win32con.WAIT_OBJECT_0:
handles[key] = handle
continue
actual = {'pid': identity['pid'], **_handle_identity(handle),
**_force_ownership(handle, identity['pid'])}
if any(actual[k] != identity[k] for k in IDENTITY_KEYS):
handles[key] = handle
handles[changed_key] = True
return _result('IDENTITY_CHANGED', '记录中的进程身份已变化,仅供观察',
session=session, state=state)
handles[key] = handle
handle = handles[key]
if handle is not None and win32event.WaitForSingleObject(handle, 0) != win32con.WAIT_OBJECT_0:
active = True
if active:
return _result('ACTIVE', '原生调用尚未确认返回', session=session, state=state)
if not release:
return _result('UNCONFIRMED', '记录中的进程已退出,需确认库占用已释放', session=session, state=state)
from .library_lock import release_retained_library
try:
proof = release_retained_library(session['library'], Path(path).parent/'release-proof',
block_name=session.get('block'))
except Exception as error:
return _result('UNCONFIRMED', str(error), session=session, state=state)
if proof.get('status') != 'RELEASE_CONFIRMED':
return _result('UNCONFIRMED', '缺少独立资源释放证明', session=session, state=state)
from .network_path import release_retained_path_leases
release_retained_path_leases(proof)
return _result('RELEASED', session=session, state=state, proof=proof)
def reconcile_session_record(path, *, evidence_root):
with ExitStack() as stack:
try:
return _reconcile(Path(path), Path(evidence_root), {}, stack)
except Exception as error:
return _result('UNCONFIRMED', str(error))
def inspect_session_record(path, *, evidence_root):
"""Observe process identities without touching the production library."""
with ExitStack() as stack:
try:
return _reconcile(Path(path), Path(evidence_root), {}, stack, release=False)
except Exception as error:
return _result('UNCONFIRMED', str(error))
def observe_retained_session(path, *, evidence_root, stop=None, progress=None, process=None,
observe_only=False):
stop = stop if stop is not None else threading.Event()
observation = Path(path).with_name('observation-' + uuid4().hex + '.json')
with ExitStack() as stack:
handles = {}
while True:
try:
result = _reconcile(Path(path), Path(evidence_root), handles, stack, release=not observe_only)
except Exception as error:
result = _result('UNCONFIRMED', str(error))
if (not observe_only and process is not None and process.poll() is not None
and getattr(process, 'cbb_library', None)):
try:
from .library_lock import release_retained_library
proof = release_retained_library(process.cbb_library, Path(path).parent/'release-proof')
if proof.get('status') == 'RELEASE_CONFIRMED':
result = _result('RELEASED', '会话记录不可读;由原进程句柄及独立资源证明确认释放',
session={'library': str(process.cbb_library)}, proof=proof)
except Exception:
pass
if result['status'] == 'IDENTITY_CHANGED' and not observe_only:
try:
from .library_lock import release_retained_library
proof = release_retained_library(result['library'], Path(path).parent/'release-proof')
if proof.get('status') == 'RELEASE_CONFIRMED':
result = {**result, 'status': 'RELEASED', 'identity_changed': True, 'release_proof': proof}
except Exception as error:
result['detail'] += f';资源释放待确认:{error}'
try:
write_session_record(observation, {**result, 'operation_id': result['operation_id'] or 'unknown'})
except (OSError, ValueError):
logging.getLogger(__name__).warning('Observation journal unavailable', exc_info=True)
if progress is not None:
try:
progress(result)
except Exception:
logging.getLogger(__name__).warning('Observation callback failed', exc_info=True)
if (result['status'] == 'RELEASED' or stop.is_set()
or observe_only and result['status'] != 'ACTIVE'):
return result
if not handles and process is not None and process.poll() is None:
stop.wait(1)
continue # The exact owned handle survives a failed journal.
if not handles and process is None:
return result
stop.wait(5 if result['status'] in ('UNCONFIRMED', 'IDENTITY_CHANGED') else 1)
def handoff_session(path, *, evidence_root, process=None, progress=None, stop=None, observe_only=False):
"""Retain exact handles in memory until the observer independently releases."""
key = str(Path(path).absolute()).casefold()
try:
session = read_session_record(path, evidence_root=evidence_root)
library = session['library']
except (OSError, ValueError, KeyError):
session = {}
library = getattr(process, 'cbb_library', None)
aliases = {_library_key(library), _library_key(Path(library).resolve())} if library else set()
with _observer_guard:
if key in _observers:
return
_retained_libraries[key] = aliases
_retained_results[key] = {'status': 'UNCONFIRMED', 'phase': 'UNCONFIRMED',
'library': str(library) if library else None, 'block': session.get('block'),
'detail': '正在核对残留会话', 'operation_id': session.get('operation_id')}
def notify(value):
with _observer_guard:
if _retained_libraries.get(key) is aliases:
previous = _retained_results[key]
_retained_results[key] = {**previous, **value,
'library': value.get('library') or previous['library'],
'operation_id': value.get('operation_id') or previous['operation_id'],
'block': value.get('block') or previous['block']}
if progress is not None:
progress(value)
def observe():
result = None
try:
result = observe_retained_session(path, evidence_root=evidence_root, progress=notify,
stop=stop, process=process, observe_only=observe_only)
if process is not None and process.poll() is not None:
process.close()
return result
finally:
with _observer_guard:
_observers.pop(key, None)
if result and result['status'] == 'RELEASED':
_retained_libraries.pop(key, None)
_retained_results.pop(key, None)
thread = threading.Thread(target=observe, name='retained-native-session', daemon=True)
_observers[key] = (thread, process)
thread.start()
def discover_sessions(evidence_root, *, library=None):
"""Only our bounded journals are inputs; never execute a stored command."""
root = Path(evidence_root)
found = []
if not root.is_dir():
return found
visited = 0
for folder, directories, files in os.walk(root, followlinks=False):
directories[:] = [d for d in directories if not (
getattr((Path(folder)/d).lstat(), 'st_file_attributes', 0) & stat.FILE_ATTRIBUTE_REPARSE_POINT)]
visited += 1
if visited > 10000:
break
if 'session.json' not in files:
continue
path = Path(folder)/'session.json'
try:
value = read_session_record(path, evidence_root=root)
completed = False
for name in files:
if not (name.startswith('observation-') and name.endswith('.json')):
continue
observation = read_session_record(Path(folder)/name, evidence_root=root)
if observation['operation_id'] == value['operation_id'] and (
observation.get('status') == 'RELEASED' and observation.get('release_proof')
or observation.get('status') == 'FINISHED'
and observation.get('native_release_confirmed') is True
and observation.get('worker_exited') is True):
completed = True
break
if completed:
continue
if library is not None and _library_key(value.get('library', '')) != _library_key(library):
continue
observed = inspect_session_record(path, evidence_root=root)
value = {**value, **observed, 'library': value.get('library'),
'block': value.get('block'), 'operation_id': value['operation_id'], 'path': str(path)}
except (OSError, ValueError):
continue
found.append(value)
return found
def active_write_records(library):
"""Called by the background residual inspector, never by a queue predicate."""
with _observer_guard:
paths = [Path(key) for key, aliases in _retained_libraries.items()
if _library_key(library) in aliases]
result = []
for path in paths:
try:
session, state, processes = _session(path, path.parent)
if processes:
from .residual_sessions import _process_handle, _handle_identity, _force_ownership
import win32con, win32event
for identity in processes:
if identity.get('role') == 'supervisor' or (
state.get('phase') not in ('NATIVE_CALL', 'CLOSING', 'UNCONFIRMED')
and identity.get('role') not in ('worker', 'helper')):
continue
try:
with _process_handle(identity['pid']) as handle:
if win32event.WaitForSingleObject(handle, 0) == win32con.WAIT_OBJECT_0:
continue
actual = {**_handle_identity(handle), **_force_ownership(handle, identity['pid'])}
if any(actual[k] != identity[k] for k in IDENTITY_KEYS if k != 'pid'):
continue
# A retained caller may be inside Close after the last journal write failed.
result.append({**session, 'phase': state['phase'] if state.get('phase') in (
'NATIVE_CALL', 'CLOSING', 'UNCONFIRMED') else 'UNCONFIRMED'})
break
except Exception as error:
if getattr(error, 'winerror', None) != 87:
result.append({**session, 'phase': 'UNCONFIRMED'})
break
except (OSError, ValueError):
result.append({'phase': 'UNCONFIRMED'})
return result
Loaded 100 of 193 files, more files were not shown because too many files have changed in this diff. Show more