Merge branch 'codex/review-residual-release' into codex/runtime-lifecycle-safety

This commit is contained in:
阿白叔 committed 2026-11-09 14:10:38 +08:00
commit d0a5ce5d17
20 files changed
+454 -72

No files matched your search

@@ -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.
@@ -9,3 +9,12 @@ Dependency rationale: library release imports native engineering identity and re
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.
@@ -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.
@@ -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。