diff --git a/docs/testing/2026-10-31-dsh-pr11-self-review.md b/docs/testing/2026-10-31-dsh-pr11-self-review.md new file mode 100644 index 0000000..20e3bef --- /dev/null +++ b/docs/testing/2026-10-31-dsh-pr11-self-review.md @@ -0,0 +1,27 @@ +# PR #11 自查修复 + +## 范围 + +以 `9552f1e` 为起点,仅审查 PR #11 的增量,并在 `codex/dsh` 同步其当前目标基线 `6f9f820` 解决冲突。窗口同时保留 DSH 监管和资源状态的初始化、刷新、原生确认门禁与关闭处理。 + +## 发现与修复 + +1. 恢复运行后,旧暂停回执的计数会使正常新增进度被判为 evidence_changed,阻断再次续跑。现在仅在运行状态、有效原检查点、旧非终态任务、当前无失败/中断且计数单调增长时识别为 expected_progress;旧终态证据变化、暂停期间变化、计数回退仍阻断。 +2. 检查点已标 completed、当前实际证据却未完成时,原实现可能直接跳过该任务。现在必须有匹配的完整、零失败、无中断证据,否则返回 checkpoint_evidence_mismatch。反向的业务结果已落盘、检查点尚为 running 的退出窗口允许重新核对并续跑。 +3. 执行后核验适配器只检查中断标记,失败项或不完整队列也可能返回 ready。现在有失败、失败计数未知或队列未完成都返回 uncertain,保留真实数量及独立阻断原因。底层已知执行回执不因补充读回结果而被覆盖。 + +## 专项验证 + +真实 TaskCenterService、原队列和 SQLite 的 25 项恢复/核验/运行记录测试通过,独立复审同样复跑通过。新增回归先复现失败后验证修复。 + +端到端队列回归使用隔离工程 fixture 和受控 cscript process_runner:A 完成后暂停,恢复完成 B 后在最终 journal.finish 前注入 BaseException 模拟退出,重新创建 RunJournal 并恢复 C。断言原 run/queue 不变,A/B/C 各执行一次,实际未命中计数保留,没有重复工程写入。 + +CLI 官方运行时门禁、模型结果裁剪和 dock 启动检查通过。未操作真实用户工程、未部署 broker、未构建 EXE。 + +DSH/UI/v1_1/打包回归:1255 passed,1 warning(已有 pywinauto STA COM 提示);耗时 382.79 秒。包含 integration 的更大范围探索运行已停止,未计为通过。最终恢复/核验 4 模块另由独立 reviewer 复跑 25 passed,覆盖最后补充的检查点证据一致性门禁。 + +## 桌面点击验收 + +两进程真实 Qt/CDP 验收通过,报告:build/dsh-pr11-self-review/20261031-144814/report.json。自然语言调用、原生确认、安全暂停、退出后原队列续跑、授权排期、审计损坏隔离及已有工单只读 GET 同步均通过。受控执行器总调用 3 次且步骤唯一,fixture 哈希及损坏审计原字节不变,两个应用进程均正常退出。realCadStarted=false,remoteIssuesCreated=false。未验证真实用户 CAD 写入。 + +PR 增量及本轮文件使用本机两份已知凭据逐字节扫描:85 文件、0 匹配,没有输出凭据。 diff --git a/docs/user/dsh-ai-supervision.md b/docs/user/dsh-ai-supervision.md index 99affd6..6db90e0 100644 --- a/docs/user/dsh-ai-supervision.md +++ b/docs/user/dsh-ai-supervision.md @@ -67,6 +67,8 @@ Harness 使用任务中心数据目录下独立的 `dsh/harness-home` 保存会 最近一次执行回执独立保留,随后查看任务列表不会覆盖它。执行历史根据本地持久记录和任务证据判断可否恢复;重启只展示记录,不自动重放。点击 **核对并恢复原队列** 后仍须原生确认。计划变化、缺少或损坏检查点、未核验中断、失败项等会阻断恢复;待核验结果从任务中心原流程处理。 +恢复执行后若在新回执写入前退出,原队列中新增的已核验进度可继续接续,不会因旧暂停回执的计数较小而重复执行或永久阻断。原先已收尾任务的证据变化、计数回退,以及完成检查点缺少匹配完成证据,仍会阻断。执行后核验只在队列完成且失败为零时报告就绪;未完成或有失败时明确显示待核验,实际保存与未命中数量分别保留。 + 原生任务中心的当前任务执行,以及不超过 20 个任务的原生批次,也写入这套运行记录。超过 20 个任务的原生分组继续使用原分组检查点,不加入 DSH 合并运行历史;不要把该分组当作可以从本页恢复的合并运行。 ### 影响范围 diff --git a/src/dx_productivity/dsh/cad_verification.py b/src/dx_productivity/dsh/cad_verification.py index 318f85a..68f30ca 100644 --- a/src/dx_productivity/dsh/cad_verification.py +++ b/src/dx_productivity/dsh/cad_verification.py @@ -51,6 +51,10 @@ class CadVerificationAdapter: 'interrupted': bool(getattr(context, 'interrupted', True))} if row['interrupted']: blockers.append('interrupted') + if type(row['failed_items']) is not int or row['failed_items'] != 0: + blockers.append('failed_items') + if row['queue_complete'] is not True: + blockers.append('queue_incomplete') except Exception as error: row = {'task_id': task_id, 'saved_items': None, 'failed_items': None, 'no_match_items': None, 'queue_complete': None, 'interrupted': True, diff --git a/src/dx_productivity/dsh/resume_diff.py b/src/dx_productivity/dsh/resume_diff.py index 41bd85c..2121f47 100644 --- a/src/dx_productivity/dsh/resume_diff.py +++ b/src/dx_productivity/dsh/resume_diff.py @@ -24,16 +24,35 @@ def resume_diff(service, inspected_run, *, current_plan_hashes=None): remaining = list(inspected_run.get('remaining_task_ids', ids)) completed = [key for key in ids if key not in remaining] baseline = {row['task_id']: row for row in (inspected_run.get('result') or {}).get('tasks', [])} + previous_entries = {row['task_id']: row.get('state') + for row in (inspected_run.get('result') or {}).get('entries', [])} evidence_changes = [] for task_id, context in contexts.items(): saved = baseline.get(task_id, {}) actual = {'interrupted': context.interrupted, **{field: getattr(context.progress, field, None) for field in ('saved_items', 'failed_items', 'no_match_items', 'queue_complete')}} + # A resumed run can crash after verified progress has landed, before + # its final receipt. The old pause receipt remains a useful baseline, + # but is immutable only for tasks that were already terminal then. + progressing = ( + inspected_run.get('status') == 'running' + and inspected_run.get('queue_available') is True + and previous_entries.get(task_id) in {'pending', 'paused', 'running', 'blocked'} + and inspected_run.get('checkpoint_states', {}).get(task_id) in {'running', 'paused', 'completed'} + and saved.get('interrupted') is False and actual['interrupted'] is False + and saved.get('queue_complete') is False and type(actual['queue_complete']) is bool + and type(saved.get('failed_items')) is int and saved['failed_items'] == 0 + and type(actual['failed_items']) is int and actual['failed_items'] == 0 + and all(type(saved.get(key)) is int and type(actual[key]) is int + and 0 <= saved[key] <= actual[key] for key in ('saved_items', 'no_match_items'))) for field, value in actual.items(): if saved.get(field) is not None and value != saved[field]: - evidence_changes.append({'task_id': task_id, 'field': field, 'saved': saved[field], 'current': value}) - if evidence_changes: + change = {'task_id': task_id, 'field': field, 'saved': saved[field], 'current': value} + if progressing: + change['expected_progress'] = True + evidence_changes.append(change) + if any(not row.get('expected_progress') for row in evidence_changes): blockers.append('evidence_changed') blockers = list(dict.fromkeys(blockers)) return {'safe': bool(inspected_run.get('resume_ready')) and not blockers, diff --git a/src/dx_productivity/dsh/run_journal.py b/src/dx_productivity/dsh/run_journal.py index 721ef70..ba2e05e 100644 --- a/src/dx_productivity/dsh/run_journal.py +++ b/src/dx_productivity/dsh/run_journal.py @@ -193,6 +193,11 @@ class RunJournal: blockers.append("interrupted") if getattr(progress, "failed_items", 0): blockers.append("failed_items") + if queue and queue['entries'][task_id]['state'] == 'completed' and not ( + context.has_evidence and progress.queue_complete is True + and type(progress.failed_items) is int and progress.failed_items == 0 + and context.interrupted is False): + blockers.append('checkpoint_evidence_mismatch') from ..trial_task_center.task_queue import _plan_hash current_hashes[task_id] = _plan_hash(service.for_project_task(task_id), task_id) if current_hashes[task_id] != row["plan_hashes"][task_id]: @@ -210,7 +215,7 @@ class RunJournal: if remaining: try: from .task_lifecycle import preflight - if not preflight(service, remaining)["ready"]: + if not preflight(service, remaining, allow_completed=True)["ready"]: blockers.append("preflight_blocked") except Exception: blockers.append("preflight_unavailable") @@ -218,6 +223,7 @@ class RunJournal: blockers.append("resources_pending") blockers = list(dict.fromkeys(blockers)) value = {**row, "queue_available": queue is not None, "tasks": tasks, "counters": counters, + "checkpoint_states": {key: entry['state'] for key, entry in queue['entries'].items()} if queue else {}, "remaining_task_ids": remaining, "blockers": blockers, "resume_ready": not blockers, "requires_confirmation": True, "cad_execution_started": False, "original_scope": {"task_ids": row["task_ids"], "plan_hashes": row["plan_hashes"]}} diff --git a/tests/dsh/test_cad_verification.py b/tests/dsh/test_cad_verification.py index af7e17c..30d344f 100644 --- a/tests/dsh/test_cad_verification.py +++ b/tests/dsh/test_cad_verification.py @@ -4,6 +4,7 @@ from types import SimpleNamespace as NS import pytest from dx_productivity.dsh.cad_verification import CadVerificationAdapter +from test_run_journal import project_service, evidence def service(tmp_path, *, probe=True, interrupted=False): @@ -43,3 +44,24 @@ def test_adapter_rejects_missing_task_without_inventing_counts(tmp_path): result = CadVerificationAdapter(svc).preflight(['missing']) assert result['status'] == 'blocked' assert result['projects'][0]['component_count'] is None + + +@pytest.mark.parametrize('state,blocker', [('batch_failed', 'failed_items'), (None, 'queue_incomplete')]) +def test_after_run_requires_successful_complete_evidence(project_service, state, blocker): + service, _, _, ids, _ = project_service + if state: + evidence(service, ids[0], state) + context = service.task_context(ids[0]) + assert context.progress.failed_items == (1 if state else 0) + assert not context.interrupted + result = CadVerificationAdapter(service).after_run(ids[:1]) + assert result['status'] == 'uncertain' + assert blocker in result['blockers'] + + +def test_after_run_keeps_verified_no_match_distinct_from_saved(project_service): + service, _, _, ids, _ = project_service + evidence(service, ids[0]) + result = CadVerificationAdapter(service).after_run(ids[:1]) + assert result['status'] == 'ready' and result['blockers'] == [] + assert result['tasks'][0]['saved_items'] == 0 and result['tasks'][0]['no_match_items'] == 1 diff --git a/tests/dsh/test_execution_recovery.py b/tests/dsh/test_execution_recovery.py new file mode 100644 index 0000000..c100850 --- /dev/null +++ b/tests/dsh/test_execution_recovery.py @@ -0,0 +1,53 @@ +"""Original queue recovery across a fatal receipt-write boundary.""" +from pathlib import Path +import subprocess +from threading import Event + +import pytest + +from dx_productivity.dsh.execution_runs import execute_confirmed_run +from dx_productivity.dsh.run_journal import RunJournal +from dx_productivity.dsh.task_lifecycle import preflight +from test_run_journal import project_service + + +def test_second_resume_after_fatal_exit_preserves_verified_progress_and_runs_each_task_once(project_service, monkeypatch): + service, _, data, ids, _ = project_service + calls = [] + active = {'stop': Event(), 'pause': True} + def worker(command, timeout): + queue, job = Path(command[3]), command[-1] + calls.append(next(key for key in ids if key in queue.parts)) + with (queue / 'status.tsv').open('a', encoding='utf8') as stream: + stream.write(f'{job}\tno_match\tbatch_closed\tControlled verified no match\n') + if active['pause']: + active['stop'].set() + return subprocess.CompletedProcess(command, 0, '', '') + for task_id in ids: + preparer = service.for_project_task(task_id)._quick_run_preparer + preparer._process_runner = worker + preparer._cscript_resolver = lambda: Path('cscript.exe') + run_id = 'd' * 32 + first = execute_confirmed_run(service, data, ids, preflight(service, ids), active['stop'], run_id) + assert first['status'] == 'paused' and calls == ids[:1] + original = RunJournal(data).inspect(service, run_id) + active['stop'] = Event() + class FatalExit(BaseException): + pass + with monkeypatch.context() as patch: + def fatal_finish(*args, **kwargs): + raise FatalExit('exit after checkpoint, before final receipt') + patch.setattr(RunJournal, 'finish', fatal_finish) + with pytest.raises(FatalExit): + execute_confirmed_run(service, data, ids, preflight(service, ids, allow_completed=True), + active['stop'], run_id, resume=original) + assert calls == ids[:2] + restarted = RunJournal(data).inspect(service, run_id) + assert restarted['status'] == 'running' and restarted['resume_ready'], restarted['blockers'] + assert restarted['remaining_task_ids'] == ids[2:] + active.update(stop=Event(), pause=False) + last = execute_confirmed_run(service, data, ids, preflight(service, ids, allow_completed=True), + active['stop'], run_id, resume=restarted) + assert last['status'] == 'completed' and last['queue_id'] == run_id + assert calls == ids + assert all(task['saved_items'] == 0 and task['no_match_items'] == 1 for task in last['tasks']) diff --git a/tests/dsh/test_resume_diff.py b/tests/dsh/test_resume_diff.py index 0aad79d..57a7a9c 100644 --- a/tests/dsh/test_resume_diff.py +++ b/tests/dsh/test_resume_diff.py @@ -1,6 +1,9 @@ from dx_productivity.dsh.resume_diff import resume_diff from dx_productivity.dsh.run_journal import RunJournal from test_run_journal import project_service, checkpoint, evidence +from dx_productivity.trial_task_center.task_queue import _plan_hash +from dx_productivity.trial_task_center.models import TaskStatus +import pytest def test_real_service_plan_changes_are_reported(project_service): @@ -46,3 +49,51 @@ def test_missing_current_hash_cannot_be_silently_ignored(project_service): 'queue_available': True, 'resume_ready': True, 'tasks': []}) assert value['safe'] is False assert value['changed'] == ids + + +def test_completed_checkpoint_requires_matching_current_evidence(project_service): + service, _, data, ids, hashes = project_service + journal = RunJournal(data) + journal.begin('c' * 32, ids, hashes) + checkpoint(data, 'c' * 32, ids, hashes, {ids[0]: 'completed'}) + report = journal.inspect(service, 'c' * 32) + assert not report['resume_ready'] + assert 'checkpoint_evidence_mismatch' in report['blockers'] + evidence(service, ids[0]) + assert journal.inspect(service, 'c' * 32)['resume_ready'] + + +@pytest.mark.parametrize('checkpoint_state', ['running', 'completed']) +def test_crash_during_resumed_run_accepts_verified_progress_but_not_old_terminal_changes(project_service, checkpoint_state): + service, store, data, ids, _ = project_service + hashes = {key: _plan_hash(service.for_project_task(key), key) for key in ids} + journal = RunJournal(data) + journal.begin('b' * 32, ids, hashes) + evidence(service, ids[0]) + for state in (TaskStatus.READY, TaskStatus.RUNNING, TaskStatus.COMPLETED): + store.set_task_status(ids[0], state) + checkpoint(data, 'b' * 32, ids, hashes, {ids[0]: 'completed', ids[1]: 'paused'}) + baseline = [] + for task_id in ids: + context = service.task_context(task_id) + baseline.append({'task_id': task_id, 'interrupted': context.interrupted, + **{key: getattr(context.progress, key) for key in + ('saved_items', 'failed_items', 'no_match_items', 'queue_complete')}}) + journal.finish('b' * 32, {'status': 'paused', 'tasks': baseline, + 'entries': [{'task_id': key, 'state': 'completed' if key == ids[0] else 'paused' if key == ids[1] else 'pending'} for key in ids]}) + assert journal.inspect(service, 'b' * 32)['resume_ready'] + journal.resume('b' * 32) + evidence(service, ids[1]) + for state in (TaskStatus.READY, TaskStatus.RUNNING, TaskStatus.COMPLETED): + store.set_task_status(ids[1], state) + checkpoint(data, 'b' * 32, ids, hashes, {ids[0]: 'completed', ids[1]: checkpoint_state}) + report = journal.inspect(service, 'b' * 32) + assert report['resume_ready'], report['blockers'] + assert report['diff']['evidence_changes'] + assert all(row['expected_progress'] for row in report['diff']['evidence_changes']) + # The run cannot rewrite or lose evidence for a task already terminal in + # the last durable receipt, even while later tasks legitimately advance. + root = service.for_project_task(ids[0])._quick_run_preparer.data_root / 'quick-run' + (root / 'status.tsv').write_text('job_id\tstate\tcode\tmessage\n', encoding='utf8') + changed = journal.inspect(service, 'b' * 32) + assert not changed['resume_ready'] and 'evidence_changed' in changed['blockers']