Repository navigation
Bound stalled startup and accepted shutdown without timing out edits - #12
Conversation
Generated-with: gpt-6.1-sol (Codex)
Generated-with: Codex
Generated-with: Codex
Generated-with: gpt-6.1-sol (Codex)
|
Independent functional review by GPT-6/Codex, separate session 01a160ec-8a99-7c83-b6e9-01276265c6a8, exact head c37f974. Read the complete 13-file diff, supervisor/worker entry paths, main startup and close/save logic, all guard tests, spec/BAT integration and language/dependency changes in an isolated checkout. No blocking finding in this startup/accepted-close containment change. The window stays disabled until the supervisor has permanently disarmed startup termination and published ACK. After that transition, supervisor/IPC errors retain the child and wait rather than imposing an elapsed editing deadline. The teardown marker is emitted only after an accepted save/discard decision and disabling the window; failed document/settings saves and cancelled closes stay outside the shutdown deadline. The guard controls its retained child only. Both build modes/spec route through the pre-Qt runtime hook; the targeted Qt 6.12.0 exclusion is consistent across dependency files. Independent Windows/Python 3.12.10 / PySide6 6.11.1: full suite 266 passed, one existing symlink-privilege skip (22.55s), including all 23 guard tests and their actual bounded child-process probes. The built-in --self-test completed with all checks successful, including temporary-copy document roundtrips. main.py py_compile, compilation of the other changed Python modules, Ruff, diff checks and German umlaut/six-language key verification passed. My custom in-tree pytest temp path initially made Ruff inspect generated fixture scripts; moving those own artifacts outside the checkout restored a clean unchanged-source lint result. All 17 current head checks are successful; main is the default base, PR is non-Draft and CLEAN/MERGEABLE. No package version or workflow edit in the diff, and CONTRIBUTING does not require DCO. No source changes made by this review. This review does not reproduce or identify the historical CPU-spin cause, certify a newly built EXE, establish visible native-window acceptance, update the installed Store package, or authorize Store submission. Those boundaries and the open historical diagnosis remain intact. |
Problem
Three installed 1.0.5 instances were found without a main window and with days
of sustained CPU use (private ticket T-20261022-397249193). The historical spin
could not be reproduced. Both affected inputs open and close successfully in
bounded tests of the actual release. No specific Qt or Markdown cause is claimed.
Change
15 seconds for native close/teardown before stopping the retained child.
refcount crash was confirmed in the Ubuntu 3.10/3.11 CI jobs.
This is a startup/accepted-close containment fix. The historical cause remains
open for diagnosis. The installed Store package is unchanged by this PR.
Validation
with an independent run of all 23 guard tests.
commit on that reviewed base (new head c37f974).
source; both reported inputs (test copies) open on native Windows and close
with exit 0, with no remaining owned processes. The supervisor loads no QtGui.