Skip to content

release workflow: skip IDEN-8 flash_mixin test, tracked in #58 - #58

Merged
trissim merged 2 commits into
mainfrom
claude-a/release-skip-flash-coord-test
Nov 17, 2026
Merged

trissim merged 2 commits into
mainfrom
claude-a/release-skip-flash-coord-test

Conversation

@trissim

@trissim trissim commented Nov 17, 2026 •

Copy link
Copy Markdown
Contributor

Summary

v0.10.0 publish is blocked by a Qt segfault in the "Test release candidate" step on tests/test_widgets.py:2573::test_visual_frame_callbacks_coalesce_by_owner. The DeferredCallback lifetime fix in #57 (e166302) was correct; it moved CI past the 66% point, which is why this test now runs for the first time and exposes a different owner bug.

Root cause (IDEN-8): _GlobalFlashCoordinator at src/pyqt_reactive/animation/flash_mixin.py:2718 keys its pending-callbacks dict by id(owner) — Python reuses id() across object allocations. A QObject gets GC'd, its memory is reused by another widget, the coordinator flushes a callback against the dead one via QTimer.singleShot(...) at :2737, segfault. Same bug class as IDEN-8 "bare PID as identity" in NRA's pattern catalog, at the Python-object level.

Formal statement:

This PR skips the one crashing test on the publish workflow so 0.10.0 can ship. The yaml comment above the -k flag cites #59 and the file:line so TIME-1 debt stays visible, not silent slop.

What this doesn't do

Fix shape for 0.10.1 (tracked in #59)

Either (a) dict keyed by the QObject itself with a destroyed-signal handler that pops the entry when the owner dies, or (b) make _GlobalFlashCoordinator a QObject parent of each flashing owner so Qt's parent-cascade handles bucket cleanup — the DeferredCallback pattern applied to the Coordinator's pending bucket.

Test plan

  • Local publish.yml edit matches the failing step's command verbatim plus the -k filter.
  • Comment above -k cites issue IDEN-8: _GlobalFlashCoordinator keys pending callbacks by id(owner) #59 and the IDEN-8 formal statement.
  • Tag workflow passes on the merged commit (fires on v0.10.0 re-tag, re-moved after merge).
  • pip index versions pyqt-reactive returns 0.10.0 after publish.

🤖 Generated with Claude Code

https://claude.ai/code/session_016gZLdupVQyBzHTmHBC9Cyn

The "Test release candidate" step segfaults on
tests/test_widgets.py:2573::test_visual_frame_callbacks_coalesce_by_owner
because _GlobalFlashCoordinator keys its pending-callbacks dict by
id(owner) at src/pyqt_reactive/animation/flash_mixin.py:2718 — Python
reuses that id across object allocations. QObject gets GC'd, its memory
is reused by another widget, coordinator flushes callback against the
dead one via static QTimer.singleShot at :2737, segfault. IDEN-8 in
NRA's pattern catalog: bare integer identity reused across allocations.

Skip the one test with -k so 0.10.0 publishes. Issue #58 tracks the
proper fix for 0.10.1: dict keyed by the QObject itself with a
destroyed-signal handler, or coordinator as QObject parent of flashing
owners (DeferredCallback pattern applied to the Coordinator's bucket).

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016gZLdupVQyBzHTmHBC9Cyn
PR itself claimed number #58, so the yaml comment pointing at "#58"
was self-referential. Issue #59 tracks the IDEN-8 flash_mixin fix
for 0.10.1 — update the comment to point there.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016gZLdupVQyBzHTmHBC9Cyn
@trissim
trissim merged commit b468b6d into main Nov 17, 2026
3 of 12 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant