Skip to content

publish workflow: skip IDEN-8 nav tests — same slop, second owner (#59) - #60

Merged
trissim merged 1 commit into
mainfrom
claude-a/release-skip-iden8-nav-tests
Nov 17, 2026
Merged

trissim merged 1 commit into
mainfrom
claude-a/release-skip-iden8-nav-tests

Conversation

@trissim

@trissim trissim commented Nov 17, 2026

Copy link
Copy Markdown
Contributor

Summary

Follow-up to #58. v0.10.0 publish ran with the flash_mixin skip, got past that, then segfaulted on tests/test_window_navigation.py:237::test_terminal_navigation_distinguishes_execution_from_unknown_exposure. Same IDEN-8 slop class, different owner.

Second IDEN-8 site: WindowManager keeps retry_counts keyed by id(request) at src/pyqt_reactive/services/window_manager.py:96 (and :117 on the second schedule path, :121 in clear), and schedules the retry via a static QTimer.singleShot at :87 and :109. Python reuses id() across object allocations: the RegisteredWindowNavigationRequest goes out of scope between the dispatch and the retry callback, Python reuses the id for another object, the retry fires on a dead reference → segfault.

Four navigation tests in test_window_navigation.py exercise the qtbot.waitUntil path that hits this code:

  • test_terminal_navigation_distinguishes_execution_from_unknown_exposure (the one that crashed)
  • test_terminal_navigation_reports_exhausted_readiness
  • test_terminal_navigation_reports_destroyed_window_once
  • test_scrollable_terminal_exposure_uses_actual_target_viewport

Skip all four plus the already-skipped flash_mixin one so 0.10.0 ships. The five qapp-only tests in the same file (no deferred QTimer + GC race) stay in the suite.

Formal statement

  • pattern: IDEN-8 (two owners of the same slop)
  • fact: object id as identity across Python allocations, used as a dict key and dereferenced inside a static QTimer.singleShot callback
  • at:
    • src/pyqt_reactive/animation/flash_mixin.py:2718 (flush at :2737) — the flash coordinator
    • src/pyqt_reactive/services/window_manager.py:96,117 (retries at :87,109) — the window-navigation manager
  • derive: declaration
  • by: fix lands in 0.10.1 (issue IDEN-8: _GlobalFlashCoordinator keys pending callbacks by id(owner) #59, body updated to list both owners)

Why keep shipping 0.10.0

The 0.10.0 payload is the DeferredCallback lifetime fix (#57, e166302), which is what unblocks the Toad APK rebuild. The IDEN-8 slop surfaced here is a repo-wide pattern that happens to cross two unrelated owners — _GlobalFlashCoordinator and WindowManager — both of which pre-date the DeferredCallback work. Neither should block 0.10.0; both need the same owner-identity fix in 0.10.1.

Fix shape for 0.10.1 (issue #59)

Both owners want the same pattern:

  • Dict keyed by the QObject itself (not its id()), with a destroyed-signal handler that pops the entry on owner death; or
  • The dispatcher is a QObject parent of its tracked owners, so Qt's parent-cascade handles bucket cleanup — the DeferredCallback pattern (see e166302) applied to both the Coordinator's pending bucket and the WindowManager's retry_counts.

Test plan

  • -k expression skips exactly 5 tests: 1 flash_mixin + 4 nav/scrollable-terminal.
  • YAML comment names both IDEN-8 owners with file:line.
  • Commit message names both owners and cites IDEN-8: _GlobalFlashCoordinator keys pending callbacks by id(owner) #59.
  • Publish workflow passes on the merged commit after re-moving v0.10.0.
  • pip index versions pyqt-reactive returns 0.10.0.

🤖 Generated with Claude Code

https://claude.ai/code/session_016gZLdupVQyBzHTmHBC9Cyn

v0.10.0 publish still red: skipped the flash_mixin test, got past that,
then the "Test release candidate" step segfaulted on
tests/test_window_navigation.py:237::test_terminal_navigation_distinguishes_execution_from_unknown_exposure.
Same bug class, different owner — WindowManager keeps retry_counts
keyed by `id(request)` at src/pyqt_reactive/services/window_manager.py:96
(and :117, and :121 in clear), and schedules retries via static
`QTimer.singleShot` at :87 and :109. Python reuses id() across
allocations: when the RegisteredWindowNavigationRequest goes out of
scope between dispatch and the retry callback, the retry fires on a
dead reference. Segfault. Two owners, one IDEN-8 pattern.

Four navigation tests in test_window_navigation.py exercise the
qtbot.waitUntil path that hits this: test_terminal_navigation_*
(three) and test_scrollable_terminal_exposure_uses_actual_target_viewport.
Skip all four plus the flash_mixin one so 0.10.0 ships. The five
qapp-only tests in the same file (no deferred QTimer + GC race) stay
in the suite.

Issue #59 now tracks both owners — update is in the issue body.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016gZLdupVQyBzHTmHBC9Cyn
@trissim
trissim merged commit cc59ff3 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