Finish 0.9.21: version bump, two test races, signed-commit rule #265

Merged
ethan-hann merged 4 commits from release-0921-version into main 2026-10-28 19:58:12 -05:00
Owner

Finishes the 0.9.21 release. The batch (PR 264) merged without the version bump, so the release workflow refused tag v0.9.21: Directory.Build.props still said 0.9.20.

  • Set the version to 0.9.21.

  • FakeEventDeliveryTests: the background-delivery test compared thread ids after an await and flaked on the Linux VM; it now checks that no event is raised on the caller's thread while the call runs.

  • #266: the Games screen dropped its loading flag before the rows had their stats (or the error card was set), so anything that saw loading end could read rows still showing a dash; it failed Linux CI on this PR. The flag now drops last, with two tests that record the screen at the moment loading ends (red before: Expected "1.1.0.4", Actual "-"; and an Assert.True on the error card).

  • AGENTS.md: every commit is signed, including agents' commits; when signing fails the coordinator arms it, and nobody commits unsigned.

Opus review: 0 critical, 0 warnings, 7 notes filed as #269. Same pattern on the profile preview: #268.

Local gates on a41656d6 (the last code commit; 134ba70b is AGENTS.md only): Windows gate passed; Linux gate on the VM passed (535 s).

Closes #266

Finishes the 0.9.21 release. The batch (PR 264) merged without the version bump, so the release workflow refused tag v0.9.21: Directory.Build.props still said 0.9.20. - Set the version to 0.9.21. - FakeEventDeliveryTests: the background-delivery test compared thread ids after an await and flaked on the Linux VM; it now checks that no event is raised on the caller's thread while the call runs. - #266: the Games screen dropped its loading flag before the rows had their stats (or the error card was set), so anything that saw loading end could read rows still showing a dash; it failed Linux CI on this PR. The flag now drops last, with two tests that record the screen at the moment loading ends (red before: Expected "1.1.0.4", Actual "-"; and an Assert.True on the error card). - AGENTS.md: every commit is signed, including agents' commits; when signing fails the coordinator arms it, and nobody commits unsigned. Opus review: 0 critical, 0 warnings, 7 notes filed as #269. Same pattern on the profile preview: #268. Local gates on a41656d6 (the last code commit; 134ba70b is AGENTS.md only): Windows gate passed; Linux gate on the VM passed (535 s). Closes #266
Check the background delivery test by the call window, not thread ids
Some checks failed
Docs / site (pull_request) Successful in 1m26s
Gate / linux (pull_request) Failing after 15m21s
Gate / windows (pull_request) Successful in 34m44s
55fa9dcf80
The test compared each event's thread id with the test's thread id after awaiting the delivery. After an await the test runs on a pool thread, and that can be the very thread that delivered the events, so the check failed by chance on the Linux VM gate:

  Assert.NotEqual() Failure: Values are equal
  Expected: Not 36
  Actual:       36

A thread-local flag now marks the caller's thread only while CaptureAsync and Complete run, with nothing awaited inside the window, and every event must arrive outside it. Mutation check: with the fake switched to inline delivery the test fails for both events (Assert.False, Actual True); with background delivery it passes.
ReloadAsync cleared IsLoading before it applied the row stats or set the
load error, so anything that saw loading end (a change notice, or a test
polling from another thread) could still find rows showing "-" or no
error card. The flag now drops last, on both the success and the failed
list paths, so loading ending means the load's result is in place.

Tests first: two deterministic checks record the state at the moment
IsLoading turns false. Observed reds before the change:
- Loading_ends_only_once_the_rows_hold_their_stats:
  Assert.Equal() Failure: Collections differ at index 0
  Expected: "1.1.0.4"  Actual: "-"
- Loading_ends_only_once_a_failed_profile_list_shows_its_error_card:
  Assert.True() Failure  Expected: True  Actual: False
The flaky wait-then-read test (rows fill in when the stats land) now
relies on that contract and passed six repeated class runs.
Require a signed commit everywhere, and arming signing instead of skipping it
All checks were successful
Docs / site (pull_request) Successful in 1m28s
Gate / linux (pull_request) Successful in 14m55s
Gate / windows (pull_request) Successful in 32m27s
Docs / site (push) Successful in 1m33s
Release / linux (push) Successful in 2m22s
Release / windows (push) Successful in 2m47s
Release / checksums (push) Successful in 33s
134ba70ba4
AGENTS.md said contributed commits need not be signed, which let an implementer offer to commit unsigned when signing was not armed. Every commit in this repository is signed; when signing fails the coordinator arms it and the commit is made again.
ethan-hann changed title from Set the version to 0.9.21 and fix a racy fake-delivery test to Finish 0.9.21: version bump, two test races, signed-commit rule 2026-10-28 19:25:08 -05:00
Sign in to join this conversation.
No reviewers
No milestone
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
ethan-hann/SavePoint!265
No description provided.