Skip to content

Fix the lobby layout test and queue merges - #434

Merged
jserv merged 4 commits into
mainfrom
fix/lobby-entries-merge-queue
Oct 22, 2026
Merged

jserv merged 4 commits into
mainfrom
fix/lobby-entries-merge-queue

Conversation

@jserv

@jserv jserv commented Oct 22, 2026 •

Copy link
Copy Markdown
Contributor

Main's browser lane has been red since fd7c23c: #410's layout test toggled only the problem card's hidden flag, but #401 wraps each card in an entry that filtering hides along with it, so cards inside hidden entries all measured at the top and counted as one row of 57. The test now hides and restores whole entries and counts the cards that are rendered, and the picker outline measures the entries the grid lays out. This is the interaction HellHBBD traced on #295.

Each pull request was green against its own base, and nothing ran them together. The workflow now also runs on merge_group, so enabling a merge queue for main tests each pull request on top of the main it will land on (main has no branch protection or ruleset today, so the queue itself is a settings change). check also collects the mutation plan and shards and is the one status to require; on a pull request that changes Rust it now waits for the shards. The weekly avatar job moves into check.yml, started only by its schedule or a dispatch, and the vendor and Chromium restores become composite actions shared by every job. build and release wait on the gate lanes rather than check, because the default success() would otherwise carry the skipped mutation jobs through check and stop every release.

Two smaller fixes ride along. A whiteboard test pressed Control+Z for a text field's own undo, which only works off macOS, so it now uses ControlOrMeta. tokio moves from 1.54.0, yanked under RUSTSEC-2026-0384, to 1.54.1, which cargo-audit now requires.

Verified locally on macOS: ./scripts/test.sh portable and ./scripts/test.sh rust pass; the browser lane passes except for timeouts under parallel load on this machine, and the affected suites pass 309/309 run serially; actionlint is clean.


Summary by cubic

Fixes the lobby picker layout test to measure filtered entries, and updates CI to test pull requests together on merge queue refs.

  • check now combines gate and mutation results; make it the required status and enable the merge queue in repository settings.
  • Moves the scheduled avatar check into check.yml and shares vendor and Chromium restore actions across jobs.
  • Keeps publishing gated on test lanes, updates whiteboard shortcut tests for host modifiers, and upgrades tokio to 1.54.1.

Written for commit 9baed37. Summary will update on new commits.

View guided diff Turn on auto-fix

cubic-dev-ai[bot]

This comment was marked as resolved.

jserv added 4 commits October 22, 2026 21:31
Since #401 each card sits in an entry that filtering hides along with
it, but #410's layout test toggled only the card and counted cards by
their attribute, so cards inside hidden entries measured as one row of
57. The test now hides and restores whole entries and counts what is
rendered, and the outline measures the entries the grid lays out.
The whiteboard test told the page it runs on Linux and then expected
Control+Z to undo typing in a field, but a field's own undo follows the
host: Chromium on macOS edits text with Cmd, so the test could only
pass on Linux and Windows. It now undoes with the host's modifier and
still presses the board's own chords in the field, so the guard that
leaves text targets alone is checked on every host.
RUSTSEC-2026-0384 reports a race in tokio::time::Interval on 1.54.0,
which is now yanked, and cargo-audit fails the portable lane on it.
1.54.1 carries the fix with no other dependency change, and the
manifest now requires it, so a fresh resolve cannot return to 1.54.0.
Two pull requests green against their own bases broke main together,
because nothing ran them side by side. The workflow now runs on
merge_group, check also collects the mutation plan and shards so it is
the one status to require, the weekly avatar job moves in beside the
gates it shares caches with, and the vendor and Chromium restores
become composite actions every job uses. Publishing waits on the lanes
rather than check, whose skipped mutation jobs would otherwise skip it.
@jserv
jserv force-pushed the fix/lobby-entries-merge-queue branch from 9fa1aca to 9baed37 Compare October 22, 2026 13:31
@jserv
jserv merged commit f8d2115 into main Oct 22, 2026
10 checks passed
@jserv
jserv deleted the fix/lobby-entries-merge-queue branch October 22, 2026 13:41
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