[matt2e]

Summary

Fixes two Vitest flakes from the 23–30 Oct CI flake sweep. The RelayStaff toast check asserted before Sonner had shown the toast. Two App.profile tests ran close to the 5 s budget on slow runners. No product code changes, and no timeouts were added or raised. This is one of six independent PRs split from the flakey-tests branch. Each is cut from main and can merge in any order.

RelayStaff.test.tsx:1903 "enforcement error finished while closed reaches reopened detail"

• CI: 39982451663 (job 120512955775), 40045357815 (120725704072), 40217567203 (121241926901), 40538472709 (122310064282), 40547430636 (122341694882), 40881478558 (123473832890), 40923691896 (123623781223).
• Cause: in the logs of 123623781223 and 123473832890, toBeVisible receives the buzz-toast-title element. The toast was in the DOM but hidden. Sonner inserts a toast with data-mounted="false", and its CSS sets opacity 0. A passive effect sets data-mounted at default priority, so the attribute lands in a later commit. The test asserted immediately after awaiting an unrelated button.
• Change: the test waits with waitFor(... toBeVisible()), so visibility is still asserted. "Native enforcement detail" renders inline and stays a direct check.
• Verified: a temporary probe held Sonner's insertion timer and released it just before the assertion, matching CI. The old assertion then fails 2/2 with the CI message ("Received element is not visible", buzz-toast-title), and the new one passes 2/2. Without the probe, the file passes 136/136.

App.profile.test.tsx :261 and :590 (5 s timeouts)

• CI: main 40812491297 (job 123245850311). :261 took 5438 ms and :590 took 5138 ms. The whole file took 65 s there, against 35 s locally, so that runner was about 3.7× slower than this machine.
• Cause: a CPU profile shows that most of each test is jsdom style resolution. Each Settings open costs about 450 ms here, mostly from Base UI's tabbable scan and TabsIndicator measuring through getComputedStyle. Two pieces of that work were not needed for what the tests assert:
  • :261 found the Notifications section button with a dialog-wide role query and clicked it. That button calls openSettingsSection (App.tsx), which opens the section through services.navigation.open with the selected community's scope. The "keeps the same unsent conversation" journey at :657 already clicks the section buttons by role.
  • :590 opened Settings, waited for the profile, and closed it before holding the refresh and reopening. A probe showed that the session's profile cache already held the viewer's kind-0 profile before that first visit, so the first visit filled nothing the reopen read.
• Change: :261 performs the button's navigation itself. It opens Notifications with the scope that openSettingsSection builds from services.communities.snapshot(). The first open stays unscoped, so the test still covers the switch from unscoped to scoped Settings with a page sidebar draft behind it, and it still waits for the Notifications heading and closes with the Close Settings button. :590 drops the first visit, opens once with the refresh held, and is renamed "opens profile settings from the session cache while refreshing". All of its assertions are unchanged.
• Verified: three full-file runs each, before and after, under the same local load. :261 went from 1607–1671 ms to 1290–1326 ms (about −20%), and :590 from 1448–1657 ms to 723–744 ms (about −53%). The file passes 23/23. If ProfileSettings' cachedProfile is made to return null, the new :590 fails (no Display name while loading), so it still guards the cache seed.
• Remaining gap: at 3.7×, :261 would still take about 4.8 s, close to the budget. The rest of its time is jsdom rendering of the product's Settings dialog, which a test-only change cannot cut. This PR reduces that risk but does not remove it.
• Overlaps: fix(composer): synchronize effort picker entrance #1730, Make page-contributed sidebars resizable #1587 and fix(messages): show native attachment download feedback #1556 edit App.profile.test.tsx, but none changes :261 or :590. fix(composer): synchronize effort picker entrance #1730 adds more work to :590.

Left out (causes not confirmed)

• InboxPage.test.tsx "reopens archived DMs durably…" (40916517646, 39996979872): the planned split at view.unmount() would not help. Timing laps show that about 65% of the test runs after the remount, in seven filter-menu changes.
• MemberAdministration.test.tsx "reuses mounted presence… (agent: false)" (40725623425): the filter and the presence unsubscribe both complete synchronously with the keystroke's commit, so there is nothing to wait for. A random-key npub containing "carl" reproduces the signature, but at about 5e-5 odds per run, so the cause is still unconfirmed.

Verification scope

tsc --noEmit and biome check ran on the two changed files. Vitest ran only App.profile.test.tsx and RelayStaff.test.tsx.

Remaining gaps

• Hosted CI has not run yet.
• A green run does not show that these flakes are gone.

🤖 Generated with Claude Code

[kalvinnchau]

🤖 One test-ordering regression to address before merging.

[wesbillman]

🤖 Star Lord’s automated source review via Wes’s account, integrating Rocket, Mantis and Groot coverage.

One test-determinism issue is described inline. The sidebar case still exercises the unscoped-to-scoped navigation transition, with representative section-button interaction retained elsewhere; the toast assertion still requires visibility.

Reviewed head ae0ea65bf9128f1106e4efa17f0830dca8bfd867 against base 36a283b2386473130430890faa3b06c6e45d4a6b. Source-only review of both changed test files, relevant callers/cache lifecycle, PR description and commit metadata; no attachments in the description. No tests or app were run by this review.

Existing hosted evidence: Actions run 41051665992 reports CI required successful. Its full-file artifacts report 23 App.profile tests and 136 RelayStaff tests passing; the two flake shards cover ten combined iterations of 24 selected tests each, with 135 other tests skipped per iteration. The missing synchronization is identified from source; the failing ordering was not reproduced. These passing runs do not prove that all flakes are eliminated, and the author’s local before/after timing claims were not independently reproduced.

[wesbillman]

🤖 Carl, an automated reviewer, commenting via Wes's GitHub account.

No blockers at 49de06e. Test-only, one file, CI green.

• :590: the new waitFor reads session.profiles.event(viewer), the same source cachedProfile in ProfileSettings.tsx seeds from, so it restores the completion barrier the removed first visit provided. If the cache never fills, the test fails loudly rather than passing vacuously.
• :261: driving navigation.open with the scope openSettingsSection builds still covers the unscoped to scoped transition behind the sidebar draft. Real section-button clicks remain at :738/:741. The cost is that this test now mirrors openSettingsSection by hand and won't notice if that function changes.

Non-blocking cleanup before merge:

• The title and body still describe the RelayStaff.test.tsx fix, but the diff no longer touches that file (#2064 already landed it on main). The body's ":590 ... cache already held the profile before that first visit" also predates the explicit wait. A squash merge would carry the stale title into history.
• :630 still says "Reopening uses the cached value immediately", but the test no longer reopens.