Repository navigation
Run browser tests in parallel on per-worker accounts - #72
Merged
Merged
Conversation
Each Playwright worker now creates its own user and group through the backend, signs in once, and reuses that session. Specs arrange trips, ideas, issues, and members through @groam/app-actions/backend and drive only the journey under test through the UI, so the suite runs with fullyParallel and CI needs two shards instead of four. The dev container also installs Chromium and its libnss3/libnspr4 dependencies so browser tests run without manual setup. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
The e2e fixture imports the backend actions, which reach packages/ai through its #ai subpath imports; map them as the seeder and app-actions tsconfigs already do. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A shared Vitest setup routes console output through a guard: tests declare output they produce on purpose with expectConsole(), which hides it, and any other output fails the test. Every Vitest package now loads it. Turning it on surfaced real problems behind the CI noise: - Scheduled Idea reviews loaded their context through the signed-in viewer, so every review failed with "Not authenticated". The review context now reads the stored proposal directly, like the title agent. - convex-test ran scheduled functions on real timers, so AI runs and Wikipedia cover lookups ran behind unrelated tests. Backend tests now fake setTimeout; agent tests opt back into real timers. - The geospatial index logged every query at INFO. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Trip and idea redirects passed the resolved Convex ID back into the URL while ReferenceProvider rewrote it to the short ID, so opening /trips/<convex id> looped on "Opening page…". Redirects now keep the reference as written and let ReferenceProvider canonicalize it once. The backend addIdea action now names ideas by title, as the UI does, and openSharedTrip waits for the canonical URL like the other open helpers. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Lint, typecheck, unit tests, browser tests, and desktop builds now start together as separate jobs instead of waiting on one quality job, and the browser suite runs as a single job. The showcase recording test declares its progress output, and createTrip waits for the trip list to load before opening the dialog. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Component and view tests run in real Chromium through the `views` setup of `vp test` (page locators, retrying expect.element) instead of jsdom and Testing Library. Browser e2e becomes the `journeys` setup: the real app against the local backend, driven by @groam/app-actions/vitest, with backend arrangements through Vitest browser commands. Playwright Test, jsdom, and Testing Library are gone; the showcase keeps Playwright for recording. Real browsers surfaced app fixes: per-group workspace remounts (no stale pagination cursors), passkeys trusting every local origin on their host, a single accessible "Attach files" control, labelled issue assignee rows, and no React warning from the theme script. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Journeys now run in Vite+ browser mode, so the Playwright layer keeps only what the showcase recording uses. Share the vitest types across packages and split the itinerary fixture data from its stateful editor. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The 4-core runner starved the local Convex backend with four browsers, so queries hit the 1s function limit. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The server ends the Better Auth session before the client hears of it, so workspace queries still mounted reran signed out and threw. The app shell now unmounts the workspace first, and journeys leave the page before switching people. CI journeys get a timeout sized for the shared runner and upload their failure screenshots. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
requireWorkspace built a full Better Auth instance and called getSession and getActiveMemberRole on every guarded function: several component round trips that made up about 70% of backend CPU in journeys and pushed queries past the 1s limit on CI. One component query now reads the session, user, and membership; the guard keeps the same errors and also rejects a session that belongs to someone else. Sign-out recovers when the request throws. CI warnings: - Remove the dead desktop copy_dir helpers and the unused Turbo cache. - Let the vendored glib allow the newer rustc lints it trips. - Size the chunk warning for the single maplibre-gl module. - Pin Ubuntu runners to 24.04. - Keep Fallow from reading the backend dev script's -C as an entry. - Share the phone list-row checks between the idea and issue lists. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…t drivers Better Auth replaces or ends the session (sign-out, two-factor on or off, password change) and switches the active group on the server before the client hears of it, so mounted Convex subscriptions reran against the old session or with the old group's pagination cursors. The app now hides the workspace in an Activity, dropping its subscriptions, until Convex or the session reports the change. A session index stops listSessions from scanning every session, which returned duplicates through the Better Auth adapter's pagination. Journey drivers lose their retry loops, visibility branches, and dead helpers; absence checks assert a landmark first; CI drops dead Turbo settings, checks formatting once, pins runner images, and caches Rust builds; docs and lint rules no longer describe Playwright journeys. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The repository only allows GitHub-owned actions and a short allowlist, so Swatinem/rust-cache failed the workflow at startup. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fallow's hotspot analysis needs git history, so Lint checks out the full history again. A failing journey now keeps its trace in the uploaded failure artifact. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
1.6.26's client session atom read itself with get() while nanostores tore it down, which mounted it again mid-teardown and left it active without a refresh manager. A remount soon after an in-flight session fetch then never refetched after sign-in or sign-up, so journeys that switch people stayed on the auth screen under load. 1.6.33 reads the value directly. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Playwright 1.63 arms ProgressController's timer with an unclamped `deadline - monotonicTime()`. Vitest polls elements through Playwright with 1 ms budgets, so on a busy CI runner the deadline has already passed and Node prints TimeoutNegativeWarning in the journeys job. Patch it the way the same file clamps its other deadline timer, until a release does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
Browser tests were the slowest CI job: 4 shards × 6–8.5 min of
Run browser tests(~29 job-minutes, ~10.5 min wall onmain@ 339509a). Every spec signed in as the one shared demo user, built its trips/ideas/issues through the UI, and the suite ran serially within each file because all tests shared data.What
@groam/app-actions/playwright/testfixture: each worker creates its own user + group through the backend, signs in once, and reuses thatstorageState. Tests share no data, sofullyParallel: true.arrangefixture creates trips, ideas (startIdeareturns the working trip for idea edits), issues, review requests, and members. Specs drive only the journey under test through the UI; second users getnewSignedOutPage().openApp,openTrip,openIdea,openIssuewait for the canonical short-ID URL and return it.signIn(demo user),addGroupMemberViaInvite, theSEED_USER_*Playwright env, and the CI seed step. Sign-up / sign-in / invite flows keep their own UI specs.listreporter so logs show per-test durations. Jobs are nowBrowser (n/2).post-create.shandlibnss3/libnspr4in the Dockerfile (Chromium could not start before). Needs a container rebuild.Verification
vp check, conventions, React Doctor, Fallow — pass.mainbelow.Timings
Run browser testsper shardmain@ 339509a🤖 Generated with Claude Code