[kixelated]

Finishes quest/m0/worker-tests.md, a follow-up of kixelated/quest#188.

Summary

• Root cause: the Worker tests run in one shared runtime (isolate: false) against one D1 database migrated once in test/setup.ts, and @cloudflare/vitest-plugin 1.x no longer isolates storage per test. Each file cleaned up a hand-picked list of tables with DELETE FROM, so any table (users, accounts, sessions, verifications, R2 objects, Durable Objects) left off a file's list carried over into later tests.
• Fix: test/setup.ts now calls reset() from cloudflare:test (it deletes every Durable Object, which also backs Miniflare's D1 and R2) and reapplies the migrations before each test. So every test starts from empty, freshly migrated storage.
• Cleanup: the per-file DELETE FROM lists in beforeEach are gone, and seeds that used INSERT OR IGNORE to tolerate leftovers are now plain INSERT, so a leak would fail loudly. This includes the second-user insert from kixelated/quest#188. Mid-test deletes that are part of what a test checks stay.
• CI check: vitest.config.ts sets sequence.shuffle: true, so every run, local or CI (just test → just worker-test), shuffles both files and tests and prints its seed. A new order dependence fails there; rerun it with vitest run --sequence.seed=<seed>.
• The full suite didn't get slower: 26s with the reset, against 31s before on the same host.

Verification

• Each of the 267 test locations (file:line; an it.each table counts as one location, and its cases still reset between them) ran alone in its own vitest process: 267/267 passed.
• Full suite, shuffled with seeds 1, 2 and 3: 333/333 passed each time, plus unshuffled runs before and after the cleanup.
• A probe test (not committed) confirmed that reset() drops a D1 table and an R2 object written by the previous test, and that the migrations are reapplied.
• prettier --check, tsc --noEmit (cloudflare) and quest check pass. The full just check / just test waited more than 30 minutes behind quest gate on a host loaded by about 15 parallel agents, so I left the rest of those to CI.

Choices made without asking

• Reset in a global beforeEach in setup.ts rather than per file, so a new test file is isolated by default.
• reset() plus reapplying the migrations, rather than deleting rows from every table. It also clears R2 and Durable Objects, needs no table list, and measured no slower.
• Shuffle on every run, rather than adding a separate shuffled CI job. One run covers both, and the printed seed reproduces any failure.
• The per-test-alone proof was a one-off script and isn't committed: running each case alone costs a vitest startup per test (about 267 processes), which is too slow for CI. The reset plus shuffle is the standing check.

Suggested follow-ups

• None required. If a module-level JS fixture ever leaks between tests (the reset covers storage only), a periodic job that runs each test alone could catch it. Not needed today: all 267 locations pass alone.

🤖 Generated with Claude Code

(Written by Claude Opus 5.5)

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-24T16:33:52.240062Z	ebbe689	New commits
🔒 Security Review	✅ Completed 2026-10-24T16:34:56.927713Z	ebbe689	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[kixelated]

Merged main into the branch (ac5fee2), no rebase.

• Conflict: quest/m0/README.md. Both sides removed a finished quest from Required: kixelated/quest#218 removed Lifecycle flakes, and this PR removes Worker tests. The resolution drops both lines. Everything else merged cleanly.
• Leftovers from the newly merged tests: none. Every DELETE FROM still in cloudflare/test is a mid-test step that the test checks, such as a lost delivery record, a disconnected project, or a finished job. The ON CONFLICT upserts in project-fixtures.ts let tests switch a project between paused and enabled within one test, so they aren't there to tolerate leftover rows. No INSERT OR IGNORE remains.
• Verification: the full Worker suite passed shuffled with seeds 1 to 5 (17 files, 347/347 tests each), and just check and just test both passed.

(Written by Claude Opus 5.5)

[kixelated]

Merged main again (ebbe689) after kixelated/quest#217 landed.

• Conflict: cloudflare/test/seed.test.ts. kixelated/quest#217 added an edit parameter to seeded(), and that function still began with a DELETE FROM batch over jobs, runners, projects, and installations. The resolution keeps main's seeded(edit) signature and drops the batch, since the global reset in test/setup.ts already gives every test empty storage.
• Verification: the Worker suite passed shuffled with seeds 1 to 5 (348/348 tests each), and just check and just test both passed.

(Written by Claude Opus 5.5)