[lucasbraum]

Supersedes #57.

Pack

vitest, >=4 <5 → >=5 <6. Status: verified (2 public repos, 100% precision on breaking findings, recall 50%). The only breaking false positive in #57 (libs/image-utils/src/jest_pdf_snapshot.ts:42) is gone since #60 folds missing-member diagnostics (TS2339/TS2551) into one finding at the repository's own declaration. Rules, expectations, ground truth and companions are unchanged from #57.

Changes over #57:

• The README edit that joined two lines is undone (the line limit no longer counts the packs table).
• verification.json, packs/queue.json and the changeset say verified; the README packs table is regenerated (pnpm docs:packs; --check and pnpm docs:links pass).
• Tests that used vitest as the example of a package without a pack (packages/cli/src/cli.test.ts, packages/core/src/fix/isolate.test.ts) use lodash instead, since vitest now has a verified pack. Only the example package changed.

Sources

• Migrating to Vitest 5.0: https://vitest.dev/guide/migration (read as docs/guide/migration/index.md of vitest-dev/vitest at 69e6369)
• Vitest 5.0.0 release notes: https://github.com/vitest-dev/vitest/releases/tag/v5.0.0
• Benchmarking guide: https://vitest.dev/guide/benchmarking
• Extending matchers: https://vitest.dev/guide/extending-matchers

No official codemod exists for this major.

Rules and behavior notes

Rules: matcher-types, jest-dom-matchers, bench-api, removed-entrypoints (rewrites four entry points), removed-entrypoints-manual, sequential-removed. Behavior notes: coverage-patterns (also a finding), clear-mocks-default, test-name-pattern, inline-projects-extend, hoisted-calls-top-level, mock-behavior, assertion-timing, pretty-format-output, fake-timers-temporal, artifact-locations, config-resolution, browser-mode, environment-requirements, node-api.

Ground-truth repositories

• votingworks/vxsuite @ 5cb3c7d9c1abc891e7b12de57c0278790dd6160a, upgrade commit bb04c5fb01aa9bcd285170d2277df04ad5d7e09c (vitest 4.1.11 → 5.0.2, with @vitest/coverage-istanbul): 30 expected findings.
• vitorvasc/opentelemetry-ecosystem-explorer @ 46d5b37d00bb (directory ecosystem-explorer), upgrade PR head 2b0497b35d0e49dbeeb0d6174ad09aed03d46697 (4.1.11 → 5.0.0): 2 expected findings.

Dropped candidates are listed in #57 (siemens/element cannot install from a public registry; the others are version-only bumps).

uptide pack test output

uptide pack test vitest · verified, recorded candidate   (before --write; now recorded verified)
fixtures      5 cases, 17 sites · pass
ground truth
  votingworks/vxsuite@5cb3c7d9c1ab             vitest 4.1.11 → 5.0.2  predicted 15333, expected 30, false positives 15319, false negatives 16
    with @vitest/coverage-istanbul 4.1.11 → 5.0.2
  vitorvasc/opentelemetry-ecosystem-explorer@46d5b37d00bb vitest 4.1.11 → 5.0.0  predicted 3621, expected 2, false positives 3619, false negatives 0

  rule                         precision  recall    tp   fp   fn
  bench-api                          15%     67%     2   11    1
  coverage-patterns                 100%      0%     0    0   13
  generic                             0%    100%     0 18927    0
  jest-dom-matchers                 100%     85%    11    0    2
  matcher-types                     100%    100%     3    0    0
  overall (sites)                     0%     50%    16 18938   16
  breaking (sites)                  100%     50%    16    0   16

PASS vitest: verified

(The ~18,900 generic (unverified) lines are omitted; the 11 bench-api false positives are unverified inferred sites, not breaking.)

Existing packs, rerun, all unchanged:

zod     verified  overall 100% / 84%, breaking 100% / 71%  (PASS)
stripe  verified  overall 100% / 90%, breaking 100% / 90%  (PASS)
ai      verified  overall 100% / 77%, breaking 100% / 36%  (PASS)
react   verified  breaking 100% / 69% (89 sites, 0 false)  (PASS)

Environment note: tldraw (react) pins Yarn Berry and the sandbox yarn is classic, so the react run used a corepack shim on PATH (sandbox only). With that shim on PATH, stripe's nextjs/saas-starter failed to install (MODULE_NOT_FOUND); stripe passes without the shim. Environment issues, not regressions.

pnpm lint, pnpm typecheck and pnpm test pass.

What is not covered

• Pack detect runs only on files that import the package, so these expected sites are missed: 13 bare-directory coverage.include/exclude entries in configs that import a shared defineConfig wrapper, and jest-dom imports in two setup files that import nothing from vitest. Breaking recall is 50%.
• check could not resolve imports in 69 of 71 vxsuite files (workspace packages are not built), so only a few of the 169 compiler diagnostics are visible there.
• bench-api misses the describe wrapper at render_ballot.bench.ts:16.
• Rules and notes with no evidence in either repo, resting on fixtures and the guide only: removed-entrypoints*, sequential-removed, nested vi.mock, and all behavior notes except coverage-patterns.
• The two repos are one monorepo and one small app; a third that really migrated would make "verified" stronger.

Tooling friction

• pnpm uptide check from source fails inside its worker (Cannot find module .../packages/core/src/errors.js); pnpm build and the dist CLI work. pack test is unaffected.
• pnpm packs:truth requires full 40-character SHAs and prints the new path with the old line number for moved files.
• A corepack shim on PATH fixes Yarn Berry repos but breaks the pnpm install of nextjs/saas-starter in the same run; the packs had to be run with and without it.
• vitest.dev is not reachable from the sandbox.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NwSXH7Dy3r4KjJhwKYBuxT

---

Generated by Claude Code

[lucasbraum]

Rerun of pack test with this PR's 3 commits cherry-picked onto current main (c0f8a70, which adds #62 after this PR's base 7ec02e0). CI on the PR head is green, there are no review comments, and the PR has no conflict. Nothing to push.

• vitest: PASS, verified. Numbers are identical to the PR body: breaking 100% precision / 50% recall (16 tp, 0 fp, 16 fn). The same 16 false negatives, and the same 11 unverified bench-api sites.
• zod: breaking 100% / 71%, overall 100% / 84%. Unchanged.
• stripe: PASS, 100% / 90%. Unchanged.
• ai: PASS, breaking 100% / 36%, overall 100% / 77%. Unchanged.
• react: FAIL in this sandbox only. The tldraw install fails because the repo pins yarn@4.7.0 and the sandbox global is Yarn 1.22.22 (environment issue, no corepack shim used this time). Without tldraw: breaking 100% / 80% (57 tp, 0 fp). The one non-breaking generic false positive is at excalidraw CollabError.tsx:22. Not a regression from this PR.

packs:fetch also ended with an ELIFECYCLE on one unrelated ground-truth install; the vitest repos installed fine.

---

Generated by Claude Code