refactor: replace @voidzero-dev/vite-plus-test with upstream vitest#1588
Open
Brooooooklyn wants to merge 110 commits into
Open
refactor: replace @voidzero-dev/vite-plus-test with upstream vitest#1588Brooooooklyn wants to merge 110 commits into
Brooooooklyn wants to merge 110 commits into
Conversation
✅ Deploy Preview for viteplus-preview canceled.
|
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 365a61de42
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Brooooooklyn
added a commit
that referenced
this pull request
May 15, 2026
…mat/typo PR #1588 CI failures: - Force-override mode (VP_FORCE_MIGRATE=1, set by test-vp-create.yml and ecosystem-ci) now re-pins any pre-existing vite-plus range to the local tgz path in monorepo workspace packages. Without this, pnpm reads the published vite-plus@0.1.21 metadata to resolve transitive deps including @voidzero-dev/vite-plus-test@0.1.21, which shadowed upstream vitest@4.1.5 at runtime and broke vp create monorepo tests. - typos CI: rename yarn-PnP to yarn Plug'n'Play (Pn→On false positive). - vp check: format packages/cli/build.ts shim-generation block. - Rename __dirname in install-failure-guard.spec.ts to satisfy eslint(no-underscore-dangle).
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 4fafa67971
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: ec69abaadb
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 381b6e2c20
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Brooooooklyn
added a commit
that referenced
this pull request
May 15, 2026
…mat/typo PR #1588 CI failures: - Force-override mode (VP_FORCE_MIGRATE=1, set by test-vp-create.yml and ecosystem-ci) now re-pins any pre-existing vite-plus range to the local tgz path in monorepo workspace packages. Without this, pnpm reads the published vite-plus@0.1.21 metadata to resolve transitive deps including @voidzero-dev/vite-plus-test@0.1.21, which shadowed upstream vitest@4.1.5 at runtime and broke vp create monorepo tests. - typos CI: rename yarn-PnP to yarn Plug'n'Play (Pn→On false positive). - vp check: format packages/cli/build.ts shim-generation block. - Rename __dirname in install-failure-guard.spec.ts to satisfy eslint(no-underscore-dangle).
Brooooooklyn
force-pushed
the
refactor/replace-vite-plus-test-with-vitest
branch
from
5c48da8 to
39efcbf
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 39efcbf239
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 2fceee5296
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b888329d76
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Member
Author
Upstream blockers (still needed after the fixes in this PR)
|
Member
Author
Upstream blockers (updated)After the latest fixes (
|
Member
Author
Final status (after b6b5b8b)The aggressive Rust source-rewrite in 21937c5 (which made Confirmed upstream blockers:
|
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: b6b5b8b0bf
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 60b7d0f0ac
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Member
Author
Final status — branch HEAD 3ffb7cdE2E now at expected steady state. The two remaining upstream items:
Everything else (npmx.dev, vinext, vue-mini, frm-stack, varlet, vitepress, reactive-resume, rollipop, dify, etc.) is now green. |
Brooooooklyn
added a commit
that referenced
this pull request
May 16, 2026
…mat/typo PR #1588 CI failures: - Force-override mode (VP_FORCE_MIGRATE=1, set by test-vp-create.yml and ecosystem-ci) now re-pins any pre-existing vite-plus range to the local tgz path in monorepo workspace packages. Without this, pnpm reads the published vite-plus@0.1.21 metadata to resolve transitive deps including @voidzero-dev/vite-plus-test@0.1.21, which shadowed upstream vitest@4.1.5 at runtime and broke vp create monorepo tests. - typos CI: rename yarn-PnP to yarn Plug'n'Play (Pn→On false positive). - vp check: format packages/cli/build.ts shim-generation block. - Rename __dirname in install-failure-guard.spec.ts to satisfy eslint(no-underscore-dangle).
Brooooooklyn
force-pushed
the
refactor/replace-vite-plus-test-with-vitest
branch
from
3ffb7cd to
a0d248e
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a0d248e65e
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 1a5b2697e5
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…installs @vitest/browser-webdriverio@4.1.7 declares `webdriverio` as a non-optional peer dep. pnpm v10's autoInstallPeers (default true) then installs webdriverio → edgedriver/geckodriver, which have postinstall scripts pnpm v10 refuses to run without explicit allowBuilds entries. Every vite-plus user on pnpm v10 hit the warning at install time, even if they don't use vitest browser mode. vp migrate now writes the allow/deny decision up front so users don't have to manage it themselves: - pnpm v10+ workspace YAML or `pnpm.allowBuilds` in package.json: writes edgedriver/geckodriver entries (true when webdriverio is a real dep, false otherwise) - pnpm v9 onlyBuiltDependencies: appends only when webdriverio is used (v9 omission already denies, so no-op otherwise) - Pre-existing user entries are respected — re-runs are idempotent - Monorepo detection scans root + all sub-packages Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…tries vp migrate now writes default edgedriver/geckodriver allowBuilds entries to pnpm-workspace.yaml on every pnpm v10+ project, so all pnpm migration snaps gain a 3-line allowBuilds block. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…lowBuilds vp create on a pnpm v10+ workspace now emits default edgedriver/geckodriver allowBuilds entries. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…eration When regenerating pnpm migration snaps for the allowBuilds change, the local run captured two environment-specific outputs that differ from CI: - command-list-yarn1: local yarn1 emitted a "react-compiler-runtime" link-missing warning that CI's yarn1 doesn't - create-next-command-monorepo-library: vp create exited 1 locally instead of scaffolding successfully (likely an environmental side-effect of the preceding regen passes) Both snaps are restored to their pre-regen content. The allowBuilds-driven 3-line additions only apply to pnpm migration paths; neither of these tests touches the allowBuilds code. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
CI's `vp check` flagged formatting in the new allowBuilds helpers. Running oxfmt locally aligns the indentation/wrapping with the rest of the file. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…uccess Reviewer cursor[bot] (PR #1588): when a pre-migration install succeeds but the post-migration reinstall fails, `installDurationMs` is the sum of both and is still > 0, so showMigrationSummary printed a green ✔ "Dependencies installed" line alongside the warning — misleading because node_modules is desynced. executeMigrationPlan now returns finalInstallOk (= finalInstallSummary .status === 'installed'); main does the same for the early-return hasVitePlusDependency branch. showMigrationSummary gates the green line on `finalInstallOk && installDurationMs > 0`. Also: .github/scripts/upgrade-deps.ts — updateTestVpCreateWorkflow now calls recordChange so vitest bumps to the workflow are tracked in the generated PR metadata, matching every other update function in the file. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…w signal Reviewer chatgpt-codex-connector[bot] (PR #1588): when a user has `@vitest/browser-webdriverio` in deps but no direct `webdriverio`, hasOwnWebdriverioDependency returned false → allowBuilds wrote edgedriver/geckodriver as `false`. But then rewritePackageJson removes @vitest/browser-webdriverio and injects `webdriverio: '*'` as its peer (via REMOVE_PACKAGES + BROWSER_PROVIDER_PEER_DEPS), so the migrated project ends up needing the drivers — and the allowBuilds decision already denied them. Now hasOwnWebdriverioDependency treats either `webdriverio` or `@vitest/browser-webdriverio` as the signal. workspaceUsesWebdriverio inherits the change transitively (delegates to the same helper). Added unit test covering the @vitest/browser-webdriverio-only fixture. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Reviewer chatgpt-codex-connector[bot] (PR #1588): hasOwnWebdriverioDependency only checked dependencies/devDependencies/optionalDependencies, but rewritePackageJson's REMOVE_PACKAGES loop also iterates peerDependencies — so `@vitest/browser-webdriverio` declared as a peer was removed and replaced with `webdriverio: '*'` while the allowBuilds decision had already denied the driver postinstalls. Extend the helper (and DependencyBag) to scan peerDependencies. Added a unit test covering the peer-only fixture. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…YAML Reviewer chatgpt-codex-connector[bot] (PR #1588): cleanupPnpmOverridesFor- WorkspaceYaml already deletes REMOVE_PACKAGES entries from `pkg.pnpm .overrides`, but the YAML-side rewritePnpmWorkspaceYaml never did the same sweep on `pnpm-workspace.yaml`'s `overrides:` block. A workspace that had pinned e.g. `@vitest/browser-playwright: 4.0.0` would keep that override after migration — and since vite-plus now direct-deps the provider at VITEST_VERSION, the stale override forced a version mismatch with the bundled vitest. rewritePnpmWorkspaceYaml now drops every REMOVE_PACKAGES key from the YAML overrides map after the legacy-wrapper-alias prune. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Reviewer chatgpt-codex-connector[bot] (PR #1588): the previous override cleanup only matched exact REMOVE_PACKAGES keys, so pnpm override selectors like `@vitest/browser-playwright@4: 4.0.0` or `vite-plus>@vitest/browser-playwright: 4.0.0` survived the migration. Those selectors then applied against vite-plus's own direct provider dependency, re-creating the version-skew bug the bare-key cleanup was meant to prevent. Added a small helper `extractOverrideTargetName` that strips both the parent-selector prefix (`parent>`) and the trailing version (`@1`) while preserving scoped names (`@scope/pkg`). The rewritePnpmWorkspaceYaml sweep now iterates the YAML overrides snapshot, parses each key through the helper, and deletes any whose target package is in REMOVE_PACKAGES. Extended the existing test fixture to also include `@vitest/browser- playwright@4`, `vite-plus>@vitest/browser-playwright`, and an unrelated `unrelated>some-other-pkg` selector — asserts the browser selectors are dropped while unrelated selectors survive. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
CI's oxlint flagged the single-statement `if (!trimmed) return trimmed;` introduced in the last commit. Wrapping it in braces aligns with the project's curly rule. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
… pkg.json overrides Two open review issues on the pnpm v10 migration path: * `allowBuilds.<driver> = false` was unconditionally written for `edgedriver` and `geckodriver` whenever the webdriverio signal forced the migration to manage them. If a project already depended on either driver directly (e.g. a non-webdriverio Selenium setup), the migration silently blocked their own postinstall. Collect `directDriverDeps` from the user's package.json and skip those keys in both the `pnpm-workspace.yaml` and `package.json` allowBuilds writers — pnpm prompts on first install as usual. * The selector-aware override cleanup added for `pnpm-workspace.yaml` in 695dcf5 did not apply to the `package.json` `pnpm.overrides` sink, so selector-shaped keys like `vite-plus>@vitest/browser-playwright` survived migration and could re-pin vite-plus's own provider dep to an incompatible version. Extract the deletion as `dropRemovePackageOverrideKeys` and invoke it from the standalone package.json path, `cleanupPnpmOverridesForWorkspaceYaml`, and the force-override root-workspace path. Reuse the shared helper inside `rewritePnpmWorkspaceYaml` so all sinks go through the same matcher. Adds five regression tests covering both fixes. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
The prefer-vite-plus-imports oxlint rule rewrote any `vitest/<sub>` to `vite-plus/test/<sub>`, including `vitest/package.json`. But upstream vitest exposes `./package.json` (resolvable today) while the generated `vite-plus` exports map deliberately omits `./test/package.json` (syncTestPackageExports skips upstream's `./package.json`). So under `vp lint --fix` the autofix turned a working metadata import into one that fails with ERR_PACKAGE_PATH_NOT_EXPORTED. Skip `vitest/package.json` in rewriteVitePlusImportSpecifier, mirroring the migrate rewriter's existing exclusion in import_rewriter.rs (`^['"]vitest/(package\.json|config)['"]$`). Other `vitest/<sub>` specifiers still rewrite normally. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…orts
Migration rewrites the `vitest/config` specifier to bare `vite-plus`
(Rust import_rewriter + the prefer-vite-plus-imports oxlint rule), so
the `vite-plus` root must re-export every runtime `vitest/config` symbol
or migrated `import { x } from 'vitest/config'` (and the require form)
resolves to undefined. index.ts (ESM) and index.cts (CJS) are separate
manual lists, so add superset guards on BOTH module systems — a future
vitest bump that adds a config export can't be fixed in one entry while
the other silently regresses. Each guard includes a non-empty sanity
assertion so it can't false-pass if enumeration breaks.
Type-only exports are out of scope here: removal-drift in index.ts's
`export type` list is already caught by the repo typecheck.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
vp migrate of a pnpm v10 monorepo wrote `allowBuilds.<driver> = false` for edgedriver/geckodriver even when the root or a workspace package already directly depended on that driver, as long as no package used webdriverio — denying a postinstall the user already relied on. The standalone path collects `directDriverDeps` (via collectDirectDriverDeps) and threads it into the allowBuilds writers, which skip user-owned drivers. The monorepo path never collected them, so both pnpm v10 sinks (pnpm-workspace.yaml and the root package.json `pnpm.allowBuilds`) got an empty set. Add `collectWorkspaceDirectDriverDeps`, mirroring workspaceUsesWebdriverio's root+sub-package traversal, and thread it through rewriteMonorepo -> rewritePnpmWorkspaceYaml / rewriteRootWorkspacePackageJson -> applyBuildAllowanceToPackageJsonPnpm. The v9 onlyBuiltDependencies path (omission == denial) is unaffected. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…is present The direct-driver skip in the pnpm v10 allowBuilds writers suppressed the write unconditionally, so in a mixed setup — a package depends on edgedriver/geckodriver directly AND webdriverio is also present — the driver key was left absent instead of `true`. Because the allowBuilds sink is workspace-global, that absent key forces a pnpm prompt for a build webdriverio actually needs. Gate the skip on `!shouldAllow`: only suppress the `false` deny path (no webdriverio, so don't override a user-managed driver). When webdriverio makes builds allowed, write `true` even for a user-owned driver. Updates the standalone webdriverio+direct-driver test to the new expectation and adds a mixed-workspace monorepo regression (package A edgedriver, package B webdriverio -> both drivers allowed). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Per review feedback ("avoid defaulting to depend on @vitest/browser-webdriverio,
its usage is low"), webdriverio is no longer a bundled runtime dependency of
vite-plus. It moves to a devDependency + optional peerDependency (catalog:), and
`vp migrate` now KEEPS the provider in the user's own deps (pinned to the bundled
VITEST_VERSION) instead of stripping it, so the rewritten
`vite-plus/test/browser-webdriverio` import still resolves under pnpm strict /
Yarn PnP. playwright/preview stay bundled and stripped as before.
Migration details:
- Drop only stale FLAT override/resolution pins that TARGET the provider, so a
leftover pin can't force the provider off the migrated 4.1.7 dep (npm EOVERRIDE
/ yarn-bun version skew). Catalog entries are preserved to avoid dangling
`catalog:` references.
- extractOverrideTargetName parses pnpm `>` parent selectors via pnpm's own
delimiter rule (/[^ |@]>/) so semver comparator ranges (`@>=4`, `@>4`, `@<5`)
aren't mis-split, and takes the trailing descriptor of a yarn from/target
selector (matching @yarnpkg/parsers).
- Detect webdriverio usage from source for pre-migration
`@vitest/browser-webdriverio*`, and the migrated
`vite-plus/test/browser-webdriverio*`,
`vite-plus/test/browser/providers/webdriverio*`, and
`vite-plus/test/plugins/browser-webdriverio` (shim) specifiers, so re-runs and
source-only projects still get the provider + `webdriverio` peer injected and
the edgedriver/geckodriver builds allowed.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The comparator-range test comment used "mis-parses"; crate-ci/typos tokenizes the hyphenated word and flags the standalone `mis` token, failing the CI Lint job before oxlint even runs. Reword to "is parsed incorrectly" — no behavior change, comment only. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
`usesVitestBrowserMode` only matched `@vitest/browser` and
`vite-plus/test/browser`, so a package importing the published browser
shims that carry no `browser` substring went undetected: the import
rewriter flattens `@vitest/browser/{client,locators,matchers,utils}` to
`vite-plus/test/{…}`, and `build.ts` also publishes the bare
`vite-plus/test/context`, the `vite-plus/test/plugins/browser*` family,
and `vite-plus/test/internal/browser`. With no `@vitest/browser*` dep
for the dep-based signal, `effectiveBrowserMode` stayed false and the
direct `vitest` pin (which the browser optimizer needs resolvable from
the package root under pnpm strict / Yarn PnP) was skipped.
Add hints covering the full published browser surface — verified
complete against every browser-surface `./test/*` export — and a
parametrized regression asserting `vitest` is pinned without injecting
`@vitest/browser`. A single `vite-plus/test/plugins/browser` prefix
covers the whole plugin family; all added subpaths are browser-only
re-exports, so they never collide with a non-browser vitest export.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Reconciles the branch's dependency changes (es-module-lexer, oxc-parser, opt-in @vitest/browser-webdriverio, removal of @voidzero-dev/vite-plus-test) with main's upstream dependency upgrades (#1713). Verified: frozen-lockfile install passes, @vitest/mocker resolves at the patched 4.1.7, and there are no stale @voidzero-dev/vite-plus-test references. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Brooooooklyn
force-pushed
the
refactor/replace-vite-plus-test-with-vitest
branch
from
62a30a2 to
d646df0
Compare
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d646df0712
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…d Vitest
`vp test` resolved the runner binary CWD-first, so a project that keeps its
own `vitest` dependency would spawn the project's Vitest while the
`vite-plus/test*` shims (`export * from 'vitest'`) resolved to vite-plus's
bundled copy — two physical Vitest instances in one run, a classic source of
internal-state and mock-hoisting mismatches.
- constants.ts: add `resolveBundled()` (CLI-dir-first, project-cwd fallback)
alongside `resolve()` (cwd-first).
- resolve-test.ts: resolve the runner binary via `resolveBundled('vitest/package.json')`
so `vp test` always runs the bundled (pinned) Vitest.
- define-config.ts: flip `vite-plus:vitest-resolver` to bundle-first. It now
re-resolves the `vitest` / `@vitest/*` family through Vite's OWN resolver
rooted at vite-plus's location and then the bundled vitest's location, before
the project. Routing through `this.resolve` (not raw `require.resolve`) honours
ESM export conditions — `require.resolve('vitest')` would pick Vitest's
`require`-condition CJS throw-stub (`index.cjs` / `config.cjs`). Two bundled
anchors are needed: `@vitest/browser*` are direct deps of vite-plus while the
nested `@vitest/*` family are deps of vitest (pnpm isolated layout). The
project root stays as the last resort.
For the common migrated layout (project depends only on vite-plus) nothing in
the family resolves from the project root anyway, so this is identical there;
it only changes the project-keeps-its-own-vitest case, which is exactly the case
to pin. Explicit project `resolve.alias`/`resolve.dedupe` on the family (not a
Vitest 4.x default) is respected as an opt-out; coverage providers, not shipped
with vite-plus, follow project resolution (Vitest version-checks them).
Adds resolve-test.spec.ts (runner is bundled, never the project copy) and
bundle-first/CJS-stub regression coverage in define-config-vitest-resolver.spec.ts.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…rovider pins
`dropRemovePackageOverrideKeys` only inspected TOP-LEVEL override keys, so an
npm/bun nested override like `{ "vite-plus": { "@vitest/browser-playwright":
"4.0.0" } }` left the nested provider pin in place. After migration strips the
provider packages from the user manifest, vite-plus owns them as direct deps, so
the surviving nested pin forces vite-plus's provider to an incompatible version.
Make the drop parent-aware. npm/bun nested overrides are SCOPED, so a provider
pin is dropped only where it affects vite-plus's provider ownership — at the top
level (a global pin reaches vite-plus's bundled copy) and inside a `vite-plus`
subtree. Under any other parent the pin targets that parent's own dependency
graph and is the user's to keep; we descend there only to reach a nested
`vite-plus`. A long-form provider override drops its own version pin (`.`) while
keeping unrelated children, and a parent we empty is pruned.
Two object-safety fixes on the same nested-override path:
- `isLegacyWrapperSpec` accepted `string | undefined` and called `.startsWith`;
`pruneLegacyWrapperAliases(pkg.overrides)` runs before the drop, so a nested
object value crashed. Broaden to `unknown` and return false for non-strings
(identical behavior for strings/undefined).
- The bun catalog rewrite passed each managed override value into
`getCatalogDependencySpec` (which calls `.startsWith`); a nested object under a
managed key (e.g. `{ "vitest": { "@vitest/runner": "4.0.0" } }`) crashed and
would clobber the user's nested override. Skip non-string values.
Adds 10 tests: vite-plus-scoped drop + prune; unrelated-parent preservation;
non-provider sibling/child preservation; top-level global + long-form `.` drop;
deep vite-plus-scoped drop; deep unrelated preservation; pre-existing empty `{}`
untouched; and a bun monorepo nested-managed-override no-crash regression.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c584b7240a
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
`vp check` (oxfmt) flagged formatting in the two new spec files added by the
prior two commits; apply the formatter (line-wrapping only, no behavior change).
Also correct an imprecise comment in define-config.ts: only the bare `vitest`
root's `require`-condition entry (`index.cjs`) is a throw-stub. `vitest/config`'s
`require` target (`config.cjs`) is a real CJS build — `require('vitest/config')`
does NOT throw, so the CJS `vite-plus` entry (`index.cts`) and migrated
`require('vite-plus')` configs work. The resolver still prefers the ESM entry for
the test server's module graph; the rationale is ESM-vs-CJS, not a throw-stub,
for subpaths like `vitest/config`.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c136966811
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
unicorn(consistent-function-scoping): the helper captures nothing from its parent scope, so the nested declaration recreated it on every call and failed `vp check` in the CLI E2E job on all platforms. Move it beside the other module-scope helpers; behavior is unchanged. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…'s own dep
The override cleanup peeled selector keys to their target package, so a
provider pin scoped under an unrelated parent (`some-app>@vitest/browser-*`
in pnpm, `some-parent/@vitest/browser-*` in yarn, or the npm/bun nested
`{ some-pkg: { @vitest/...: x } }`) was deleted even though it never
touches vite-plus's own (now direct-dep) provider — silently changing that
parent's transitive resolution.
Decide reachability by matching the full parent chain against the protected
edge `<root> -> vite-plus -> @vitest/provider` (a single `vite-plus`
ancestor): `extractOverrideParentSegments` returns the ordered chain and
`parentChainReachesVitePlus` drops `**` wildcards then requires at most one
concrete ancestor that glob-matches `vite-plus`. Version/range wildcards
(`parent@*`) are stripped before glob classification, and the nested
npm/bun traversal threads the accumulated ancestor chain so it matches the
flat-selector semantics. Global pins, `vite-plus`-parent selectors, and
globs that match `vite-plus` (`**`, `vite-*`) are still dropped.
A selector whose parent is the project's own package name is preserved as a
documented, accepted edge (see comment) per the posted PR review.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: cbc007fb90
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
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
2 participants
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.
Summary
Deletes the bundled
@voidzero-dev/vite-plus-testwrapper and consumes upstreamvitest@4.1.5(plus@vitest/browser*) directly. The vite redirection role that drove the wrapper is now handled cleanly by package-manager overrides (vite→@voidzero-dev/vite-plus-core), so the bundle was dead weight that lagged upstream releases.Public API contract preserved:
vite-plus/test*IS the public test API — existing user code (import { vi } from 'vite-plus/test', etc.) is NEVER rewritten.vitestor@vitest/*separately; they come in transitively as direct deps ofvite-plus.vp migrateon an upstream-vitest project still forward-migratesvitest,vitest/*,@vitest/browser*, declare-module specifiers, and/// <reference types>directives to thevite-plus/test*surface (one-time transition).Notable changes:
packages/cli/build.ts:syncTestPackageExportsauto-generates./test/*shims from upstreamvitest's exports map, plus./test/<provider>and./test/browser/providers/<short>shims projected from each@vitest/browser-*package's exports.packages/cli/package.json: adds@vitest/browser,@vitest/browser-playwright,@vitest/browser-preview,@vitest/browser-webdriverioas direct catalog deps pinned to4.1.5.crates/vite_global_cli/src/commands/version.rs: vitest ToolSpec points at thevitestpackage directly.packages/cli/src/resolve-test.ts: resolvesvitest/package.jsonand readsbin.vitestsovp testinvokes upstream vitest.packages/cli/src/utils/constants.ts: dropsvitestfromVITE_PLUS_OVERRIDE_PACKAGES; onlyviteremains a managed key.packages/cli/src/migration/migrator.ts:isVitestAdjacentflag that flipsneedVitePlus = truefor projects with packages likevitest-browser-svelteeven when there's no vite/oxlint/tsdown to migrate.pruneLegacyWrapperAliases/pruneYamlMapLegacyWrapperAliasessweeps that rewrite stalevitest: npm:@voidzero-dev/vite-plus-test@*aliases to^4.1.5(so existingcatalog:refs keep resolving) and drop other stale wrapper-targeted keys.packages/cli/src/migration/bin.ts: adds ahandleInstallResulthelper so failed reinstalls warn the user, append toreport.warnings, and flipprocess.exitCodeinstead of being silently reported as success.User-visible behavior changes
vp test -hand live test runs now show vitest's native banner (vitest/<semver>,RUN v<semver> <cwd>) instead of the wrapper-rebranded output (vp test/<semver>,RUN <cwd>). This is the tradeoff for delegating directly to upstream vitest without a wrapper layer.Test plan
cargo test -p vite_migration --lib: 167 tests passpnpm exec vitest run(packages/cli): 374 tests passpnpm bootstrap-clisucceedspnpm -F vite-plus snap-test-global+snap-test-local: all fixtures regenerated; diffs only reflect expected behavior (forward import rewrites,@vitest/browser*removed from user devDeps,playwright/webdriveriopreserved as peers, stalevite-plus-testcatalog aliases normalized to^4.1.5).vp teston a fixture usingimport { vi } from 'vite-plus/test';vi.mock(...). See "Follow-up" below.vp migrateon a fresh upstream-vitest project.pnpm installclean: zero@voidzero-dev/vite-plus-testreferences in the lockfile, browser-provider packages installed transitively viavite-plus.Follow-up
@vitest/mockerhoistsvi.mock(...)calls only when the import source is the literal string'vitest'(@vitest/mocker@4.1.5/dist/chunk-hoistMocks.jshardcodeshoistedModule = "vitest"). User code that importsvifrom'vite-plus/test'won't get its mocks hoisted, which silently breaks mocking. The plan is to contribute upstream ahoistedModule?: string | string[]option tohoistMocks()so vite-plus can pass['vitest', 'vite-plus/test']. Tracked as a separate effort — this PR ships the wrapper removal independently; the mocker fix can land later as a@vitest/mockerpoint release.🤖 Generated with Claude Code
Note
Medium Risk
Changes affect daily dep upgrades, release/preview publish lists, and e2e/create smoke installs where Vitest peer resolution (especially Bun) must stay correct.
Overview
This diff wires upstream Vitest into release automation and CI after dropping
@voidzero-dev/vite-plus-test, and updates user-facing docs to only overridevite(notvitest).Dependency bumps (
upgrade-deps.ts) now keepvitest, all pinned@vitest/*catalog entries,VITEST_VERSION,test-vp-createoverride pins, and the@vitest/mockerpnpm patch key/file in sync on daily upgrades—replacing the oldpackages/testbump path.CI packaging stops packing/publishing the test wrapper:
e2e-test,test-vp-create,release,prepare_release, andpkg.pr.newonly handle core + cli. Pack jobs pin0.0.0tgz names and addrepack-vite-tgzso Bun can satisfy Vitest’speer vitevia a masqueradedvite-7.99.0.tgz;test-vp-createpins Vitest packages by semver instead of a test-wrapper tgz.vp --versionreports Vitest from the realvitestpackage. Migration walks.cjs/.ctsfor legacyrequire('vite-plus/test/...')rewrites. README manual migration dropsvitestoverride examples.Reviewed by Cursor Bugbot for commit cbc007f. Bugbot is set up for automated code reviews on this repo. Configure here.