projects does not work with browser mode

[christophe-g]

Describe the bug

running vitest in a monorepo, fails to work (stalls - noting happens) when packages use browser mode.

This is the exact same behavior as described here: #9164

Happening on 5.0.0-beta.1 as well as on latest v4.

When running non headless vitest page display 'no test found'

Test run in non-browser environment (happy-dom).

Reproduction

tried simplified repro without success.

However, I suspect version conflict between root and package. Below is the output trace after a Ctrl+c.

Vitest caught 1 unhandled error during the test run.
This might cause false positive tests. Resolve unhandled errors to make sure your tests are not affected.

⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ Unhandled Error ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯
Error: Failed to run the test /home/christophe/Programming/preignition/accessibleData/app/app-survey/renderer/renderer.test.ts.
 ❯ node_modules/.pnpm/vitest@5.0.0-beta.1_@opentelemetry+api@1.9.1_@types+node@25.6.0_@vitest+browser-playwri_53592e796b1c9a9e525229a5f01a8ec0/node_modules/vitest/dist/chunks/cli-api.MLtHkHGf.js:2627:17
 ❯ processTicksAndRejections node:internal/process/task_queues:105:5

Caused by: Error: [vitest] Browser connection was closed while running tests. Was the page closed unexpectedly?
 ❯ WebSocket.<anonymous> node_modules/.pnpm/@vitest+browser@5.0.0-beta.1_vite@8.0.10_@types+node@25.6.0_esbuild@0.27.7_terser@5.46._83b033ee639ed61e333e1cbaf5290842/node_modules/@vitest/browser/dist/index.js:3055:16
 ❯ WebSocket.emit node:events:519:35
 ❯ WebSocket.emitClose node_modules/.pnpm/ws@8.20.0/node_modules/ws/lib/websocket.js:273:10
 ❯ Receiver.receiverOnFinish node_modules/.pnpm/ws@8.20.0/node_modules/ws/lib/websocket.js:1214:20
 ❯ Receiver.emit node:events:507:28
 ❯ finish node:internal/streams/writable:953:10
 ❯ node:internal/streams/writable:934:13
 ❯ processTicksAndRejections node:internal/process/task_queues:90:21

Caused by: Error: [birpc] rpc is closed, cannot call "createTesters"
 ❯ node_modules/.pnpm/@vitest+browser@5.0.0-beta.1_vite@8.0.10_@types+node@25.6.0_esbuild@0.27.7_terser@5.46._83b033ee639ed61e333e1cbaf5290842/node_modules/@vitest/browser/dist/index.js:2917:34
 ❯ Proxy.$close node_modules/.pnpm/@vitest+browser@5.0.0-beta.1_vite@8.0.10_@types+node@25.6.0_esbuild@0.27.7_terser@5.46._83b033ee639ed61e333e1cbaf5290842/node_modules/@vitest/browser/dist/index.js:2916:18
 ❯ WebSocket.<anonymous> node_modules/.pnpm/@vitest+browser@5.0.0-beta.1_vite@8.0.10_@types+node@25.6.0_esbuild@0.27.7_terser@5.46._83b033ee639ed61e333e1cbaf5290842/node_modules/@vitest/browser/dist/index.js:3055:9
 ❯ WebSocket.emit node:events:519:35
 ❯ WebSocket.emitClose node_modules/.pnpm/ws@8.20.0/node_modules/ws/lib/websocket.js:273:10
 ❯ Receiver.receiverOnFinish node_modules/.pnpm/ws@8.20.0/node_modules/ws/lib/websocket.js:1214:20
 ❯ Receiver.emit node:events:507:28
 ❯ finish node:internal/streams/writable:953:10
 ❯ node:internal/streams/writable:934:13

System Info

System:
    OS: Linux 7.0 Linux Mint 22.3 (Zena)
    CPU: (16) x64 AMD Ryzen 7 7840U w/ Radeon  780M Graphics
    Memory: 20.41 GB / 30.65 GB
    Container: Yes
    Shell: 5.2.21 - /bin/bash
  Binaries:
    Node: 23.6.1 - /home/christophe/.nvm/versions/node/v23.6.1/bin/node
    npm: 11.2.0 - /home/christophe/.nvm/versions/node/v23.6.1/bin/npm
    pnpm: 11.0.3 - /home/christophe/.nvm/versions/node/v23.6.1/bin/pnpm
  Browsers:
    Chrome: 147.0.7727.137
    Firefox: 150.0.1
    Firefox Developer Edition: 150.0.1
  npmPackages:
    @vitest/browser: catalog: => 5.0.0-beta.1 
    @vitest/browser-playwright: catalog: => 5.0.0-beta.1 
    @vitest/browser-preview: catalog: => 5.0.0-beta.1 
    playwright: catalog: => 1.59.1 
    vite: catalog: => 8.0.10 
    vitest: catalog: => 5.0.0-beta.1

Used Package Manager

npm

Validations

• Follow our Code of Conduct
• Read the Contributing Guidelines.
• Read the docs.
• Check that there isn't already an issue that reports the same bug to avoid creating a duplicate.
• Check that this is a concrete bug. For Q&A open a GitHub Discussion or join our Discord Chat Server.
• The provided reproduction is a minimal reproducible example of the bug.

[github-actions]

Hello @christophe-g. Please provide a minimal reproduction using a GitHub repository or StackBlitz (you can also use examples). Issues marked with needs reproduction will be closed if they have no activity within 3 days.

[dokson]

@christophe-g your suspicion about the version conflict was right. I confirmed the duplication on the related repro from #9164: pnpm creates two physical .pnpm/ snapshots of the same vitest version when a peer dep (in your case likely @types/node, in the linked repro @vitest/browser-playwright) resolves to different versions across packages. Node loads each as a distinct module, the browser orchestrator state is registered in copy A but the project's vitest.config.ts resolves copy B, and the runner hangs waiting for a connection that lives on the other side.

Opened #10262 with a targeted detection: the browser server now resolves vitest and @vitest/browser from each project root at startup, compares to the running paths, and throws an actionable error (with pnpm why vitest + the recommended hoist-to-root fix) instead of letting the run hang silently. Also adds a Duplicate Vitest Installation section in docs/guide/common-errors.md.

[christophe-g]

@dokson - that's great, thank you !

I also tried deduping most of what I could (` resolve: { dedupe: [ 'vite', 'vitest', '@vitest/...']}), without much success on my end.

[AriPerkkio]

@dokson please post that reproduction here. 🙏

[dokson]

@AriPerkkio sure — the reproduction is @Andrei486's vitest-project-repro, which is the same repro from #9164. It's flaky as @sheremet-va noted (pnpm's deduplication on macOS/Node 24 sometimes collapses the two snapshots into one), but the underlying duplication is deterministic and verifiable from the lockfile alone.

Lockfile evidence (from a fresh pnpm install --recursive on that repo):

$ grep 'vitest@4' pnpm-lock.yaml
vitest@4.0.15:                                       ← snapshot A (root)
vitest@4.0.15(@vitest/browser-playwright@4.0.15):    ← snapshot B (packages/browser)

Same version, two distinct .pnpm/ snapshots because the root has no peer-dep hash on vitest while the sub-package's transitive resolution adds the @vitest/browser-playwright peer.

Runtime evidence (Node createRequire from each project root, Windows):

ROOT             → ...\.pnpm\vitest@4.0.15_@vitest+browser-playwright@4.0.15\node_modules\vitest
packages/browser → ...\.pnpm\vitest@4.0.15\node_modules\vitest                              ← different
packages/normal  → ...\.pnpm\vitest@4.0.15\node_modules\vitest

The root runner uses snapshot A and registers the browser orchestrator state there; the sub-project's vitest.config.ts resolves snapshot B when its plugins import @vitest/browser/vitest, so the two never share state and the run hangs.

Why @christophe-g hits the same on v5 beta (#10261): same shape, but the diverging peer is @types/node (root: ^24.12.0, sub-package transitively: 25.x), so the two snapshots are vitest@5.0.0-beta.1_@types+node@24.12.0_... vs vitest@5.0.0-beta.1_@types+node@25.3.5_....

#10262 doesn't fix the duplication itself (that needs either a single-resolution policy across projects or moving the orchestrator out-of-process), but it detects it at startup with createRequire from each project.config.root and throws an actionable error pointing to pnpm why vitest and the hoist-to-root fix instead of letting the run hang silently.

[AriPerkkio]

@dokson are you actually able to reproduce this with https://github.com/Andrei486/vitest-project-repro? What steps did you run? I'm seeing tests pass constantly with the repro steps. 🤔

[dokson]

git clone https://github.com/Andrei486/vitest-project-repro
cd vitest-project-repro
sed -i 's|"vitest": "latest"|"vitest": "5.0.0-beta.1"|; s|"@vitest/browser-playwright": "latest"|"@vitest/browser-playwright": "5.0.0-beta.1"|' package.json packages/*/package.json
pnpm install:all
pnpm exec playwright install
pnpm test

Hangs deterministically (8/8 timeouts at 90 s on Windows 11 / Node 25 / pnpm 10.32) with the stack from #10261:

[vitest] Browser connection was closed while running tests. Was the page closed unexpectedly?
[birpc] rpc is closed, cannot call "createTesters"

Lockfile shows the duplicate snapshots:

vitest@5.0.0-beta.1:                                                          ← copy A (root, no peers in hash)
vitest@5.0.0-beta.1(@vitest/browser-playwright@5.0.0-beta.1)(vite@8.0.10):    ← copy B (packages/browser)

[anilanar]

It would be great to have a solution that works even with physical duplication of modules, namely if there were no global/module level state.