vp check panics on EAGAIN while replaying large lint diagnostics

[hyoban]

Describe the bug

vp check panics while replaying a large captured Oxlint diagnostic block when stdout temporarily returns EAGAIN / WouldBlock.

Expected behavior: vp check should handle stdout backpressure without panicking or truncating diagnostics. If output cannot be written, it should retry appropriately or return a regular I/O error.

Actual behavior: the process prints the warning heading, invokes the Vite+ panic hook, drains only part of the buffered diagnostics, and exits with code 129. The following pnpm lint:eslint command is never executed.

This occurs in a public GitHub Actions job, so it is not specific to a local terminal or Codex output pipe.

Reproduction

https://github.com/langgenius/dify/actions/runs/29259173815/job/86847662089

The failing source revision is:

langgenius/dify@00196da

Steps to reproduce

1. Check out the failing revision:

   git clone https://github.com/langgenius/dify.git
   cd dify
   git checkout 00196da9c166c2191fd33937afe3b01369a893c1

2. Install Vite+ 0.2.4 dependencies and run the repository check under GitHub Actions or another stdout consumer that can apply non-blocking backpressure:

   vp install
   pnpm -w check

3. The package script runs:

   vp check && pnpm lint:eslint

4. Formatting succeeds for 8,557 files. After linting finishes, vp check starts replaying the captured diagnostics and panics with os error 11.

The trigger is the large diagnostic set: this revision produces approximately 2,426 warnings.

A follow-up commit replaced vp check with vp fmt --check && vp lint --quiet:

langgenius/dify@0178010

The corresponding job succeeds on the same runner type and reports all 2,426 warnings and 0 errors:

https://github.com/langgenius/dify/actions/runs/29259903686/job/86850424110

Locally, with the same project and tool versions:

• vp check reproducibly panics when writing to a backpressured output pipe.
• vp lint successfully emits the same 2,427 diagnostics.
• vp check >/dev/null completes successfully.

This differential points to the vp check capture/analyze/replay path rather than Oxlint itself.

System Info

$ vp env current
Environment:
  Version       22.23.1
  Source        devEngines.runtime
  Source Path   /path/to/dify/package.json
  Project Root  /path/to/dify

Tool Paths:
  node  ~/.vite-plus/js_runtime/node/22.23.1/bin/node
  npm   ~/.vite-plus/js_runtime/node/22.23.1/bin/npm
  npx   ~/.vite-plus/js_runtime/node/22.23.1/bin/npx

Package Manager:
  Name          pnpm
  Version       11.10.0
  Source        packageManager
  Source Path   /path/to/dify/package.json
  Project Root  /path/to/dify
  Bin Path      ~/.vite-plus/package_manager/pnpm/11.10.0/pnpm/bin/pnpm

$ vp --version
vp v0.2.4

Local vite-plus:
  vite-plus  v0.2.4

Tools:
  vite             v8.1.3
  rolldown         v1.1.4
  vitest           v4.1.10
  oxfmt            v0.57.0
  oxlint           v1.72.0
  oxlint-tsgolint  v0.24.0
  tsdown           v0.22.3

Environment:
  Package manager  pnpm v11.10.0
  Node.js          v22.23.1 (devEngines.runtime)

The public CI reproduction runs on depot-ubuntu-24.04-4. A local reproduction also occurs on macOS, where the equivalent error is os error 35.

Used Package Manager

pnpm

Logs

$ vp check && pnpm lint:eslint
pass: All 8557 files are correctly formatted (4122ms, 4 threads)
warn: Lint or type warnings found
Vite+ panicked. This is a bug in Vite+, not your code.

thread '<unnamed>' (3117) panicked at library/std/src/io/stdio.rs:1166:9:
failed printing to stdout: Resource temporarily unavailable (os error 11)
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

[diagnostics continue briefly, then truncate]

ELIFECYCLE Command failed with exit code 129.

Initial analysis

In Vite+ 0.2.4, vp check forces Oxlint's default reporter and captures its output before analyzing and replaying it:

https://github.com/voidzero-dev/vite-plus/blob/v0.2.4/packages/cli/binding/src/check/mod.rs#L156-L180

The failure path then calls print_stdout_block followed by print_summary_line:

https://github.com/voidzero-dev/vite-plus/blob/v0.2.4/packages/cli/binding/src/check/mod.rs#L199-L213

print_stdout_block currently discards both write_all results:

https://github.com/voidzero-dev/vite-plus/blob/v0.2.4/packages/cli/binding/src/check/analysis.rs#L130-L142

let _ = stdout.write_all(trimmed.as_bytes());
let _ = stdout.write_all(b"\n");

The same code is still present on main at the time of filing. A likely sequence is:

1. Replaying the large captured diagnostic block fills the non-blocking stdout pipe.
2. write_all returns WouldBlock, but the error is discarded.
3. The following summary/output helper prints while stdout is still unavailable and panics through Rust's standard printing path.

This is related to #1285 because that report contains the same stdout panic signature. However, #1285 addressed the misleading Rolldown panic hook attribution. The underlying EAGAIN handling and diagnostic truncation remain reproducible in Vite+ 0.2.4 and on the current main implementation.

Validations

• Read the Contributing Guidelines.
• Checked that there is not already an issue tracking the underlying vp check stdout/backpressure bug.
• Confirmed this is a Vite+ issue rather than an upstream Vite, Vitest, tsdown, Rolldown, or Oxc issue through the vp check versus vp lint differential.
• Provided a deterministic, publicly runnable reproduction. The large diagnostic set is part of the trigger.

[liangmiQwQ]

Hello @hyoban, thanks for the reproduction! Would you be interested in working on a fix?