fix(cli): handle stdout backpressure#2169
Conversation
✅ Deploy Preview for viteplus-preview canceled.
|
|
CI update: The Linux/macOS failures are unrelated snapshot drift: a run-day install date (already addressed by #2150) and newer package versions resolved from the registry. This PR does not touch those paths, so I have not included unrelated snapshot updates here. |
|
@codex review |
|
Codex Review: Didn't find any major issues. Breezy! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
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". |
cpojer
requested review from
jong-kyung and
wan9chi
and removed request for
cpojer
|
Thank you for tracking this down and putting together a fix! This PR correctly identifies the underlying issue. However the same problem can affect any code that writes to stdout or stderr in the Vite+ process, including |
Fixes #2165. ## Root cause Node marks non-TTY stdio as non-blocking, and the flag is shared with inherited child descriptors. Rust's standard I/O and embedded dependencies assume blocking writes, so a full consumer pipe can return `EAGAIN`, truncate diagnostics, or panic. ## Why this approach #2169 retries writes in code we control. Dependencies like vite-task write directly to the same stdio, so those retries cannot protect them. Clearing `O_NONBLOCK` at startup covers every writer.
Fixes #2165. ## Root cause Node marks non-TTY stdio as non-blocking, and the flag is shared with inherited child descriptors. Rust's standard I/O and embedded dependencies assume blocking writes, so a full consumer pipe can return `EAGAIN`, truncate diagnostics, or panic. ## Why this approach #2169 retries writes in code we control. Dependencies like vite-task write directly to the same stdio, so those retries cannot protect them. Clearing `O_NONBLOCK` at startup covers every writer.
4 participants
Summary
poll(POLLOUT)and propagate non-retryable I/O errorsvp checkoutput helpers return write failures instead of discarding them or panickingWouldBlock, interrupted writes, terminal errors, and a real non-blocking Unix streamRoot cause
vp checkcaptures Oxlint diagnostics and replays them to stdout. When a consumer could not drain a non-blocking stdout pipe quickly enough,write_allreturnedWouldBlock. The diagnostic replay path discarded that error, so the diagnostics were truncated and a later standard-library print could panic while stdout was still unavailable.The new write loop preserves the unwritten suffix, waits until the stream is writable, and retries. Permanent failures are returned through the existing command result path.
Fixes #2165.
Validation
cargo test -p vite_shared waits_until_a_nonblocking_writer_is_ready_before_retryingjust checkjust lintjust testpnpm -F vite-plus buildjust snapshot-test check_(36 passed)