[1egoman]

Stacked on #2167. Follow-up to this review thread.

#2167 unregisters an incoming data stream when its source ReadableStream is cancelled. But when a consumer stops reading early, TextStreamReader and ByteStreamReader only call releaseLock(), which doesn't cancel the stream. A consumer stops early by aborting the signal passed to readAll/withAbortSignal, by leaving a for await with break/return, or when the reader throws one of its own errors (for example, more bytes than totalLength). The stream then stays registered, and later chunks pile up in a queue nobody reads until the trailer arrives or the sender disconnects. On uncompressed streams nothing caps that buffering.

This PR makes the iterator's early-exit cleanup call reader.cancel(reason) before releasing the lock. The cancel travels up the pipe chain to the source stream, and the cancel hook from #2167 removes the stream from textStreamControllers/byteStreamControllers. Later packets for that stream are dropped as if it had never been opened. Streams that finish normally never run this cleanup, so they aren't affected. This also matches how the built-in ReadableStream async iterator behaves on early exit.

Behavior change: before this change, each [Symbol.asyncIterator]() call took a fresh lock, so a consumer could break out of a for await and resume the same stream later. That is no longer possible: once a reader stops early, the stream is cancelled. This wasn't documented, and I don't expect anyone to rely on it, but the changeset calls it out.

Testing

• New regression test, modeled on the Incoming data stream stays registered after a missing-chunk error, so later chunks and the trailer throw TypeErrors #2166 test: it aborts a text stream's readAll partway through, then checks that a late chunk is dropped and that a header reusing the stream ID opens a fresh stream, before any trailer for the old stream has arrived. It fails without the StreamReader change.
• vitest: all tests pass except smoke-tests/smoke.spec.ts, which fails with a missing @playwright/test import and fails the same way on the base branch. tsc --noEmit and eslint are clean.

🤖 Generated with Claude Code

[changeset-bot]

🦋 Changeset detected

Latest commit: 1c2e4cf

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
livekit-client	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.

[github-actions]

size-limit report 📦

Path	Size
dist/livekit-client.esm.mjs	115.83 KB (+0.05% 🔺)
dist/livekit-client.umd.js	125.12 KB (-0.02% 🔽)
dist/core/core.mjs	78.55 KB (0%)