fix zlib throwing memory limit exceeded

[anonrig]

Fixes #6286

[ask-bonk]

The change looks correct. Let me do a final review of all the pieces.

Summary of my analysis:

1. Root cause: The do...while loop in syncProcessBuffer continues when avail_out == 0. On the next iteration, addChunk() → grow() fails because the buffer is already at maxCapacity. This is an off-by-one: the output fit exactly but the loop tried to grow anyway.

2. Fix: Insert a check after adjustUnused() — if avail_out == 0 and the buffer is at max capacity, check whether the stream is finished. If yes, the data fit exactly, so break. If no, the output exceeds the limit, so throw.

3. isStreamEnd() implementations: Each compression context correctly implements isStreamEnd():

   • ZlibContext: err == Z_STREAM_END — standard zlib end signal
   • BrotliEncoderContext: tracks streamEnd flag set from BrotliEncoderIsFinished()
   • BrotliDecoderContext: lastResult == BROTLI_DECODER_RESULT_SUCCESS
   • ZstdEncoderContext: !ZSTD_isError(lastResult) && lastResult == 0
   • ZstdDecoderContext: !ZSTD_isError(lastResult) && lastResult == 0

4. No interaction issues with getError(): The getError() check runs before the new check. For all contexts, when the stream is truly done, getError() returns kj::none, so the flow correctly reaches the isStreamEnd() check.

5. Test coverage: Covers inflateRaw, inflate, gunzip, brotliDecompress, and the too-small case. No zstd test though — but zstd is newer and less commonly used with maxOutputLength.

6. No backward compat concern: This only fixes a case that previously threw an error incorrectly. No behavior is removed; a previously-broken path now works. No compat flag needed.

The code is clean, correct, and well-tested. There are no actionable issues.

LGTM

github run