[cabelo]

Overview

Additional information

Requirements

• I have read and agree with the contributing guidelines
• AI usage disclosure: NO

Version 0.11.26

• Fixed a bug where the ends of sounds are cutoff. This affects ma_sound.
• Fixed a crash when passing in null for the read or seek callbacks for a decoder.
• Fixed an error with node initialization.
• Fixed build for Arm64EC on MSVC.
• Fixed audio glitches during volume smoothing, including SSE2 and multichannel processing.
• Fixed a race when changing sound and sound-group volume during playback.
• Fixed incorrect handling of scheduled node start times.
• Fixed repeated audio and a possible ring-buffer overflow during duplex processing with resampling.
• Fixed swapped input and output channel maps returned by data converter queries.
• Miscellaneous compiler compatibility and warning fixes.
• Resource Manager: Fixed a possible crash when loading fails.
• Resource Manager: Fixed races during asynchronous loading and teardown that could access freed memory.
• PulseAudio: Improvements to format negotiation to better handle cases when PulseAudio reports s24_32 as its native format.
• PulseAudio: Fixed playback startup stalls, device queries after server restarts, and initialization errors involving missing devices and resource leaks.
• AAudio: Fixed a possible crash during rerouting and preserved device settings when reopening streams.
• OpenSL: Fixed draining of duplex devices and handling of buffer-queue query failures.
• Web Audio: Fixed AudioWorklet initialization and unsafe stack cleanup.
• Emscripten: Fixed an error with ALLOW_MEMORY_GROWTH.
• Emscripten: Fixed some issues relating to handling of 64-bit pointers.

[taronaeo]

/bot review

[ggml-gh-bot]

Automated code review

This is a clean, single-purpose vendor bump (miniaudio 0.11.25 -> 0.11.26), so most area checklists reduce to verifying the update is genuine, complete, and non-breaking for the only in-tree consumer.

Verification performed:

• vendor/miniaudio/miniaudio.h at this commit is byte-identical to both the pinned upstream commit in sync_vendor.py (877ea4a...) and the upstream 0.11.26 release tag. No local modifications, no partial application, no unrelated edits riding along.
• The only other change is the one-line URL/hash update in scripts/sync_vendor.py, which is required for the check-vendor CI job (it re-runs the sync script and diffs) to stay green. That will pass as-is.
• The only in-tree consumer is tools/mtmd/mtmd-helper.cpp, which uses the memory-based decoder/encoder API only: ma_decoder_config_init, ma_decoder_init_memory, ma_decoder_get_length_in_pcm_frames, ma_decoder_read_pcm_frames, ma_decoder_uninit, and the ma_encoder_* family. None of these signatures changed in 0.11.26. The changed semantics around ma_decoder_read_proc/ma_decoder_seek_proc (null-callback crash fix, partial-read treated as EOF) only affect custom callbacks, which are not used here.
• Most of the fixed subsystems (AAudio, OpenSL, PulseAudio, Web Audio, Resource Manager, engine/node graph, volume smoothing) are compiled out anyway: mtmd-helper.cpp defines MA_NO_DEVICE_IO, MA_NO_RESOURCE_MANAGER, MA_NO_NODE_GRAPH, MA_NO_ENGINE, MA_NO_GENERATION. The decoder bugfixes (end-of-sound cutoffs, decoder init errors) are the relevant ones here and are upstream fixes, not local risk.
• The changelog in the PR description matches the actual 0.11.26 release notes, and the update includes genuine correctness/race fixes rather than new attack surface. Vendored header stays public-domain/MIT-0, no new dependencies.

Findings:

Blocking: none.

Will slow the review: none.

Nits (optional):

• The pinned commit 877ea4a is content-identical to the 0.11.26 tag, so the tag URL from the commented-out line above could be used instead. The pinned-commit pattern is deliberate (per the existing comment referencing the issue in PR 17179), so keeping it is fine - just noting the two are currently equivalent.
• The PR template's Overview section is empty; one sentence ("routine miniaudio version bump to pick up upstream 0.11.26 fixes") would help reviewers. Minor.

No action needed on the vendored file itself - it should be merged as an unmodified upstream copy, which it is.

This review was generated automatically by pi coding agent using zai-org/GLM-5.3. It may contain mistakes. Maintainers make the final call.