Context
The recent PR backlog sweep landed the low-risk hot-path wins on master and closed the overlapping PR queue. A few ideas from that sweep are still worth preserving, but they need deliberate follow-up rather than direct PR merges.
Keep In Mind
- Re-evaluate zero-allocation polling APIs for
AudioEngine / AudioSegmentProcessor.
The idea is valid, but any future attempt must preserve current ownership/encapsulation guarantees for getStats() and getStateInfo() and avoid leaking nested mutable state through reused output objects.
- Revisit bar-history storage independently from return-buffer reuse semantics.
The circular/shadow-buffer idea may still be useful, but it should be evaluated separately from API changes that reuse the same returned Float32Array reference across UI updates.
- Only pursue deeper
UtteranceBasedMerger object-copy reductions with profiling.
The safe normalizeWords() and pendingEnd cleanup already landed. Further changes to finalized/immature word copying should be guided by real profiling because they touch correctness-sensitive transcript state.
- Keep callback subscription semantics explicit.
The Set-based unsubscribe PRs were closed because they silently changed duplicate-subscription and mid-dispatch behavior. If we ever optimize subscriber removal, define and test the callback contract first.
Not A Goal
- Reopening the old bot PR variants.
- Landing API-semantics changes as backlog cleanup without focused tests.
Reference
Landed sweep commit: 3a46116
Context
The recent PR backlog sweep landed the low-risk hot-path wins on
masterand closed the overlapping PR queue. A few ideas from that sweep are still worth preserving, but they need deliberate follow-up rather than direct PR merges.Keep In Mind
AudioEngine/AudioSegmentProcessor.The idea is valid, but any future attempt must preserve current ownership/encapsulation guarantees for
getStats()andgetStateInfo()and avoid leaking nested mutable state through reused output objects.The circular/shadow-buffer idea may still be useful, but it should be evaluated separately from API changes that reuse the same returned
Float32Arrayreference across UI updates.UtteranceBasedMergerobject-copy reductions with profiling.The safe
normalizeWords()andpendingEndcleanup already landed. Further changes to finalized/immature word copying should be guided by real profiling because they touch correctness-sensitive transcript state.The
Set-based unsubscribe PRs were closed because they silently changed duplicate-subscription and mid-dispatch behavior. If we ever optimize subscriber removal, define and test the callback contract first.Not A Goal
Reference
Landed sweep commit:
3a46116