catap 0.5.0

• BREAKING: Replaced streaming callbacks from on_data(data, num_frames) with
  on_buffer(buffer: AudioBuffer).

• Added callback metadata types: AudioBuffer and AudioStreamFormat.

• Added stream_format accessors on AudioRecorder and RecordingSession;
  use fields such as stream_format.sample_rate instead of scalar format
  convenience properties.

• Added a native macOS Core Audio dylib for the recorder IOProc and made it
  required for recording. The old pure-Python IOProc fallback has been removed.

  # Before
  def on_data(data: bytes, num_frames: int) -> None:
      ...
  
  # After
  def on_buffer(buffer: AudioBuffer) -> None:
      data = buffer.data
      num_frames = buffer.frame_count

v0.4.3

Changes

• Tightened README, CLI help, and package metadata around the process-tap support boundary and unsupported capture scenarios.

Validation

• uv run --group dev ruff check .
• uv run --group dev ty check --error-on-warning src tests
• uv run --group dev pytest
• uv run --group dev python -m build
• uv run --group dev twine check dist/*

v0.4.2

• Added CI coverage for free-threaded CPython 3.13t and 3.14t on macOS.
• Replaced the recorder buffer-pool deque with queue.SimpleQueue to avoid relying on CPython deque atomicity in free-threaded builds.
• Documented local free-threaded test commands and the opt-in real-recording smoke check.

v0.4.1

• Added Python 3.11 support by replacing 3.12-only type-alias syntax and broadening package metadata.

v0.4.0

• Added strict tap stream format validation before capture starts. Unsupported
  Core Audio layouts now raise UnsupportedTapFormatError instead of risking
  plausible but corrupt WAV output.
• Reject non-linear PCM, big-endian PCM, non-packed PCM, non-interleaved audio,
  unsigned integer PCM, padded frames, invalid rates/channels/bit depths, and
  floating-point formats other than packed float32.
• Treat malformed callback buffers as capture failures instead of guessing:
  multi-buffer AudioBufferList layouts, missing data pointers, mismatched or
  missing channel counts, and partial-frame byte counts are surfaced on stop.
• Make WAV output transactional. Recorders now write to a sibling temporary file
  and publish it only after clean shutdown, preserving existing output files on
  failed startup or failed writes.
• Stop active capture sessions before destroying Core Audio resources and clear
  started state after stop attempts so cleanup paths are more deterministic.
• Reject non-integer max_pending_buffers values such as True, floats, and
  strings.
• Expanded pytest coverage around Core Audio lifecycle cleanup, recorder format
  handling, malformed callback buffers, output-file safety, public exports, and
  session queue-bound validation.

v0.3.0

• Refactored the recorder into focused internal capture-engine, worker, support,
  and session-backend modules.
• Preserved recorder cleanup failures across stop and close paths so secondary
  teardown errors are not lost.
• Surfaced Core Audio callback failures on recorder stop instead of silently
  swallowing them.
• Added synthetic and live profiling probes for worker throughput, conversion
  cost, callback timing, queue depth, and dropped-buffer behavior.
• Replaced the old private-internals profiling scripts with the new profiling
  harnesses.
• Added performance and real-time notes for the recorder callback, queueing
  model, CPython deque assumptions, and known tradeoffs.
• Tightened README wording and added the current tested macOS hardware/version.
• Consolidated Core Audio binding discovery helpers and expanded related tests.
• Updated locked development dependencies.

Full Changelog: v0.2.0...v0.3.0

v0.2.0

• Initial public release of catap.
• Added device stream discovery so taps can target a specific input/output stream.
• Added discovery of existing process taps and the ability to record from them.
• Added shared-tap support: create, extend, and delete shared taps from the core lab demo.
• Raised clearer errors for stale shared taps and preserved zero-tap device stream metadata.
• Hardened audio recorder concurrency and cleanup lifecycle for free-threaded Python builds.
• Consolidated recorder structs and unified the cleanup cascade.
• Moved helper tone tooling into an internal devtools package and added regression coverage.
• Added worker queue latency profiling and restored synthetic profiler compatibility.
• Expanded the core lab demo with recording playback controls, helper tone device selection, shared-tap workflows, and bench-style chrome.
• Bumped supported Python floor metadata to include 3.14.
• Slimmed the README and split implementation notes into docs/.