Skip to content

Liquid Glass and Apple Music touches for the Spotify iOS app. A GPL-3.0 fork of spoti.pw 0.21.1.

C
5.1 KB
Raw Blame

Native lossless decoder probe

An isolated simulator app using generated sine waves. It does not load Spotify, contact providers, play audible audio, or change production playback. Requires Xcode with an arm64 iOS simulator, FFmpeg and Python 3.9+. Build products default to the ignored build/ directory; override LOSSLESS_PROBE_OUT to keep them elsewhere.

sh harness/lossless/build.sh
python3 harness/lossless/serve.py harness/lossless/build/LosslessProbe.app

In another terminal, with the intended simulator booted:

xcrun simctl install booted harness/lossless/build/LosslessProbe.app
SIMCTL_CHILD_LOSSLESS_HTTP_BASE=http://localhost:18764 \
  xcrun simctl launch --console booted dev.vitrine.lossless-probe > /tmp/lossless-probe.log 2>&1
python3 harness/lossless/check.py /tmp/lossless-probe.log

Omit SIMCTL_CHILD_LOSSLESS_HTTP_BASE for local tests only. The server binds loopback and serves only four generated fixture names, supporting HEAD, byte ranges, 206 and 416. Stop it with Ctrl-C afterward. The app exits after its probes, with a 60-second watchdog. Check the log with check.py: exiting successfully only means the probe finished, not that all APIs supported the media.

What is exercised

  • 12-second stereo FLAC at 16-bit/44.1 kHz and 24-bit/96 kHz; ALAC at 24-bit/48 kHz in M4A with the moov box at the end and an otherwise identical fast-start copy with it at the beginning.
  • AudioFileStream in 257-byte input chunks. Collected packets and the magic cookie then go through AudioConverter to Float32 PCM. This intentionally keeps all packets in memory: it is not a bounded streaming implementation, network-underflow test or proof of seeking compressed streams.
  • ExtAudioFile whole-file decode at native rate and 44.1 kHz, exact total-frame checks, and a seek to file frame floor(3.125 * sourceRate). The next 1,024 native-rate frames are compared to the known sine waves; converted seeks check successful reads, not waveform alignment or converter latency.
  • AVAssetReader over local and loopback HTTP assets, requesting one second from 3.125 seconds. Native-rate timestamp/frame bounds are checked locally. HTTP behavior is reported, not required to fail on every OS.

Observed on 2026-10-08

iOS 27.0 arm64 simulator, SDK 27.0, minimum deployment target 16.1. Not run on iOS 16.1 or 26, or on a physical phone. The deployment target alone does not establish those runtimes' capabilities. Raw structured output: results/ios27-simulator.jsonl.

Fixture Local native decode Seek max absolute error Incremental parser / packet decode
CD FLAC 529,200 frames 0.00001525 Parser succeeds; packet path gives 528,384 frames, 816 short
96 kHz FLAC 1,152,000 frames 0.000000119 Parser succeeds; packet path gives 1,150,976 frames, 1,024 short
ALAC, tail metadata 576,000 frames 0.000000119 Parser refuses with optm / 1869640813 (not optimized)
ALAC, fast start 576,000 frames 0.000000119 Parser and packet decode return all 576,000 frames

All four full-file conversions to 44.1 kHz returned 529,200 frames. All four local asset-reader seeks completed within one sample of the requested start/duration. All four HTTP assets loaded tracks but startReading returned false with AVFoundationErrorDomain -11838 (AVErrorOperationNotSupportedForAsset); the fixture server recorded successful 206 responses.

The FLAC packet-path tail deficit is unresolved; do not silently pad it or call streaming complete. The source ASBD reported zero bits per channel for these compressed formats. That value cannot be used as a verified bit-depth badge. Fixture bit depth is known from generation; future source validation needs codec/container metadata and decode inspection, separate from Float32 output precision.

These results support a complete temporary-file decoder as the first controlled playback experiment. They do not prove Spotify clock continuity, stage composition, seamless switching, network streaming, low memory use, background playback, hi-res output, or any provider's compatibility. See the integration investigation for those remaining steps.

Audio handoff prototype

handoff-build.sh builds a second app from the worktree's production SGPreparedPCM, SGPCMReplacement and SpeedPitch.x. It replaces generated native PCM before a stand-in stage, using a real converter/mixer/RemoteIO chain. It emits quiet audio. See the prototype's calling contract and tests.

THEOS="$HOME/theos" sh harness/lossless/handoff-build.sh
xcrun simctl install booted harness/lossless/build/handoff/LosslessHandoff.app
xcrun simctl launch --console booted dev.vitrine.lossless-handoff > /tmp/lossless-handoff.log 2>&1

Require a final RESULT failures=0, not just successful launch. The app has a 30-second watchdog. The default decoder probe above and its check.py remain separate; they use a different log format. The handoff checks do not run actual Spotify or Sing's ML/look-ahead engine. The production source selector has no automatic registration or settings switch.