mirror of
https://github.com/navidrome/navidrome.git
synced 2026-10-11 03:47:18 +02:00
Reading the embedded picture out of the WASM module to hash it in Go cost +28% in tag extraction and doubled allocations (93 real MP3s: 32.5ms -> 41.7ms, 91MiB -> 188MiB per pass), because the fork instantiates a module per file and the image read grows its memory and copies the bytes twice. go-taglib now fingerprints each embedded picture inside the module during the existing properties call (length, head, tail and 64 sampled stripes) and exposes it as ImageDesc.Hash. The extractor stores that value, so no picture bytes cross the boundary. Extraction is back to master speed (32.0ms, p=0.24) and the scanner benchmark is unchanged (10.88ms vs 10.89ms). Also adds tests for albums whose only art is an external file: no fingerprint is stored, tracks keep the album id, and hydration skips the album lookup. |
||
|---|---|---|
| .. | ||
| deezer | ||
| gotaglib | ||
| lastfm | ||
| listenbrainz | ||