v0.2.0

The crate is split in two. nanomp3-core is the frame decoder that used to be nanomp3; nanomp3 re-exports it and adds a port of minimp3's minimp3_ex helpers. Both crates are versioned together, and nanomp3 depends on exactly the same nanomp3-core version.

Breaking changes

• Decoder::decode now returns (usize, Result<FrameInfo, DecodeError>) instead of (usize, Option<FrameInfo>). DecodeError says why no audio came out (NoFrame, ReservoirUnavailable, Corrupt, UnsupportedLayer) and, for all but NoFrame, carries the frame's FrameInfo. Code that used if let Some(info) becomes if let Ok(info).
• Decoder::decode is generic over the output sample type (f32, i8, u8, i16, u16, i32, u32). Buffers whose element type was only pinned down by the old &mut [f32] signature (such as vec![0.0; n]) need an explicit type.
• FrameInfo is #[non_exhaustive] and has a new layer field.
• The minimum supported Rust version is 1.89 (1.80 without the wide feature).

Behavior changes

• Mono/stereo switches decode correctly. The original translation used minimp3's MINIMP3_NONSTANDARD_BUT_LOGICAL filterbank update, which deviates from the ISO reference on streams that switch between mono and stereo (51 dB PSNR on the conformance vector, versus 117 dB now). Output now matches minimp3's standard build bit for bit.
• MPEG Layer I and II (MP1/MP2) frames are decoded (feature layer12, on by default). Previously they were skipped.
• i16 rounding is fixed. minimp3 rounds samples in (−1.5, −0.5] to 0 instead of −1; nanomp3 now rounds half away from zero. Decoder::new_minimp3_compat() restores minimp3's exact output.
• Inputs shorter than 4 bytes, and several malformed inputs, no longer panic.

Added

• nanomp3: Reader (over any Read + Seek) and SliceReader (over bytes in memory): seekable, sample-accurate readers that skip tags, trim the encoder delay and padding for gapless playback, and know the stream length. Options configures them, including minimp3_compat.
• nanomp3: decode_all / decode_all_reader decode a whole stream into a Vec.
• nanomp3: frames / for_each_frame_reader iterate over frames without decoding; VbrTag parses Xing/Info/LAME tags; detect / detect_reader recognize MPEG audio; strip_tags, strip_trailing_tags and id3v2_len handle ID3v2, ID3v1 and APEv2 tags.
• Decoder::new_minimp3_compat(), Sample, DecodeError.
• Cargo features: wide (SIMD via the wide crate) and layer12 in both crates; alloc and std in nanomp3.

Fixed minimp3_ex bugs

The minimp3_ex port matches minimp3_ex exactly with Options::minimp3_compat. By default, these bugs are fixed, and the I/O reader behaves exactly like the in-memory one:

• Header-less APEv2 tags are over-trimmed by 32 bytes of audio.
• The I/O reader strips trailing tags again every time it reaches the end of the stream.
• Junk longer than the 128 KiB read buffer ends the I/O scan early (short length, incomplete seek index).
• The I/O frame iterator overstates the bytes available to the decoder.
• An I/O seek fails when a frame it needs is cut short by the end of the file.
• After a failed seek, minimp3_ex's next read returns memory past its buffer; nanomp3 stays in its error state until the next successful seek.

Internal

• No unsafe: both crates are #![forbid(unsafe_code)].
• Decoding is 36–79% faster than 0.1 and faster than minimp3's own SIMD build, with output still bit-identical to minimp3's reference build.
• parity/ compares nanomp3 against upstream minimp3 and minimp3_ex (a git submodule) bit for bit; fuzz/ has differential and robustness fuzz targets.
• Items marked #[doc(hidden)] (nanomp3_core::__private, nanomp3::__compat, nanomp3::__Raw, __raw(), Sample::from_synth) are internal and not covered by semver.