StuffIt decoders: classic method 13 (LZ+Huffman), 5 (LZAH), RLE90; SIT5 Arsenic

[MagicalTux]

fstool now has a read-only StuffIt reader (src/fs/archive/sit.rs, behind the sit feature) for the classic SIT! container. It decodes data-fork method 0 (store) today; the compressed methods need codecs here and currently read as a clean Unsupported.

Classic SIT! methods needed (per fork):

• 13 — LZ + Huffman. The common StuffIt Deluxe method — highest priority.
• 5 — LZAH (LZ with adaptive Huffman).
• 1 — RLE90. (Shared with ARC method 3 — see the RLE90 question in the ARC issue.)
• Lower priority: 3 (Huffman), 2 (LZW/Compress), 8 (MW).

StuffIt 5 (StuffIt (c)1997… magic) is an entirely different, largely proprietary container; fstool currently routes it to a detection-only fallback. Its main entropy stage is the Arsenic range coder. This is lowest priority / may be declined — happy to leave SIT5 unsupported.

API: Decoder-shaped, decode-only, driven via DecoderReader with the fork's known uncompressed size from the entry header.

Reference: The Unarchiver / libxad XADStuffIt* parsers (clean-room RE; decode-only there too).

[MagicalTux]

Partly addressed by #68 (sit13 feature) — but note method 13 is not yet a working decoder. The format is proprietary and the only public reference (The Unarchiver / XADMaster) is LGPL, which can't be copied into this MIT clean-room crate. So #68 shipped only method 13's building blocks — streaming bit reader, Kraft-validated Huffman decoder, bounds-checked LZSS window, all unit-tested — behind a decoder that returns Unsupported. A conformant payload decoder needs either a public spec or a clean-room functional spec + real fixtures.

Status of the rest:

• Method 5 (LZAH) — prototyped but held back: it only round-trips with itself and is not byte-compatible with real StuffIt LZAH (no public wire spec). Preserved on a tagged branch (held/lzah).
• SIT5 Arsenic — prototyped (full BWT + range-coder pipeline) but held back for the same reason; tagged held/arsenic-sit5. (You noted this may be declined — that's where it stands.)
• Method 1 (RLE90) — not shipped; needs a dedicated rle90 codec (see the RLE90 note on SEA ARC codecs: crunch (dynamic LZW) + squeeze (RLE+Huffman); confirm RLE90 coverage #66; the existing rle is a different scheme).
• Methods 3 (Huffman), 2 (LZW), 8 (MW) — not started.

To finish 13 / 5 / Arsenic properly we'd use a clean-room functional spec derived from the LGPL reference by a separate party (a whiteroom-analysis prompt + black-box fixture plan is ready), then implement against that. Happy to do so once those artifacts exist.

[MagicalTux]

Done: the three headline methods all shipped and validated bit-exactly against real archives — method 13 (#72, 28 forks), method 5/LZAH (#71, 17 forks), and SIT5 Arsenic (#73, 46 forks + SHA vs unar; this one you'd flagged as possibly-declined). Remaining minor work split into follow-ups: #75 (RLE90 = method 1, shared with ARC) and #77 (methods 2/3/8). Closing in favor of those.