xdms-rs

A pure-Rust, dependency-free library for unpacking DMS (Disk Masher System) Amiga disk archives into raw ADF disk images.

Origin

A clean-room Rust port of xDMS, the portable public-domain DMS unpacker written by André Rodrigues de la Rocha and maintained by Heikki Orsila. DMS was the de-facto Amiga standard for storing non-DOS disks (games, demos); emulators expect ADF/ADZ instead. The original C decoder is the reference for this port. Dual-licensed MIT OR Apache-2.0.

Usage

// One-liner: decompress a .dms to a .adf
xdms::unpack_file("disk.dms", "disk.adf")?;

// Or get the bytes directly
let adf: Vec<u8> = xdms::unpack_to_vec(std::io::stdin().lock())?;
# Ok::<(), xdms::Error>(())

use xdms::DmsArchive;

// Full control: metadata, password, salvage mode, custom sink
let mut archive = DmsArchive::open("disk.dms")?.with_password("secret");
println!("{}", archive.info());          // human-readable summary
let summary = archive.unpack_to(std::io::stdout().lock())?;
println!("{} tracks", summary.tracks);

// Integrity check only (no output):
DmsArchive::open("disk.dms")?.verify()?;
# Ok::<(), xdms::Error>(())

Test coverage

All seven compression modes are implemented. How each is verified:

Encryption, banner / FILEID.DIZ capture, salvage mode, and header/CRC/checksum validation each have tests too.

Fixtures are never committed as large blobs: the HEAVY disk images download on demand (cargo run --example fetch_fixtures) into a gitignored directory and their tests skip when absent, while the tiny RLE archive is committed under tests/data/. "Byte-exact" means the decoded ADF's SHA-256 matches the reference output. Run the suite with cargo test.

Performance

Pure decode throughput vs the reference C xdms, both decoding the same public-domain disks to memory (Rust --release, C -O2, Apple arm64, startup amortized over 2000 iterations). Output is byte-identical.

no_std

The crate is no_std + alloc when built with default-features = false; the std feature (on by default) adds the std::io Read/Write API.

Design goals

The port was planned around a few deliberate choices:

• Deep modules, small surface. A handful of types (DmsArchive, Info, Summary, Error) and a few one-liner functions; every decompressor, bit reader, Huffman table, and sliding window is hidden behind one Decompressor.
• Idiomatic, expressive Rust — not a C transliteration. The C's magic integers become enums/newtypes (Mode, DiskType, GenInfo, TrackFlags), its #defines become named constants, and byte parsing/validation lives in From/TryFrom impls. Names say what they mean (packed_len, not pklen1).
• Very few dependencies — in fact zero. No runtime or dev dependencies; CRC-16, the bit reader, the Huffman builders, and even the tests' SHA-256 are implemented in-crate.
• core/alloc-first. The engine uses only core and alloc; std::io is confined behind a default std feature, so a no_std build is mechanical.
• Faithful and byte-exact. Output is validated to match the reference C xdms bit-for-bit; decode loops mirror the original algorithms closely.
• Test-driven, with an oracle. Built red→green→refactor, using the original C binary (and the independent adf2dms encoder) as golden references.
• Portable and linted. Pure-Rust engine with no platform APIs; CI covers Linux/macOS/Windows plus a no_std build, with Clippy's nursery lints on.
• Comments explain why, not what — reserved for format quirks and porting rationale.

Assistance of Claude Opus 4.7 was used, adhering to the LLVM AI tool use policy.

License

Licensed under either of MIT or Apache-2.0 at your option.