Development and checks

Read the contributing guide and architecture index before changing the implementation. The main crate has no dependency on another Zstandard implementation. The C reference is used only by development, tests, and benchmarks.

Set up

Requires Rust 1.89+, a C compiler for the reference oracle, and Git:

git clone --recurse-submodules https://github.com/M-Compressors/mzstd.git
cd mzstd

For an existing clone, initialize git submodule update --init --recursive.

Format, lint, and test

Differential tests run in release mode because debug execution is too slow:

cargo fmt --all -- --check
cargo clippy --workspace --all-targets --all-features --locked -- -D warnings
cargo clippy --workspace --all-targets --locked -- -D warnings
cargo clippy --workspace --all-targets --features framed --locked -- -D warnings
cargo test --workspace --release --locked
cargo test --workspace --all-features --release --locked
cargo test --workspace --features framed --release --locked

All features enable no_std, removing std-only APIs, so the default and framed std configurations need separate runs. CI also checks each codec separately, the MSRV, bare-metal compilation, rustdoc, packaging, and index overflow correction.

Fuzzing, memory, and proofs

AFL and libFuzzer are excluded packages with their own feature checks and regression commands. Follow the contributing guide and fuzzing specification. Never commit generated findings or corpus output directories; minimize failures into regression inputs.

Use scripts/miri.sh and the documented AddressSanitizer commands after changing SIMD or uninitialized-output paths. Inspect function coverage with cargo llvm-cov; generated reports stay local. Changes to verified format primitives require refreshing Charon/Aeneas output and rebuilding Lean proofs. See CI memory checks and formal verification.

Documentation and benchmarks

Keep architecture specifications and Mermaid diagrams current, add new specifications to the index, and record user-visible changes in the changelog. All guide and architecture pages are published together. Validate the book and its runnable recipes:

python3 -m unittest discover -s scripts -p test_docs.py
python3 scripts/docs.py test-examples
python3 scripts/docs.py build

Install the pinned tools and preview as described in documentation website. The benchmarking guide describes cold comparisons, profiling, raw measurements, and chart generation. Performance changes need representative measurements of throughput and output size, with the command, machine, input, and before/after results recorded.