vulture-rs 1.0.0

Rust reimplementation of vulture (dead code finder) - output-identical to vulture 2.13

vulture-rs

A Rust reimplementation of vulture (the Python dead-code finder), built on the ruff parser for performance. Versioned independently; the upstream parity target (currently vulture 2.13) is stated here, in the package description, and in the CHANGELOG (a dated entry per release, enforced at release time). Drop-in compatible: the binary is named vulture, accepts the same CLI, and produces byte-identical output — tools that shell out to vulture via PATH pick it up unchanged (the one visible difference: --version reports this package's own version, not 2.13).

Install

pip install vulture-rs   # ships the `vulture` binary

Wheels for linux x86_64/aarch64 and macOS arm64, published via the manually-dispatched Release to PyPI workflow (wheels only, no sdist); every wheel is smoke-tested before it ships.

Results

A large production codebase (6,553 files), 8 cores, median of 5:

tool	wall time	speedup
Python vulture 2.13	14.31 s	1x
vulture-rs (8 threads)	0.43 s	33x
vulture-rs (1 thread)	1.06 s	13.5x

Peak memory is comparable (~100 MiB).

Parity

Output is byte-identical (stdout + exit codes) to vulture==2.13:

• on the 6,553-file production codebase above — 2,687 findings, zero diff — across assorted flag combinations;
• across tests/differential.sh: an edge-case corpus (noqa quirks, whitelist masking, elif chains, encodings incl. latin-1/BOM/CRLF, unreachable code, invisible bindings, decorated defs) times a 12-entry flag matrix;
• on two alien corpora: CPython 3.12's stdlib (2,621 findings, byte-identical incl. normalized stderr) and a 22,286-file site-packages corpus (58k findings, 137s -> 3.4s, 41x). There, 144 of 58,008 stdout lines and 11 stderr lines differ - every one triaged to registered deviations (D1 type comments, D16 CPython SyntaxWarning chatter), zero parity bugs;
• on six popular open-source repos (requests, flask, click, rich, celery, django; 3,736 files, 25-54x) — byte-identical.

Known, deliberate deviations (error-message prose, type comments, crash-path cleanups) are enumerated in docs/vulture-rust-spec.md §1; the full behavioral contract lives in docs/vulture-spec.md. Both specs were reverse-engineered from vulture 2.13 and review-hardened before this implementation.

The parity reference is vulture 2.13 on CPython 3.12: the parser is version-gated to 3.12, so syntax newer than that (PEP 696 type-param defaults, t-strings, …) is a syntax error here exactly as it is there. Bumping the reference is a deliberate task.

Build & test

cargo build --release            # target/release/vulture
cargo test                       # CPython-generated vector tests
tests/differential.sh <python-vulture> target/release/vulture
tests/og_corpus.sh <python-vulture> target/release/vulture

CI (.github/workflows/ci.yml) runs fmt, clippy, the vector tests, and both differential harnesses against a pinned vulture==2.13 on every PR and push to main.

Regenerate the test vectors with python3 tests/gen_vectors.py (needs CPython 3.12).

The ruff_* dependencies are git-pinned to a ruff release tag (the crates.io releases under those names are stale placeholders). Treat upgrades as a deliberate task gated on the differential suite.

Usage

Same CLI as Python vulture:

vulture <paths> [--exclude PATTERNS] [--ignore-names PATTERNS]
        [--ignore-decorators PATTERNS] [--min-confidence N]
        [--sort-by-size] [--make-whitelist] [-v]

Exit codes: 0 clean, 1 invalid input, 2 invalid arguments, 3 dead code found.