semanticversion-rs — Rust port of python-semanticversion

Port Mortem 2026 · Track D (Python → Rust) · solo build

A complete, memory-safe Rust reimplementation of python-semanticversion: SemVer 2.0 parsing/comparison plus npm-style SimpleSpec / NpmSpec / LegacySpec range matching.

Headline

The original, unmodified pytest suite passes against the Rust build via a PyO3/maturin extension named semantic_version — 54 tests + 586 subtests green, zero test edits. The port is 100% safe Rust (zero unsafe), validated by a 24,500-pair differential fuzz + 2.5M crash-fuzz runs (0 divergence, 0 crashes), documented in a 20-entry decision log, and benchmarked at up to 60× faster spec matching, ~11× faster parsing, 21% lower memory.

Verify it yourself (one command)

make        # builds the extension + runs the ORIGINAL unmodified pytest → 54 passed / 16 skipped

make = maturin develop && pytest tests/original -q (exits non-zero on any failure). The 16 skips are test_django.py ("Django not installed") — identical to the original baseline.

Architecture (all DONE)

Module	Status	Role
error	✅	SemverError → PyValueError
identifiers	✅	prerelease / build identifier rules
version	✅	parse / display / coerce / ordering / partial
clause	✅	Clause AST + Range matching + empty-marker policies
simple_spec	✅	SimpleSpec grammar + AST emission
npm_spec	✅	x/hyphen/caret/tilde + prerelease OR-expansion
bindings (PyO3)	✅	the semantic_version extension (dunder surface, Spec/LegacySpec)

Proof layer

• Original pytest unmodified: 54 passed / 16 skipped / 586 subtests.
• Differential fuzz: 24,500 random pairs (Rust binding vs original Python) → 0 behavioral divergences; 2,554,822 crash-fuzz runs → 0 panics (fuzz/log.txt).
• 8 latent port bugs caught & fixed by the fuzzer (e.g. 18 u64 overflow-panic sites → saturating_add; ~*/^* rejection; empty-prerelease 1.2.3-..; || empty-group; AllOf-wrap shape; hyphen LTE fence). See DECISIONS.md D18.
• Zero unsafe: grep -rn unsafe src/ → empty.
• Decision log: 20 entries (D00–D19) in DECISIONS.md.

Benchmarks (bench/, see bench/methodology.md)

Aggregate 9×, top 60× (npm matching), ~11× parsing, 21% lower RSS. Honest note: the PyO3 precedence-key path drags the aggregate (Python tuple overhead); native precedence runs at ~386 ns p50. Hackathon cloud VM, not bare-metal.

Demo

Team

in.rahul.dev — Rahul Gupta (a.k.a. Pranav) · Discord pranav_dev. · GitHub rahulgupta0-dev · solo.

License

BSD-2-Clause — same as the original project; see LICENSE. This is a from-scratch Rust port.