wolf-lang Public

Notifications Fork 0 Star 0
trunk
Branches bench-data trunk Tags v0.1.0
2 Branches 1 Tags
Go to file T
Code
mfwolffe merge s84: iteration that does not allocate
27cdb14 311 Commits
.cargo scaffold cargo workspace: pipeline crates with locked dependency direction, spec and corpus stubs
.github bench: measure the right binary, against a baseline that executes
bench str: words, lines and split iterate without allocating
corpus str: words, lines and split iterate without allocating
crates str: words, lines and split iterate without allocating
docs midend: rle_and_forward guards escape, and the call-site fact lands in its sound half
fuzz wolf build --release end-to-end: 94 of 94 run-phase corpus files at identical verdict and stdout sha across the tiers, overflow trapping identically at O2 (X3 is profile-independent); the metadata-strip fuzz rig with the three historical shapes and the CFG-duplication stressor as permanent regressions; the tier-R compile bench lane reports 0.41s clean vs 0.075s debug
spec spec: the clauses words, lines and split answer to
xtask differ: publish lane coverage, and gate it
.editorconfig pin toolchain, add deny config and repo conventions
.gitattributes gitattributes: LF checkout everywhere — byte-exact spans require byte-stable sources
.gitignore s44: the M2 measurement protocol, and the verdict it returns
CONTRIBUTING.md Update commit message guidelines in CONTRIBUTING.md
Cargo.lock docs/api generated over wolf-std: 358 items, 357 documented, 309 with an example
Cargo.toml docs/api generated over wolf-std: 358 items, 357 documented, 309 with an example
LICENSE license: GPL-3.0-or-later with the wolf Runtime Library Exception on wolf_rt (D41 as amended) — the GCC model: copyleft protects the toolchain, the exception keeps compiled programs their authors' own
README.md docs: the prose pass — plain sentences, no antithesis tic, no unearned adjectives; twelve claims corrected against the source (the release notes' code count, the corpus counts, the tier the compiler actually ships)
clippy.toml pin toolchain, add deny config and repo conventions
deny.toml pin toolchain, add deny config and repo conventions
rust-toolchain.toml pin toolchain, add deny config and repo conventions
rustfmt.toml pin toolchain, add deny config and repo conventions

Contributing to wolf

Commits

  • Commit often, in chunks: one logical change per commit. Never git add -A.
  • Messages: terse, imperative, under 250 characters.
  • Tests land in the same commit as the code they test. Benchmarks land with the perf claims they prove.

Quality bars

  • cargo xtask ci must be green before pushing. It runs cargo fmt --check, clippy -D warnings, and the workspace tests, then every verify lane in turn (deps-check, corpus, abi-check, debug-check, and the rest). The step list is the ci() table in xtask/src/main.rs; a lane added there is a lane you have to pass.
  • The crate dependency direction is locked (see Cargo.toml and cargo xtask deps-check): span ← diag ← {lex, ast} ← parse ← sema ← mem ← wir ← backend ← codegen_* ← driver; wolf_rt may depend on wolf_span at most.
  • Every diagnostic gets a reviewed snapshot. A corpus surface change updates corpus/ in the same commit as the spec change, so the corpus never drifts from the locked surface (s02 discipline; review checklist item).
  • Diagnostic messages follow crates/wolf_diag/VOICE.md, and a PR that adds or changes one quotes the guide in review. Every code needs a registry entry with a real explanation and at least one snapshot fixture; cargo xtask diag-catalog gates that. Compiler-phase crates never print (cargo xtask print-gate).
  • Platform-agnostic by default: code must not assume linux/x86-64. The CI matrix (linux x86-64/aarch64, macOS aarch64, windows x86-64, freebsd cross-build) is the arbiter.

Testing conventions

  • Property tests use proptest. Case count follows PROPTEST_CASES, which is small in PR CI and large in nightly CI. Exemplar: crates/wolf_span/tests/.
  • Snapshots use insta. Run cargo insta review to update one deliberately; exemplar: xtask/tests/directive_snapshots.rs.
  • Benchmarks: cargo xtask bench --track=<runtime|compile> emits JSONL to bench-results/. cargo xtask bench diff <base> <cand> [--gate] is the variance-aware comparison (median + 3×MAD noise floor, 2% practical floor, N≥10 runs). Reference kernels live in bench/kernels/.
  • Fuzz targets live in fuzz/. cargo xtask fuzz-smoke builds them where cargo-fuzz (nightly) is available, and skips loudly where it is not.

Toolchain

  • Pinned in rust-toolchain.toml (rustup/CI) and rust-version (everyone). Bump it in its own commit, with CI green.

Where the plan lives

  • .docs/ (untracked): sprint plan, decision log, research reports. The sprint files under .docs/sprints/ are the implementation contract; the decision log .docs/planning/02-decisions.md is the design authority.