leonardomso__rust-skills/checks

checks — compile-verify the rule examples

A dev tool that type-checks the ```rust code blocks in ../rules/*.md so the "Good" examples we tell agents to write actually compile. Not part of the published skill.

Run

# structural / link / index checks (no toolchain needed)
python3 checks/validate.py

# compile-check the examples
cd checks
python3 gen.py                                              # extract blocks -> examples/
cargo check --examples --keep-going --message-format=json > check.json
python3 analyze.py check.json                               # classify results
python3 analyze.py check.json --check-baseline baseline.txt # CI gate: fail on NEW suspects

Both run in CI (.github/workflows/ci.yml): validate (Python only) and examples (pinned to Rust 1.95.0, the toolchain baseline.txt was generated on).

Updating the baseline

baseline.txt lists the currently-accepted suspects (fragments/pseudocode the heuristics can't auto-classify). The CI gate fails only on signatures not in it. After intentionally adding/changing examples, regenerate it on the pinned toolchain and review the diff:

rustup run 1.95.0 cargo check --examples --keep-going --message-format=json > check.json
python3 analyze.py check.json --emit-baseline > baseline.txt

When bumping the pinned toolchain in ci.yml, regenerate baseline.txt on the same version in the same commit.

How it works

gen.py extracts each candidate block into examples/<name>.rs, wrapping fragments in an async fn -> Result<...> so ? and .await type-check. It skips blocks that can't compile standalone by design: ## Bad anti-patterns, nightly #![feature] gates, procedural-macro code, placeholder crate names (my_crate, …), and bare ... pseudocode.

analyze.py buckets each failing example by compiler error code:

• fragment — every error is name resolution (undefined symbol/crate). These reference helpers defined elsewhere in the rule; expected, ignored.
• artifact — caused by extraction (a &self method body wrapped as a free fn, pseudocode .../??? tokens, dangling doc comments). Not real bugs.
• low — only "type annotations needed"; compiles in the rule's real context.
• SUSPECT — anything else (type mismatch, no-method, bad syntax, wrong arity, missing trait impl). These are the ones to review and fix.

Notes

• Run on Rust ≥ 1.95: some examples use APIs stabilized in 1.95 (e.g. the MaybeUninit array From conversions) and will spuriously fail on older toolchains.
• Generated files (examples/, *.json, manifest.json, target/) are gitignored; only the source (gen.py, analyze.py, Cargo.toml) is tracked.