alint — rust-lang/rust

rust-lang/rust ships its own ~5kLoC custom linter binary because no off-the-shelf tool can express its structural rules. alint covers the canonical 30 % of those rules in 18 lines of YAML — and the gap analysis points at exactly four new rule kinds that would push that to ~55 %.

Why this matters

rust-lang/rust is the canonical “we own a compiler, of course we shipped our own linter binary” repo. src/tools/tidy/ is a ~5kLoC Rust binary that runs 32 named tidy checks plus an extra_checks dispatcher that fans out to ruff, eslint, clang-format, shellcheck, and typos. Every change to a check requires a rebuild of the binary. Every new contributor structural rule needs Rust + a pub mod foo; line + a check!() registration

a unit test. The orchestration plumbing — walk the tree, dispatch the linter, collect the diffs — is the same shape alint already expresses.

This is the second-strongest case study in the catalogue (behind kubernetes) for the launch positioning, and uniquely valuable for the “polyglot monorepo / project-with-its-own-linter- binary” audience — the readers who already know generic linters don’t fit their needs and have built bespoke tooling.

Headline catch

~13 of the 32 tidy modules become declarative. The 30 % that fit alint’s grammar map cleanly to 18 lines of YAML — including the most-often-cited tidy::style checks (line length, file length, trailing whitespace, line endings, no-TODO), edition pin, extdeps allowlist, known-bug headers, rustdoc-gui description lines, Windows-illegal filenames, no-#[test] in stdlib, debug-artifact guards. Plus the extra_checks dispatcher folds into one command: rule per scope.

Even on the rest, alint replaces the outer “walk the tree, dispatch the linter, collect the diffs” plumbing while shelling out to ruff/eslint/clang-format via the command rule kind. Net: alint can take over the orchestration of ~13 of the 32 tidy modules + most of the verify-*.sh scripts with one declarative config, which is what src/tools/tidy/ was bespoke-built to do.

The pitch lands as: “we’re not asking you to throw away your custom linter, we’re asking you to push the 30 % of mechanical checks down into a declarative layer so you can focus your handwritten Rust on the AST-aware domain logic that actually needs a custom binary.”

Where alint earns its keep here

The 30 % declarative subset is one declarative file vs. 13 modules + a lib.rs registration + a main.rs check!() call + per-module unit tests. Adding a rule to alint is 5-10 lines of YAML; adding one to tidy is a new Rust module.
tidy::alphabetical is the canonical demand-driver for the v0.10 ordered_block rule kind. sortedness between marker pairs is the single most-requested missing rule kind across the case-study catalogue — [dependencies] ordering in every Cargo workspace, requirements.txt, imports blocks. The // tidy-alphabetical-start/-end markers are the canonical example.
tidy::triagebot is a 6-source confirmation for v0.10 registry_paths_resolve. Alongside clap + cpython + arrow + pytorch + tensorflow — past saturation. The rule generalises to CODEOWNERS validation in every GitHub repo, ESLint overrides[].files, Cargo [[bin]].path, and dozens of similar config-driven path registries.
A tidy@v1 bundled-ruleset draft is a one-line adoption story. Packaging the canonical 30 % as alint://bundled/tidy/rust@v1 (with scope_filter to exclude src/llvm-project/ and src/gcc/ by default, and parameterised line-length thresholds for .goml / error-code markdown) would raise the case study’s pitch from “18 lines of YAML” to “1 line of YAML” for every rust-lang/rust contributor.
The boundary is honest. The 14 out-of-scope tidy modules (AST analysis, codegen drift, feature/error-code cross-reference, target-policy heuristics) are explicitly recommended to stay on ./x test tidy. Same framing as the kubernetes case study — alint’s non-goals are deliberate.

Future story angles

Launch tile candidate — second tile after kubernetes for the “project-with-its-own-linter-binary” audience. The contrast lands: kubernetes has 50 verify-scripts → alint replaces 17; rust-lang/rust has one custom linter → alint covers 30 % of it declaratively.
ordered_block v0.10 ship narrative — rust-lang/rust is the canonical example. When the rule kind ships, the case study’s “13 of 32” jumps to “14 of 32” and the headline gets to add “and tidy-alphabetical-* is now declarative too.”
Bundled-ruleset adoption ladder rung — alint://bundled/tidy/ rust@v1 is a launch-story candidate; the “1 line of YAML for rust-lang/rust contributors” framing is sharper than the current “18 lines.”
scope_filter evolution refactor — current config repeats src/llvm-project/** + src/gcc/** excludes across 5 rules; v0.9.17’s named-scope evolution centralises this into a single source of truth and is itself a “what does the v0.10 surface buy you?” demonstration.