prettier/prettier

Why this case study matters

prettier is the cleanest “structural floor on top” data point in the Wave 1+2 inventory so far: prettier is itself a code-formatter (it dogfoods), its existing tooling is mature and tightly curated (ESLint flat-config + 9 internal prettier-internal-rules/* + the self-prettier --check + cspell + knip + tsc + 5 custom node validation scripts), and yet the plugin-shape conventions that the entire codebase architecture rests on are encoded only in code review and folder-name memory.

For any plugin-architecture project — webpack, rollup, vite, babel, postcss — the “every plugin has shape X” rule is universal and rarely enforced. prettier shows what the structural floor looks like when it lands cleanly under a mature tool stack.

Headline catch

prettier ships eight plugins under src/language-*/ (js, css, html, markdown, yaml, json, graphql, handlebars). Each one must export index.js (parser/printer/options) plus languages.evaluate.js (linguist-languages lookup table consumed at build time). A plugin that drops languages.evaluate.js would silently be omitted from the production bundle. Yet nothing on disk asserts this contract. Same thing for the 1:1 mapping between plugins and changelog_unreleased/<lang>/ category directories: a new plugin landing without its category dir ships fine, and PR notes for it have nowhere to go.

Alint’s for_each_dir over src/language-* is the missing structural floor. Five net-new gates that prettier’s existing ESLint + prettier-itself + cspell + knip + tsc stack does not enforce today:

• prettier-each-language-plugin-has-index
• prettier-each-language-plugin-has-languages-evaluate
• prettier-each-plugin-package-shape
• prettier-plugin-package-prettier-scope
• prettier-changelog-categories-exist

Where alint earns its keep here

• Single declarative gate catches both directions of the plugin ↔ changelog-category drift in one rule.
• Script consolidation — prettier has 5 custom node scripts (lint-changelog.js, check-deps.js, format-test-lint.js, ensure-no-files-changed.js, clean-cspell.js) totalling ~330 LoC, plus 8 yarn lint scripts, plus a 16-step GitHub workflow. Five of six invariants in lint-changelog.js and one of two in check-deps.js collapse to declarative alint rules. That’s ~80 LoC of bespoke JS replaced by ~50 lines of declarative YAML — easier to read in PR review, easier to extend, schema-validated at config-load time.
• Speed delta — yarn lint runs the 8 lint scripts in parallel via npm-run-all2 -p. Each script does its own fs walk; cspell
  • prettier + eslint + tsc each pay startup cost per process. Typical wall-clock on a clean checkout: ~30-45 s on a developer laptop. alint runs the alint-replaceable subset in ~0.5-1 s expected.
• Cross-cutting v0.10 candidate driver — prettier’s ensure-no-files-changed.js is the canonical command_idempotent shape (most-demanded primitive in the inventory: prettier, ruff, kubernetes, airflow, turbo). format-test-lint.js is the canonical for_each_leaf_dir shape (prettier, rust-lang/rust, ruff, any test-corpus tree).

Future story angles

• Plugin-architecture launch tile — generalise from prettier to webpack, rollup, vite, babel, postcss. Every plugin-architecture project has the “every plugin has shape X” universal-and-unenforced rule. prettier is the canonical demo.
• “Adopt alint to consolidate the orchestration layer” angle — pair with cpython for the case where mature tooling already exists but the structural floor is the missing piece.
• command_idempotent v0.10 launch post — prettier’s idempotency check is the cleanest example of the most-demanded primitive in the inventory.
• for_each_leaf_dir v0.10 launch post — prettier’s format-test-lint pattern recurs in every test-corpus tree (rust-ui, ruff snapshots, deno specs).