[paoValle]

Keeps extractor class names in the browser bundles, so parse() reports extractorType: 'github' instead of 'a' for anyone consuming dist/index.js or dist/index.full.js.

Fixes #414.

Why

extractorType is derived from constructor.name, which Terser replaces with short names. Node loads the unbundled CJS and reports the site correctly, so the defect only shows up in a browser extension or any bundler that consumes the published bundles — where the field is documented to name the extractor that ran, and came back as a, i or v.

What changed

keep_classnames: /Extractor$/ on the TerserPlugin minimizer both bundles share, so only the classes the field depends on are kept and the bundle's other classes still minify. Measured cost, npm run build + npm run size: dist/index.js 90.2 → 90.4 KB gzip, dist/index.full.js 206.9 → 207.1 KB, both inside the budgets in scripts/check-bundle-size.mjs (92 and 215 KB), so no budget change.

scripts/check-extractor-names.mjs asserts the built bundles still contain the extractor class names, as npm run lint:extractor-names, wired into the build job and the release checks next to the size budget. Vitest runs on the TypeScript source, so nothing in the suite could notice this disappearing.

I went with the build option rather than a static name per extractor because the latter means touching every extractor and guarding the convention anyway; if you would prefer that shape, say so and I will redo it.

Verification

• npm run build then a probe that loads the built full bundle with jsdom against a minimal LWN page and prints result.extractorType: "a" before this change, "lwn" after.
• npm run lint:extractor-names → passes with the option, fails (✗ dist/index.js is missing BaseExtractor, GitHubExtractor, YoutubeExtractor) with it removed.
• npx vitest run → 789 tests passed. The six website test files fail to load in a fresh clone (Cannot find package '@lezer/html', @lezer/highlight) and fail the same way with this change stashed.

One path this does not cover: website/scripts/build.mjs bundles the worker with esbuild (minify: true) and no keepNames, so the playground and the worker may still see a mangled name. It is a one-line change there, but it is a separate build with its own dependencies, so I left it out of this PR — happy to add it or open a follow-up if you want it.

[paoValle]

Rebased onto main 0.19.6 (e4f5cca → 643dc40); the branch had gone conflicting.

The only conflict was CHANGELOG.md, where both sides added a bullet to the 0.19.6 section — kept both, so the entry reads the security fix and then this one. package.json, webpack.config.js, the two workflow files and scripts/check-extractor-names.mjs applied cleanly.

Re-run after the rebase, with npm ci and npm ci --prefix website (the six website test files fail to import @lezer/* without the second one, which is the step CI runs):

• TZ=UTC npx vitest run — 45 files, 848 tests passed;
• TZ=UTC DOM=jsdom npx vitest run — 45 files, 848 tests passed;
• npm run build, npm run lint:extractor-names, npm run lint:innerhtml, npm run lint:unused — all clean.

The CI run on this branch is action_required (the fork's first workflow run needs a maintainer's approval), so nothing has executed here yet.