feat(syntax-language): own the extension-to-grammar map

[tmccoy14]

CodeBlock highlights through microlighter, so which grammar handles which file is a fact about this package's own dependency. Guardian Studio and Explorer each kept a byte-identical copy of that map, and each declared microlighter as a devDependency purely to test it — reaching past equality to inspect its transitive dependency, with independently pinned ranges that could resolve to a version equality does not ship. Copilot flagged that as critical on eqtylab/explorer#62.

Owning the map here lets both consumers drop that devDependency.

Surface

New subpath, @eqtylab/equality/syntax-language:

GRAMMAR_BY_EXTENSION: Record<string, string>   // 55 entries
GRAMMAR_BY_FILENAME:  Record<string, string>   // 5 entries
grammarForFilename(fileName): string | undefined
hasFileExtension(fileName): boolean

Deliberately not re-exported from the root barrel. It is data, not a component, and reaching it through the barrel pulls every CSS module into a plain Node consumer — which fails outright in a test runner with Unknown file extension ".css". The subpath bundle has no imports at all.

This is purely additive: nothing importing from the root barrel changes.

Only what this release ships

The map names only grammars microlighter 2.1.0 actually ships, so it describes what a file will really look like. An earlier draft pre-named grammars that exist upstream but are unreleased (astro, ini, nginx, nix, ron); that was dropped, because a .ini file rendering plain while the map says ini: 'ini' reads as a bug to anyone debugging it.

Nothing is lost by waiting: the check below reports a newly shipped grammar as unreachable and names it, so the build tells you what to add — and the extension stays a deliberate choice (.cfg → ini is a judgement call, not a given).

Why a build check instead of tests

Nothing upstream can generate this map. The grammar modules carry no file-type metadata (0 of 37 declare fileTypes), and microlighter's own alias list covers markdown fence tags rather than filenames — it has nothing for cc, hpp, h, m, ps1, patch, and nothing for extensionless names like Dockerfile. Some entries are also editorial: xml → html is a substitute, since the HTML grammar already describes its tag-and-attribute shape.

So the data stays hand-written, and what gets automated is keeping it honest. src/scripts/check-grammars.js runs inside build:lib, which means code-quality.yaml gates it on every PR and publish.yaml gates it on every release — a drifting map fails the build rather than shipping code blocks that silently render plain.

Because nothing is pre-named, the invariant is now an exact correspondence in both directions:

Checked 37 mapped grammars against 37 shipped

and when it drifts:

- Maps onto 1 grammar(s) the highlighter does not ship: notagrammar.
    Either the name is wrong, or the grammar is not in this release yet and the entry should wait.
- The highlighter ships 1 grammar(s) no filename reaches: sql.
    Add an extension for each, or list it in NOT_FILE_SHAPED if no filename would select it.

Verification

• check:grammars passes against 2.1.0 — 37 mapped, 37 shipped, exact correspondence
• Confirmed it exits 1 on both failure directions
• build:lib succeeds; dist/lib/syntax-language.{js,cjs,d.ts} emitted and the subpath resolves with no imports
• prettier and eslint clean

Consumers are ready and waiting on this release: they move to the subpath, delete their copies, and drop the microlighter devDependency. Verified locally against this build — Explorer 36 tests green across 3 builds, Studio CI green at 138 tests.

🤖 Generated with Claude Code