cairn-lang-tree-sitter

tree-sitter grammar for the Cairn build description language.

• Rust: use cairn_lang_tree_sitter::LANGUAGE; (also exports HIGHLIGHTS_QUERY, LOCALS_QUERY, INJECTIONS_QUERY).
• Node: require('tree-sitter-cairn').
• WASM: attached to each GitHub Release as tree-sitter-cairn.wasm.

See ../../docs/superpowers/specs/2026-07-23-tree-sitter-cairn-design.md for the design and ../../docs/superpowers/plans/2026-07-23-tree-sitter-cairn.md for the implementation plan.

Development

cd crates/cairn-lang-tree-sitter
pnpm install
pnpm run generate  # regenerate src/parser.c after grammar.js edits
pnpm test          # run tree-sitter corpus tests
cargo test -p cairn-lang-tree-sitter  # run Rust integration tests

Grammar edits without a matching pnpm run generate will fail the generate-check job in .github/workflows/tree-sitter.yml.

Corpus conventions

tree-sitter test strips field labels from the parsed tree whenever the expected S-expression has none, so most cases in test/corpus/ stay terse and pin node shape only. Dropping a field(...) from grammar.js leaves that shape intact, which is why every rule that declares fields also has exactly one case whose expected tree spells the labels out. Those cases carry (field labels) in their name — for example def declaration (field labels):

(source_file
  (def_decl
    name: (identifier)
    args: (attribute_list
      (attribute key: (identifier) value: (size_literal (integer) (integer))))
    body: (struct_body
      ...)))

Labelling is all-or-nothing per case: one label anywhere in the expected tree makes the whole tree compare with labels, so a partially labelled case fails. When editing a (field labels) case, keep every label in place; when adding a field(...) to grammar.js, label one representative case for the new rule.