Cooklang Tree-Sitter Grammar

This is a Tree-Sitter grammar for Cooklang.

TODO

1. Align version across all files — Cargo.toml, pyproject.toml, Makefile say 0.0.1; package.json, tree-sitter.json say 1.0.0.
2. Align license across all files — Cargo.toml, pyproject.toml say MIT; package.json, tree-sitter.json say ISC.
3. Sections are parsed as dividers, not containers — fixing this means: change parsing so sections are containers, add a section fold rule to folds.scm, and document section folding in queries/README.md.
4. Strengthen Go and Python binding tests — the Go test (bindings/go/binding_test.go) only checks that the language loads; it does not parse anything, so it would not catch a broken external scanner. No Python test exists at all. Both should parse a snippet that exercises the scanner (e.g. @flour{1%cup}) and assert the expected tree shape.
5. Check and add language extensions — Two steps: 1/ tests to identify where extensions break the grammar and fix those and 2/ decide whether some/all extensions require specific parsing.

Notable Differences

• A step with multiple lines will have children node(s) per line whereas the rust parser will aggregate consecutive text lines. This is probably acceptable.
• It appears that the Cooklang BNF doesn't actually allow for punctuation (eg. -) in words (like ingredient names), but the compiler allows for it. I explicitly put it in this grammar since it seems useful. This is hacky.
• Newline and whitespace characters are handled slightly differently due to the way Tree-Sitter views them.
• The quantity delimiters { and } are not highlighted as @punctuation.bracket. They are produced by the external scanner (unlike ( / ) which appear as anonymous nodes in the grammar), so capturing them in a highlight query would require a dedicated quantity_open / quantity_close node type. The trade-off is accepted: adding those nodes would loosen the parse tree for a purely cosmetic gain.
• % is exposed as a named percent_separator node in the AST. A separator (either % or a space) is required between value and unit; bare concatenation like {2cups} is a parse error.

Testing

Tests use tree-sitter system:

tree-sitter test

tests are organized by the main aspect they're validating (see files test/corpus/):

tree-sitter test --file-name ingredient.txt

Or we can run test on a restricted scope using a regex against the test names:

tree-sitter test -i ingredient

Highlighting

Highlight queries (queries/highlights.scm) are tested with:

test/test_highlights.sh

The script contains inline .cook snippets, runs tree-sitter query against each one, and asserts that the expected captures fire. For interactive spot-checking, tree-sitter highlight works on .cook sfiles once the grammar's parent directory is listed in parser-directories in ~/.config/tree-sitter/config.json (note: the parent, not the grammar directory itself).

Releasing

Before tagging a release, bump the version in every file listed below to the same value, then create a git tag with that version (e.g. v0.1.0). The Go binding has no explicit version field — it is versioned by the tag itself, so the tag is the single source of truth for Go consumers.

File	Field
Makefile	VERSION (line 1)
package.json	version
tree-sitter.json	version (appears in both grammar and language blocks)
pyproject.toml	version under [project]
Cargo.toml	version under [package]

References

• OUTDATED Cooklang EBNF
• Python Tree-Sitter Grammar (Good for looking at newline stuff.)