tree-sitter-carve

Tree-sitter grammar for Carve, a post-Markdown lightweight markup language with visual mnemonics.

This grammar is built from the proven Djot Tree-sitter scanner architecture and changes the public syntax to Carve:

• /italic/, *bold*, and /*bold italic*/
• =highlight=, plus the braced-only {^superscript^} and {,subscript,} (a bare ^ or , is literal text)
• $ + backtick math spans, and $$ + backtick display math spans
• :name[content] inline extensions, @mentions, #tags, and :emoji:
• %% line comments, trailing inline %% comments, and %%% fenced comments
• Djot-style blocks retained where Carve keeps them: headings, lists, tables, fenced code, links, images, attributes, footnotes, captions, and divs

Development

git clone --recurse-submodules https://github.com/markup-carve/tree-sitter-carve
npm install
npm run generate
npm test

If you cloned without --recurse-submodules, fetch the shared spec corpus with git submodule update --init.

Shared-corpus conformance

The canonical Carve spec corpus is vendored as a git submodule at spec/ (spec/tests/corpus/NN-slug.crv). Two checks run it against this grammar:

npm run test:coverage    # every spec category is classified (covered or skip)
npm run test:conformance # every covered category parses with no ERROR/MISSING
npm run test:corpus      # both, in order

test/coverage.json is the coverage matrix. A category is covered when the grammar parses every one of its corpus inputs cleanly; otherwise it is listed in skip with a reason recording the unmodeled construct. Adding a new spec category (after bumping the submodule) fails test:coverage until it is classified, which is intentional: it forces a decision rather than silently dropping coverage.

Status

Initial grammar. It is intended for editor support and structural parsing. The Carve conformance corpus remains the source of truth for renderer behavior.