treesitter-indent

Tree-sitter grammar for Indent Markup Language used by diagram-app (.inml files).

Grammar shape

The grammar is intentionally generic -- there is no special-cased "edge"/"reference"/"include" concept at this layer, matching the hand-written parser in indent-parser:

document  := NEWLINE? statement*
statement := type attribute* NEWLINE block?
block     := INDENT statement+ DEDENT
attribute := attribute_name "=" value
value     := string | number | boolean

• type is any bareword up to the first space (e.g. org, service, ->, =>, include -- all just ordinary keywords; meaning is assigned entirely by downstream consumers).
• Indentation is 4 spaces per level, tracked via external NEWLINE / INDENT / DEDENT tokens (see src/scanner.c), since indentation sensitivity isn't context-free.
• A trailing \ joins the current line with the next physical line; this is modeled as an invisible extras token in grammar.js rather than scanner logic (see the comment at the top of src/scanner.c for why).

Building

npm install         # also runs `node-gyp-build` to compile the native binding
tree-sitter generate

Testing

The corpus test suite in test/corpus/statements.txt is the primary correctness check:

tree-sitter test

It covers: attributes (quoted strings, unquoted numbers/booleans), single- and multi-level nesting, multi-level dedent back to a shallower depth, full-line comments and blank lines, backslash line-continuation, bareword kinds like ->, and EOF without a trailing newline.

You can also sanity-check against real fixtures directly:

tree-sitter parse ../indent-parser/test/fixtures/sample.inml
tree-sitter parse ../test.inml

Both should produce zero ERROR nodes.

Highlighting

queries/highlights.scm provides syntax-highlighting captures for editors that support Tree-sitter directly (Neovim, Helix, Zed, etc.). VS Code does not consume Tree-sitter grammars natively -- see the sibling vscode-indent-lang package for a plain TextMate grammar instead.