karet-treesitter

Shared tree-sitter parse host (incremental parsing, tree cache, query running) for karet.

A single, reusable tree-sitter front end — parser pooling, incremental re-parsing, and query execution — so syntax highlighting, diffing and structural navigation all share one parse of each buffer. Tree-sitter is the sole syntax backend, by design.

It also resolves language injection: LayeredParser parses a document as a root tree plus one tree per embedded region, so a markdown code fence is parsed as the language it names, an HTML <script> as JavaScript, and a Rust doc comment as markdown (whose own ```rust fences then nest back into Rust). Injected layers are parsed over the shared source with set_included_ranges, so every node already carries document byte offsets and a consumer can merge their queries with no coordinate translation.

The grammar registry also exposes semantic-structure and outline queries for code, Markdown, markup, shell, structured-data, query, and schema grammars. Grouped query matches keep declarations, names, scopes, headers, and bodies associated so higher-level engines can derive navigation and sticky-scroll models from the same retained parse.

Part of the karet workspace.

License

Licensed under either of MIT or Apache-2.0 at your option.