This requires a Tree-sitter grammar for each supported language, along with a set of Tree-sitter queries that describe how to use that grammar's syntax tree for tasks like syntax highlighting, auto-indent, etc. The hard part about letting extensions add Tree-sitter parsers to Zed is that Tree-sitter parsers are expressed as C code. This division between the two phases enables a unique architecture in which we load a parser from a wasm file, but most of the static data in that wasm file is copied out of the wasm linear memory into a native data structure.