Language Support

Differens parses code with tree-sitter and diffs the resulting trees. Any language with a tree-sitter grammar gets structural diffing. The real question is how much semantics a language gets: Differens either recognizes a node as a Function or a Class, or it just sees a tree-sitter node type like function_item.

Two levels of support

Semantic extractors (T5, L6). Four languages ship a per-language extractor: TypeScript/JavaScript, Python, Rust, and Go. An extractor maps tree-sitter node types to canonical concepts like Function, Class, Method, Import, Export, Variable, Type, Enum, Struct, and Trait, then pulls the declaration name out of the tree. With an extractor, a rename is a rename and a move is a move. Narration can say “renamed computeTotal to calculateTotalAmount”. Generic structural diff (T5, L5). Every other language with a tree-sitter grammar still gets structural diffing. The CST is parsed into a tree and matched structurally, but nodes keep their raw tree-sitter type names instead of canonical concepts, and labels come from the generic name-field rule. Reformatting and restructure are caught. Semantic narration is not.

Languages with full extractors

Language	Tier	Extractor	What it detects
TypeScript / JavaScript	T5 (L6)	`TypeScriptExtractor`	functions, arrow functions, methods, classes, interfaces, type aliases, enums, imports, exports, variables, JSX, control flow
Python	T5 (L6)	`PythonExtractor`	functions (including async), classes, imports, decorators, assignments, comprehensions, type annotations, control flow
Rust	T5 (L6)	`RustExtractor`	functions, structs, enums, traits, impls, macros, `use` imports, modules, types
Go	T5 (L6)	`GoExtractor`	functions, methods, structs, interfaces, imports, variables, constants, goroutines
Any other tree-sitter grammar	T5 (L5)	n/a	structural diff with raw tree-sitter node type labels

Anything tree-sitter has a grammar for falls back to the generic structural level (T5, L5). No extractor needed. If no grammar is registered at all, the file falls down the tier ladder to a line diff.

The LanguageExtractor interface

An extractor is a few dozen match arms, not a parser:

export interface LanguageExtractor {
  /** Language identifier, e.g. "typescript" */
  readonly language: string;
  /** File extensions this extractor handles */
  readonly extensions: string[];
  /** Map a tree-sitter node type to a canonical concept */
  extractConcept(nodeType: string): string;
  /** Extract the label (name) from a node if it has one */
  extractLabel(node: Parser.SyntaxNode, source: string): string | undefined;
  /** Node types whose label does NOT come from a plain `name` field */
  readonly labelFallbackTypes?: ReadonlySet<string>;
}

The tree-sitter cursor walk reads the name field directly for every node. extractLabel only runs for the types listed in labelFallbackTypes. That is a set lookup instead of a childForFieldName call per node.

Adding a language

To give a language semantic diffs, implement LanguageExtractor and register the extractor with the code tier. See Adding a language extractor in the contributing guide.

Semantic Extractors

Tier Adapters

Language Support

Language Support

Two levels of support

Languages with full extractors

The LanguageExtractor interface

Adding a language

Language pages

​Language Support

​Two levels of support

​Languages with full extractors

​The LanguageExtractor interface

​Adding a language

​Language pages

Language Support

Two levels of support

Languages with full extractors

The LanguageExtractor interface

Adding a language

Language pages