Architecture

                    tree-sitter.zig
                           │
        ┌──────────────────┼──────────────────┐
        │                  │                  │
      Parser             Tree              Query
        │                  │                  │
   ┌────┼────┐       ┌─────┼─────┐       ┌────┼────┐
   │    │    │       │     │     │       │    │    │
 Lexer Stack Actions Node Cursor Edit  Pattern Capture Matcher
   │
 Input
   │
 Unicode
   │
 Language

In plain language

Source text flows downward: input bytes are tokenized by the lexer, assembled into a tree by the parser using language tables, and then read through nodes, cursors, or queries. Edits flow back up: an edit plus an old tree reuses subtrees to build a new tree cheaply.

Dependency direction

Low-level code never imports high-level code:

utils → core → memory → input/unicode → language → tree → lexer → parser → query

core (points, ranges, edits, symbols) knows nothing about parsing; utils is pure helpers over the standard library. Each subsystem exposes a facade (parser/parser.zig, tree/tree.zig, …) re-exported by src/treesitter.zig.

Related pages

• Development: Architecture, Development: Source Layout
• Parser, Memory Model