Parser

Simple explanation

The parser reads the source left to right and progressively builds a structured tree. At each step it asks one question: given everything parsed so far (summarized as a state) and the next token (lookahead), should it shift the token onto its stack, reduce the stack top into a syntax node, accept the finished tree, or recover from an error?

Technical explanation

parser/parser.zig implements a table-driven LR loop over language/tables.zig data:

1. Lookahead (parser/lookahead.zig): longest-match tokenization with extras skipped, one token of lookahead held in ParseState.
2. Shift: create a token subtree, push (target_state, node).
3. Reduce: pop N values, drain any pending error nodes inside the span, build the interior node, push the goto target.
4. Accept: the end token in an accept state finishes the parse; leftover error nodes attach to the root positionally.
5. Recovery (parser/recover.zig): skip bad input into ERROR spans, insert MISSING tokens at end of input, or unwind to a final error-marked root — always terminating via bounded budgets.

Reductions record each node's LR start state (reuse_state), which is what later makes incremental reuse sound.

Related pages

• Lexer, Parser Stack, Parse Actions, Reductions
• Parser Runtime, Parser API