Parser Runtime

src/parser/ implements the loop in parseLoop (parser.zig):

1. Reuse probe (tryReuse): if no lookahead is held, lex one token and install it (advancing the tokenizer — the single-lexing invariant). Scan the monotonic candidate cursor for the largest span starting at the token with a matching start state and a valid shift/goto target; clone it, push, jump. Successful probes continue.
2. Action dispatch on (state, lookahead): accept (attach leftovers, return root), shift (token node + push), reduce (pop N, drain in-span errors, build node, goto, push), or recover.
3. Termination: every iteration consumes input, shrinks the stack, or pushes new state; recovery budgets bound the rest; a step guard (8n + 1024) backstops pathological grammars into finalize, which always returns a (possibly error-marked) root.

ParseState (state.zig) carries the tokenizer, stack, installed lookahead, error span, pending error nodes, and the detected start symbol (the non-terminal whose goto reaches the accept state).

Related pages

Parser, Incremental Runtime, Stack Runtime