Parser

Parsing tokens into a source-shaped AST.


Status	Active — §2.1 leaders + expressions (chumsky)
Owners	`echo_parser`, `echo_ast`
Library	chumsky 0.9 (ADR 0011)
Related	`docs/syntax.md`, `docs/lexer.md`, ADR 0003

Pipeline

echo_source → echo_lexer → echo_parser (chumsky) → echo_ast
                ↓
         echo_diagnostics
         echo_syntax (leader facts)

Hosts call echo_parser::parse (or xo ast); they do not reimplement grammar.

Input is tokens from echo_lexer (leaders already classified).
Output is echo_ast::File (source-shaped; no types / resolution).
CLI: xo ast [--kinds] [--diag-codes] <file>
Suite: e26 requires .ast via ast --kinds --diag-codes
No trailing commas (matches syntax.md); do not re-enable allow_trailing without a syntax change
Field assigns accept dotted chains: ~ a.b.c = e, ~ .a.b = e (AST AssignTarget::Field with nested Expr::Field base)
Index assigns: ~ xs[i] = e, ~ a.b[i] = e (AssignTarget::Index with index); list push: ~ xs[] = e, ~ a.b[] = e (AssignTarget::Index with index: None; field path before [ is allowed)
Edge cases must be justified (see AGENTS.md — Justified edge cases only)

Richer recovery strategies as the grammar grows (must not invent meaning)
Pratt combinator vs layered expr when expr surface expands
Multi-bind (~ a = 1, b = 2 / $ x = 1, y = 2) is locked and expanded to sequential binds after parse (no trailing comma)