Error Recovery

Simple explanation

Real-world code is often broken — half-typed, mid-edit, or mid-paste. Instead of failing, the parser marks the broken parts (ERROR for skipped text, MISSING for absent tokens) and keeps building a useful tree around them.

Technical explanation

parser/recover.zig defines the budgets (max_recovery_ops, max_missing_inserts) and costs (per skipped byte, per insertion). The engine, on a missing table action:

1. At end of input, inserts a MISSING token — preferring a symbol whose shift target itself handles end-of-input (this closes open parens correctly) — and continues.
2. Otherwise skips the offending token or byte into an open error span, which closes into an ERROR leaf attached to the next enclosing reduction or, at worst, the root.
3. Gives up gracefully into an error-marked root wrapping all fragments when budgets exhaust.

Because recovery only ever adds information (error flags, extra children), valid regions of broken files still produce exactly the nodes a fresh parse of the fixed file would.

Related pages

• Error Recovery guide, Errors API
• Upstream Conformance