Performance

rtfstruct uses streaming tokenization, list-backed text buffering, capped diagnostics, and parser safety limits to avoid common malformed-input blowups.

Guardrails

• ParserOptions.max_group_depth caps nested RTF groups.

• ParserOptions.max_document_chars caps emitted document text.

• ParserOptions.max_diagnostics caps retained diagnostics.

• Image payloads are captured only through explicit parser state and can be omitted from JSON output.

Benchmark

A dependency-free generated corpus benchmark is available:

python benchmarks/parse_generated.py --paragraphs 10000

The benchmark is intentionally simple. It is meant for local regression checks before adopting heavier benchmark tooling.