Performance and Benchmarks

Benchmarks live in benchmarks/benchmark_validation.py. They compare normal and compiled validation without external dependencies.

python benchmarks/benchmark_validation.py
python benchmarks/benchmark_validation.py --json

Measured scenarios

• flat_basic: Flat schema with core validators
• nested_payload: Nested dictionaries and lists
• collect_errors: Multiple error collection mode
• class_schema: Class-style schema

How to read results

speedup greater than 1 means compiled validation is faster. Specialized validators and custom callbacks reduce the gap.

What compiled validation optimizes

compile(schema) uses separate generated functions for normal validation and collect_errors=True. Compile a schema once and reuse it on hot paths.

collect_errors=True creates multiple ErrorDetail objects, so the speedup is usually smaller than normal validation. Prefer normal validation for high-volume valid payloads, and use collect_errors=True when callers need a full error list.