property-based-testing

Use the Hypothesis library with the @given decorator and strategies from hypothesis.strategies to generate inputs. Define properties like roundtrip (decode(encode(x)) == x) or idempotence (f(f(x)) == f(x)) rather than specific example values.

Hypothesis is the property-based testing library for Python, while fast-check serves JavaScript and TypeScript. Both provide input generators and automatic shrinking of failing cases, with equivalent concepts like strategies and arbitraries.

Yes, Echidna and Medusa are property-based fuzzers for EVM contracts that test state invariants. You define functions like echidna_balance_invariant that must always return true, and the fuzzer attempts to violate them.

Avoid PBT for simple CRUD operations without transformation logic, UI code, integration tests with complex external setup, and one-off scripts. It works best for unit-level pure functions, serializers, validators, and normalizers.

Vacuous tests usually result from contradictory or overly restrictive assume() calls that filter out nearly all generated inputs. Fix this by building constraints directly into the input strategy instead of filtering after generation.

First reproduce the shrunk minimal failing example, then verify the property against docstrings, type hints, and documented preconditions. Classify the failure as a test bug, ambiguous specification, or genuine bug before reporting it.