Property-Based Testing
Instead of example inputs, state the invariant that must hold for ALL inputs — then let generated cases hunt for the violation.
From examples to properties
An example test: sort({3,1,2}) == {1,2,3}. A property: for every vector, sorting yields the same multiset, non-decreasing. One property is worth a hundred examples because it covers the inputs you forgot. Classic C++ properties:
- Round-trip: parse(serialize(x)) == x (parsers, codecs, string formatting).
- Involution: compress(decompress(x)) correctness,
reverse(reverse(v)) == v. - Idempotence:
sort(sort(v)) == sort(v), dedup applied twice is applied once. - Oracles: compare against a slow-but-obviously-correct implementation.
Random needs structure
Pure random bytes rarely reach interesting code. Generate structured randoms: valid-ish inputs with occasional boundary values (0, 1, INT_MAX, empty). Deterministic seeds keep CI reproducible; the failing case is printed and shrunk to the minimal witness.
Properties as documentation
A property states the contract so precisely that reading the test is reading the spec — "for all n >= 0: buildRange(n).size() == n" teaches the requirement better than prose. The graded exercises here grade property implementations: your checker function must accept correct implementations and reject subtly broken ones.