퍼징과 불변식

Unit tests check specific inputs you thought of. But bugs often hide in inputs you did not think of. Property-based testing flips this: you state a property that should always hold, and the tool generates many random inputs trying to break it.

Foundry supports two flavors: fuzz tests and invariant tests.

A fuzz test is simply a test function with parameters. Foundry automatically calls it many times with randomized argument values, hunting for a counterexample.

Random inputs may be unrealistic. Constrain them with:

• vm.assume(cond) — discard runs that fail the condition
• bound(x, min, max) — map any value into a range

Prefer bound for ranges since assume can waste runs by rejecting too many inputs.

A good fuzz property is a statement that must hold for all valid inputs. Examples:

• Depositing then withdrawing returns the same amount
• Total supply never changes on a transfer
• A user can never withdraw more than they deposited

Think in terms of universal truths, not specific values.

When a fuzz test fails, Foundry prints the exact counterexample that broke the property and the number of runs. It also stores a corpus so the failing input is replayed on future runs until you fix it.

Control fuzzing in foundry.toml. The runs setting is how many random inputs each fuzz test gets. More runs increase confidence but take longer.

Invariants go a step further. Instead of one function call, Foundry executes long random sequences of calls to your contract, then checks that a property still holds after every sequence.

Invariant functions are named with the invariant_ prefix.

For meaningful invariant tests you usually write a handler contract that exposes a curated set of actions. The fuzzer calls the handler's functions in random order, keeping the call sequences valid and focused.

Register target contracts with targetContract in setUp().

Handlers often track ghost variables: bookkeeping totals updated as actions run. Invariants compare the contract's real state against these ghost values.

For example, summing every deposit in the handler and asserting it equals the vault's total assets catches accounting drift.

Knowing which to reach for:

• Fuzz tests a single function against random arguments — good for input validation and pure logic
• Invariant tests random sequences of actions — good for stateful systems like vaults, AMMs, and token accounting

Use both: fuzz for unit-level properties, invariants for system-level guarantees.

Effective property testing:

• State properties as universal truths, not examples
• Use bound over heavy assume
• Increase runs for critical contracts
• Write focused handlers for invariants
• Track ghost variables for accounting checks

Property tests catch the edge cases humans miss.

You learned property-based testing:

• Fuzz tests take parameters; Foundry generates random inputs
• Use bound and vm.assume to constrain inputs
• Invariant tests run random call sequences via handlers
• Ghost variables track expected state for accounting invariants
• Counterexamples are saved and replayed

Next: the cast and anvil CLI tools.