Property-Based Testing

Instead of testing specific examples, define properties that must hold for ALL inputs. The framework generates hundreds of random inputs and finds the smallest failing case.

When to Use PBT

Use Case	Property
Serialization roundtrip	deserialize(serialize(x)) === x
Sort function	Output is ordered AND contains same elements
Parser	Never crashes on any input
Encoder/decoder	decode(encode(x)) === x
State machine	Invariants hold after any sequence of operations
Math/financial	Associativity, commutativity, identity, bounds
API handler	Never returns 500 on valid input
Data transformation	Output schema matches specification

When NOT to Use PBT

• UI rendering tests (use visual regression)
• Integration tests with external services (use contract tests)
• Tests that need specific business scenarios (use example tests)
• Tests where the oracle is as complex as the implementation

fast-check (JavaScript/TypeScript)

Setup

npm install --save-dev fast-check

Basic Property

import fc from 'fast-check'

// Property: sorting is idempotent
test('sort is idempotent', () => {
  fc.assert(
    fc.property(fc.array(fc.integer()), (arr) => {
      const sorted = [...arr].sort((a, b) => a - b)
      const sortedTwice = [...sorted].sort((a, b) => a - b)
      expect(sorted).toEqual(sortedTwice)
    })
  )
})

// Property: serialization roundtrip
test('JSON roundtrip preserves data', () => {
  fc.assert(
    fc.property(fc.jsonValue(), (value) => {
      expect(JSON.parse(JSON.stringify(value))).toEqual(value)
    })
  )
})

Custom Arbitraries

// Generate valid email addresses
const emailArb = fc.tuple(
  fc.stringOf(fc.constantFrom(...'abcdefghijklmnopqrstuvwxyz0123456789'.split('')), { minLength: 1 }),
  fc.constantFrom('gmail.com', 'example.com', 'test.org')
).map(([local, domain]) => `${local}@${domain}`)

// Generate valid user objects
const userArb = fc.record({
  id: fc.uuid(),
  name: fc.string({ minLength: 1, maxLength: 100 }),
  email: emailArb,
  age: fc.integer({ min: 0, max: 150 }),
  role: fc.constantFrom('admin', 'user', 'viewer')
})

// Generate valid but adversarial strings
const adversarialStringArb = fc.oneof(
  fc.constant(''),
  fc.constant(' '),
  fc.constant('\0'),
  fc.constant('<script>alert(1)</script>'),
  fc.constant("Robert'); DROP TABLE users;--"),
  fc.constant('../../../etc/passwd'),
  fc.unicodeString(),
  fc.string({ minLength: 10000, maxLength: 100000 })  // Very long
)

Stateful Testing (Model-Based)

// Test a cache against a simple Map model
class CacheModel {
  private model = new Map<string, string>()

  set(key: string, value: string): void { this.model.set(key, value) }
  get(key: string): string | undefined { return this.model.get(key) }
  delete(key: string): void { this.model.delete(key) }
  size(): number { return this.model.size }
}

const cacheCommands = [
  fc.tuple(fc.string(), fc.string()).map(([k, v]) => ({
    check: (model: CacheModel) => true,
    run: (model: CacheModel, real: Cache) => {
      model.set(k, v)
      real.set(k, v)
      expect(real.get(k)).toBe(model.get(k))
    },
    toString: () => `set(${k}, ${v})`
  })),
  fc.string().map((k) => ({
    check: (model: CacheModel) => true,
    run: (model: CacheModel, real: Cache) => {
      model.delete(k)
      real.delete(k)
      expect(real.get(k)).toBe(model.get(k))
    },
    toString: () => `delete(${k})`
  }))
]

test('cache behaves like Map', () => {
  fc.assert(
    fc.property(fc.commands(cacheCommands), (cmds) => {
      const model = new CacheModel()
      const real = new Cache()
      fc.modelRun(() => ({ model, real }), cmds)
    })
  )
})

Hypothesis (Python)

Setup

pip install hypothesis

Basic Properties

from hypothesis import given, strategies as st, settings

@given(st.lists(st.integers()))
def test_sort_preserves_length(xs):
    assert len(sorted(xs)) == len(xs)

@given(st.lists(st.integers()))
def test_sort_preserves_elements(xs):
    assert sorted(sorted(xs)) == sorted(xs)

@given(st.text())
def test_encode_decode_roundtrip(s):
    assert s.encode('utf-8').decode('utf-8') == s

# With settings
@settings(max_examples=1000, deadline=None)
@given(st.dictionaries(st.text(), st.integers()))
def test_dict_operations(d):
    import json
    assert json.loads(json.dumps(d)) == d

Custom Strategies

from hypothesis import strategies as st

# Valid email strategy
emails = st.builds(
    lambda local, domain: f"{local}@{domain}",
    local=st.from_regex(r'[a-z0-9]{1,20}', fullmatch=True),
    domain=st.sampled_from(['gmail.com', 'example.com'])
)

# Valid user strategy
users = st.fixed_dictionaries({
    'name': st.text(min_size=1, max_size=100),
    'email': emails,
    'age': st.integers(min_value=0, max_value=150),
    'role': st.sampled_from(['admin', 'user', 'viewer'])
})

gopter (Go)

Setup

go get github.com/leanovate/gopter

Basic Property

func TestSortIdempotent(t *testing.T) {
    properties := gopter.NewProperties(gopter.DefaultTestParameters())

    properties.Property("sort is idempotent", prop.ForAll(
        func(xs []int) bool {
            sorted := make([]int, len(xs))
            copy(sorted, xs)
            sort.Ints(sorted)

            sortedTwice := make([]int, len(sorted))
            copy(sortedTwice, sorted)
            sort.Ints(sortedTwice)

            return reflect.DeepEqual(sorted, sortedTwice)
        },
        gen.SliceOf(gen.Int()),
    ))

    properties.TestingRun(t)
}

Property Catalog

Algebraic Properties

Property	Definition	Example
Identity	f(x, identity) === x	add(x, 0) === x
Commutativity	f(a, b) === f(b, a)	add(a, b) === add(b, a)
Associativity	f(f(a, b), c) === f(a, f(b, c))	add(add(a, b), c) === add(a, add(b, c))
Idempotency	f(f(x)) === f(x)	sort(sort(xs)) === sort(xs)
Roundtrip	g(f(x)) === x	decode(encode(x)) === x
Invariant	property(f(x)) === true	length(sort(xs)) === length(xs)

Safety Properties

Property	Check
No crash	Function never throws for any valid input
Bounded output	Output size is proportional to input size
No mutation	Input is not modified by the function
Deterministic	Same input always produces same output
Monotonic	If a <= b then f(a) <= f(b)

Shrinking

When a property fails, the framework automatically shrinks the failing input to the smallest case that still fails:

Original failing input: [482, -1, 0, 99, -384, 7, 42, 0, -1]
Shrunk to: [1, 0]

This tells you the bug is about: handling zero in a list with other elements

Tips:

• Custom arbitraries should define custom shrinkers
• If shrinking takes too long, limit with { endOnFailure: true }
• Shrunk examples make great regression tests

Integration with vibecosystem

• tdd-guide agent: Recommend PBT for pure functions and serialization
• qa-engineer agent: Use PBT for edge case discovery
• arbiter agent: Run PBT suites as part of test validation
• mocksmith agent: Generate test data using PBT arbitraries

Inspired by Trail of Bits property-based-testing plugin.