QuickCheck for Swift.
For those already familiar with the Haskell library, check out the source. For everybody else, see the Tutorial Playground for a beginner-level introduction to the major concepts and use-cases of this library.
SwiftCheck is a testing library that automatically generates random data for
testing of program properties. A property is a particular facet of an algorithm
or data structure that must be invariant under a given set of input data,
basically an XCTAssert
on steroids. Where before all we could do was define
methods prefixed by test
and assert, SwiftCheck allows program properties and
tests to be treated like data.
To define a program property the forAll
quantifier is used with a type
signature like (A, B, C, ... Z) -> Testable where A : Arbitrary, B : Arbitrary ... Z : Arbitrary
. SwiftCheck implements the Arbitrary
protocol for most Swift
Standard Library types and implements the Testable
protocol for Bool
and
several other related types. For example, if we wanted to test the property
that every Integer is equal to itself, we would express it as such:
func testAll() {
// 'property' notation allows us to name our tests. This becomes important
// when they fail and SwiftCheck reports it in the console.
property("Integer Equality is Reflexive") <- forAll { (i : Int) in
return i == i
}
}
For a less contrived example, here is a program property that tests whether Array identity holds under double reversal:
property("The reverse of the reverse of an array is that array") <- forAll { (xs : [Int]) in
// This property is using a number of SwiftCheck's more interesting
// features. `^&&^` is the conjunction operator for properties that turns
// both properties into a larger property that only holds when both sub-properties
// hold. `<?>` is the labelling operator allowing us to name each sub-part
// in output generated by SwiftCheck. For example, this property reports:
//
// *** Passed 100 tests
// (100% , Right identity, Left identity)
return
(xs.reversed().reversed() == xs) <?> "Left identity"
^&&^
(xs == xs.reversed().reversed()) <?> "Right identity"
}
Because SwiftCheck doesn't require tests to return Bool
, just Testable
, we
can produce tests for complex properties with ease:
property("Shrunken lists of integers always contain [] or [0]") <- forAll { (l : [Int]) in
// Here we use the Implication Operator `==>` to define a precondition for
// this test. If the precondition fails the test is discarded. If it holds
// the test proceeds.
return (!l.isEmpty && l != [0]) ==> {
let ls = self.shrinkArbitrary(l)
return (ls.filter({ $0 == [] || $0 == [0] }).count >= 1)
}
}
Properties can even depend on other properties:
property("Gen.one(of:) multiple generators picks only given generators") <- forAll { (n1 : Int, n2 : Int) in
let g1 = Gen.pure(n1)
let g2 = Gen.pure(n2)
// Here we give `forAll` an explicit generator. Before SwiftCheck was using
// the types of variables involved in the property to create an implicit
// Generator behind the scenes.
return forAll(Gen.one(of: [g1, g2])) { $0 == n1 || $0 == n2 }
}
All you have to figure out is what to test. SwiftCheck will handle the rest.
What makes QuickCheck unique is the notion of shrinking test cases. When fuzz testing with arbitrary data, rather than simply halt on a failing test, SwiftCheck will begin whittling the data that causes the test to fail down to a minimal counterexample.
For example, the following function uses the Sieve of Eratosthenes to generate a list of primes less than some n:
/// The Sieve of Eratosthenes:
///
/// To find all the prime numbers less than or equal to a given integer n:
/// - let l = [2...n]
/// - let p = 2
/// - for i in [(2 * p) through n by p] {
/// mark l[i]
/// }
/// - Remaining indices of unmarked numbers are primes
func sieve(_ n : Int) -> [Int] {
if n <= 1 {
return []
}
var marked : [Bool] = (0...n).map { _ in false }
marked[0] = true
marked[1] = true
for p in 2..<n {
for i in stride(from: 2 * p, to: n, by: p) {
marked[i] = true
}
}
var primes : [Int] = []
for (t, i) in zip(marked, 0...n) {
if !t {
primes.append(i)
}
}
return primes
}
/// Short and sweet check if a number is prime by enumerating from 2...⌈√(x)⌉ and checking
/// for a nonzero modulus.
func isPrime(n : Int) -> Bool {
if n == 0 || n == 1 {
return false
} else if n == 2 {
return true
}
let max = Int(ceil(sqrt(Double(n))))
for i in 2...max {
if n % i == 0 {
return false
}
}
return true
}
We would like to test whether our sieve works properly, so we run it through SwiftCheck with the following property:
import SwiftCheck
property("All Prime") <- forAll { (n : Int) in
return sieve(n).filter(isPrime) == sieve(n)
}
Which produces the following in our testing log:
Test Case '-[SwiftCheckTests.PrimeSpec testAll]' started.
*** Failed! Falsifiable (after 10 tests):
4
Indicating that our sieve has failed on the input number 4. A quick look back at the comments describing the sieve reveals the mistake immediately:
- for i in stride(from: 2 * p, to: n, by: p) {
+ for i in stride(from: 2 * p, through: n, by: p) {
Running SwiftCheck again reports a successful sieve of all 100 random cases:
*** Passed 100 tests
SwiftCheck implements random generation for most of the types in the Swift
Standard Library. Any custom types that wish to take part in testing must
conform to the included Arbitrary
protocol. For the majority of types, this
means providing a custom means of generating random data and shrinking down to
an empty array.
For example:
import SwiftCheck
public struct ArbitraryFoo {
let x : Int
let y : Int
public var description : String {
return "Arbitrary Foo!"
}
}
extension ArbitraryFoo : Arbitrary {
public static var arbitrary : Gen<ArbitraryFoo> {
return Gen<(Int, Int)>.zip(Int.arbitrary, Int.arbitrary).map(ArbitraryFoo.init)
}
}
class SimpleSpec : XCTestCase {
func testAll() {
property("ArbitraryFoo Properties are Reflexive") <- forAll { (i : ArbitraryFoo) in
return i.x == i.x && i.y == i.y
}
}
}
There's also a Gen.compose
method which allows you to procedurally compose
values from multiple generators to construct instances of a type:
public static var arbitrary : Gen<MyClass> {
return Gen<MyClass>.compose { c in
return MyClass(
// Use the nullary method to get an `arbitrary` value.
a: c.generate(),
// or pass a custom generator
b: c.generate(Bool.suchThat { $0 == false }),
// .. and so on, for as many values and types as you need.
c: c.generate(), ...
)
}
}
Gen.compose
can also be used with types that can only be customized with setters:
public struct ArbitraryMutableFoo : Arbitrary {
var a: Int8
var b: Int16
public init() {
a = 0
b = 0
}
public static var arbitrary: Gen<ArbitraryMutableFoo> {
return Gen.compose { c in
var foo = ArbitraryMutableFoo()
foo.a = c.generate()
foo.b = c.generate()
return foo
}
}
}
For everything else, SwiftCheck defines a number of combinators to make working with custom generators as simple as possible:
let onlyEven = Int.arbitrary.suchThat { $0 % 2 == 0 }
let vowels = Gen.fromElements(of: [ "A", "E", "I", "O", "U" ])
let randomHexValue = Gen<UInt>.choose((0, 15))
let uppers = Gen<Character>.fromElements(in: "A"..."Z")
let lowers = Gen<Character>.fromElements(in: "a"..."z")
let numbers = Gen<Character>.fromElements(in: "0"..."9")
/// This generator will generate `.none` 1/4 of the time and an arbitrary
/// `.some` 3/4 of the time
let weightedOptionals = Gen<Int?>.frequency([
(1, Gen<Int?>.pure(nil)),
(3, Int.arbitrary.map(Optional.some))
])
For instances of many complex or "real world" generators, see
ComplexSpec.swift
.
SwiftCheck supports OS X 10.9+ and iOS 7.0+.
SwiftCheck can be included one of two ways:
Using The Swift Package Manager
Package.swift
file's dependencies section:.package(url: "https://github.com/typelift/SwiftCheck.git", from: "0.8.1")
Using Carthage
carthage update
Frameworks
Using CocoaPods
$ pod install
in your project directory.Framework
SwiftCheck is released under the MIT license.