A safe, efficient and simple QuickCheck-inspired library to generate shrinkable random data mainly oriented towards generative/property/exploratory testing.

In Brief

The purpose of the library is to test general properties of a program rather than very specific examples as you would with unit tests.

• When writing a checkito test (called a check), you first construct a generator by specifying the bounds that make sense for the inputs (ex: a number in the range 10..100, an alpha-numeric string, a vector of f64, etc.).
• Generators can produce arbitrary complex values with its combinators in a similar way that Iterators can.
• Given a proper generator, checkito will sample the input space to find a failing case for your test.
• Once a failing case is found, checkito will try to reduce the input to the simplest version of it that continues to fail (using a kind of binary search of the input space) to make the debugging process much easier.
• Note that checkito does not guarantee any kind of exhaustive search of the input space (the size of it gets out of hand rather quickly) and is meant as a complement to other testing strategies.
• It is recommended to write a regular unit test with the exact failing input to prevent a regression and to truly guarantee that the failing input is always tested.

Main Traits

• Generate: is implemented for many of rust's standard types and allows the generation of any random composite/structured data through combinator (such as tuples, Any, Map, Flatten and more). It is designed for composability and its usage should feel like working with Iterators.
• Shrink: tries to reduce a generated sample to a 'smaller' version of it while maintaining its constraints (ex: a sample usize in the range 10..100 will never be shrunk below 10). For numbers, it means bringing the sample closer to 0, for vectors, it means removing irrelevant items and shrinking the remaining ones, and so on.
• Prove: represents a desirable property of a program under test. It is used mainly in the context of the Check::check or Checker::check methods and it is the failure of a proof that triggers the shrinking process. It is implemented for a couple of standard types such as (), bool and Result. A panic!() is also considered as a failing property, thus standard assert!() macros (or any other panicking assertions) can be used to check the property.

To ensure safety, this library is #![forbid(unsafe_code)].

Cheat Sheet

use checkito::*;

/// The `#[check]` attribute is designed to be as thin as possible and
/// everything that is expressible with it is also ergonomically expressible as
/// _regular_ code (see below). Each `#[check]` attribute expands to a single
/// function call.
///
/// An empty `#[check]` attribute acts just like `#[test]`. It is allowed for
/// consistency between tests.
#[check]
fn empty() {}

/// The builtin `letter()` generator will yield ascii letters.
///
/// This test will be run many times with different generated values to find a
/// failing input.
#[check(letter())]
fn is_letter(value: char) {
    assert!(value.is_ascii_alphabetic());
}

/// Ranges can be used as generators and will yield values within its bounds.
///
/// A [`bool`] can be returned and if `true`, it will be considered as evidence
/// that the property under test holds.
#[check(0usize..=100)]
fn is_in_range(value: usize) -> bool {
    value <= 100
}

/// Regexes can be used and validated either dynamically using the [`regex`]
/// generator or at compile-time with the [`regex!`] macro.
///
/// Usual panicking assertions can be used in the body of the checking function
/// since a panic is considered a failed property.
#[check(regex("{", None).ok(), regex!("[a-zA-Z0-9_]*"))]
fn is_ascii(invalid: Option<String>, valid: String) {
    assert!(invalid.is_none());
    assert!(valid.is_ascii());
}

/// The `_` and `..` operators can be used to infer the [`FullGenerate`]
/// generator implementation for a type. Specifically, the `..` operator works
/// the same way as slice match patterns.
///
/// Since this test will panic, `#[should_panic]` can be used in the usual way.
#[check(..)]
#[check(_, _, _, _)]
#[check(negative::<f64>(), ..)]
#[check(.., negative::<i16>())]
#[check(_, .., _)]
#[check(negative::<f64>(), _, .., _, negative::<i16>())]
#[should_panic]
fn is_negative(first: f64, second: i8, third: isize, fourth: i16) {
    assert!(first < 0.0);
    assert!(second < 0);
    assert!(third < 0);
    assert!(fourth < 0);
}

/// `color = false` disables coloring of the output.
/// `verbose = true` will display all the steps taken by the [`check::Checker`]
/// while generating and shrinking values.
///
/// The shrinking process is pretty good at finding minimal inputs to reproduce
/// a failing property and in this case, it will always shrink values over
/// `1000` to exactly `1000`.
#[check(0u64..1_000_000, color = false, verbose = true)]
#[should_panic]
fn is_small(value: u64) {
    assert!(value < 1000);
}

/// Multiple checks can be performed.
///
/// If all generators always yield the same value, the check becomes a
/// parameterized unit test and will run only once.
#[check(3001, 6000)]
#[check(4500, 4501)]
#[check(9000, 1)]
fn sums_to_9001(left: i32, right: i32) {
    assert_eq!(left + right, 9001);
}

/// Generics can be used as inputs to the checking function.
///
/// [`Generate::map`] can be used to map a value to another.
#[check(111119)]
#[check(Generate::map(10..1000, |value| value * 10 - 1))]
#[check("a string that ends with 9")]
#[check(regex!("[a-z]*9"))]
fn ends_with_9(value: impl std::fmt::Display) -> bool {
    format!("{value}").ends_with('9')
}

pub struct Person {
    pub name: String,
    pub age: usize,
}
/// Use tuples to combine generators and build more complex structured types.
/// Alternatively implement the [`FullGenerate`] trait for the [`Person`]
/// struct.
///
/// Any generator combinator can be used here; see the other examples in the
/// _examples_ folder for more details.
///
/// Disable `debug` if a generated type does not implement [`Debug`] which
/// removes the only requirement that `#[check]` requires from input types.
#[check((letter().collect(), 18usize..=100).map(|(name, age)| Person { name, age }), debug = false)]
fn person_has_valid_name_and_is_major(person: Person) {
    assert!(person.name.is_ascii());
    assert!(person.age >= 18);
}

/// The `#[check]` attribute essentially expands to a call to [`Check::check`]
/// with pretty printing. For some more complex scenarios, it may become more
/// convenient to simply call the [`Check::check`] manually.
///
/// The [`Generate::any`] combinator chooses from its inputs. The produced
/// `Or<..>` preserves the information about the choice but here, it can be
/// simply collapsed using [`Generate::unify<T>`].
#[test]
fn has_even_hundred() {
    (0..100, 200..300, 400..500)
        .any()
        .unify::<i32>()
        .check(|value| assert!((value / 100) % 2 == 0));
}

fn main() {
    // `checkito` comes with a bunch of builtin generators such as this generic
    // number generator. An array of generators will produce an array of values.
    let generator = [(); 10].map(|_| number::<f64>());

    // For more configuration and control over the generation and shrinking
    // processes, retrieve a [`check::Checker`] from any generator.
    let mut checker = generator.checker();
    checker.generate.count = 1_000_000;
    checker.shrink.items = false;

    // [`check::Checker::checks`] produces an iterator of [`check::Result`] which
    // hold rich information about what happened during each check.
    for result in checker.checks(|values| values.iter().sum::<f64>() < 1000.0) {
        match result {
            check::Result::Pass(_pass) => {}
            check::Result::Shrink(_pass) => {}
            check::Result::Shrunk(_fail) => {}
            check::Result::Fail(_fail) => {}
        }
    }

    // For simply sampling random values from a generator, use [`Sample::samples`].
    // Just like in the checking process, samples will get increasingly larger.
    for _sample in generator.samples(1000) {}
}

See the examples and tests folder for more detailed examples.

Contribute

• If you find a bug or have a feature request, please open an issues.
• checkito is actively maintained and pull requests are welcome.
• If checkito was useful to you, please consider leaving a star!

Alternatives

• proptest
• quickcheck
• arbitrary
• monkey_test