test-strategy

This crate provides two procedural macros, #[derive(Arbitrary)] and #[proptest].

Each of these macros is an alternative to the following proptest's official macros.

The macros provided by this crate have the following advantages over the proptest's official macros.

• Supports higher-order strategies. (#[derive(Arbitrary)] and #[proptest])
• Code formatting is not disabled. (#[proptest])

However, the syntax of this crate's macros are not compatible with the syntax of the official macros.

Install

Add this to your Cargo.toml:

[dependencies]
test-strategy = "0.4.0"
proptest = "1.5.0"

Example

You can use #[derive(Arbitrary)] to automatically implement proptest's Arbitrary trait.

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct TestInputStruct {
    x: u32,

    #[strategy(1..10u32)]
    y: u32,

    #[strategy(0..#y)]
    z: u32,
}

#[derive(Arbitrary, Debug)]
enum TestInputEnum {
    A,
    B,
    #[weight(3)]
    C,
    X(u32),
    Y(#[strategy(0..10u32)] u32),
}

You can define a property test by adding #[proptest] to the function.

use test_strategy::proptest;

#[proptest]
fn my_test(_x: u32, #[strategy(1..10u32)] y: u32, #[strategy(0..#y)] z: u32) {
    assert!(1 <= y && y < 10);
    assert!(z <= y);
}

Attributes

Attributes can be written in the following positions.

#[derive(Arbitrary)]

You can implement proptest::arbitrary::Arbitrary automatically by adding #[derive(Arbitrary)] to struct or enum declaration.

By default, all fields are set using the strategy obtained by proptest::arbitrary::any().

So the following two codes are equivalent.

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct TestInput {
    x: u32,
    y: u32,
}

use proptest::{
    arbitrary::{any, Arbitrary},
    strategy::{BoxedStrategy, Strategy},
};

#[derive(Debug)]
struct TestInput {
    x: u32,
    y: u32,
}
impl Arbitrary for TestInput {
    type Parameters = ();
    type Strategy = BoxedStrategy<Self>;

    fn arbitrary_with(_args: Self::Parameters) -> Self::Strategy {
        let x = any::<u32>();
        let y = any::<u32>();
        (x, y).prop_map(|(x, y)| Self { x, y }).boxed()
    }
}

#[strategy]

You can specify a strategy to generate values for the field by adding #[strategy(...)] to the field.

In the following example, the value of field x will be less than 20.

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct TestInput {
    #[strategy(0..20u32)]
    x: u32,
}

In #[strategy], the values of other fields can be used by following # to the name of the field.

In the following example, the value of y is less than or equal to x.

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct TestInput {
    x: u32,
    #[strategy(0..=#x)]
    y: u32,
}

#[any]

Instead of writing #[strategy(any_with::<Type>(expr))], you can write #[any(expr)].

use proptest::collection::size_range;
use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug, PartialEq)]
struct TestInput {
    #[any(size_range(0..16).lift())]
    x: Vec<u16>,
}

Instead of writing an expression to be passed to any_with, you can write only the value of the field to be changed from the default value.

Therefore, the following TestInputA, TestInputB and TestInputC are equivalent.

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct TestInputA {
    #[any(InnerArgs { upper : 20, ..InnerArgs::default() })]
    a: Inner,
}
#[derive(Arbitrary, Debug)]
struct TestInputB {
    #[any(InnerArgs::default(), upper = 20)]
    a: Inner,
}
#[derive(Arbitrary, Debug)]
struct TestInputC {
    #[any(upper = 20)]
    a: Inner,
}

#[derive(Default)]
struct InnerArgs {
    lower: i32,
    upper: i32,
}

#[derive(Arbitrary, Debug)]
#[arbitrary(args = InnerArgs)]
struct Inner {
    #[strategy(args.lower..args.upper)]
    x: i32,
}

#[weight]

By default, all variants appear with equal probability.

You can add #[weight] to the variant to change the probability of the variant appearing.

In the following example, TestInput::B is twice as likely to appear as TestInput::A.

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
enum TestInput {
    A,

    #[weight(2)]
    B,
}

If you add #[weight(0)] to a variant, the variant does not appear, so you can use a type in that variant that cannot be used as Arbitrary.

use test_strategy::Arbitrary;

#[derive(Debug)]
struct NotArbitrary;

#[derive(Arbitrary, Debug)]
enum TestInput {
    A,

    #[allow(dead_code)]
    #[weight(0)] // Removing this `#[weight(0)]` will cause a compile error.
    B(NotArbitrary),
}

#[map]

Instead of using prop_map in #[strategy(...)], #[map(...)] can be used.

The following codes mean the same thing.

use proptest::arbitrary::any;
use proptest::strategy::Strategy;
use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct TestInput1 {
    #[strategy(any::<u32>().prop_map(|x| x + 1))]
    x: u32,
}

#[derive(Arbitrary, Debug)]
struct TestInput2 {
    #[strategy(any::<u32>())]
    #[map(|x| x + 1)]
    x: u32,
}

#[derive(Arbitrary, Debug)]
struct TestInput3 {
    #[map(|x: u32| x + 1)]
    x: u32,
}

References to other fields in the function applied to prop_map or #[map(...)] will generate different strategies.

Referencing another field in #[strategy(...)] will expand it to prop_flat_map, even if it is in prop_map.

use proptest::arbitrary::any;
use proptest::strategy::{Just, Strategy};
use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct T1 {
    x: u32,

    #[strategy(any::<u32>().prop_map(move |y| #x + y))]
    y: u32,
}
// The code above generates the following strategy.
let t1 = any::<u32>()
    .prop_flat_map(|x| (Just(x), any::<u32>().prop_map(move |y| x + y)))
    .prop_map(|(x, y)| T1 { x, y });

On the other hand, if you refer to another field in #[map], it will expand to prop_map.

use proptest::arbitrary::any;
use proptest::strategy::Strategy;
use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct T2 {
    x: u32,

    #[map(|y: u32| #x + y)]
    y: u32,
}
// The code above generates the following strategy.
let t2 = (any::<u32>(), any::<u32>()).prop_map(|(x, y)| T2 { x, y });

If the input and output types of the function specified in #[map] are different, the value type of the strategy set in #[strategy] is the type of the function's input, not the type of the field.

use proptest::arbitrary::any;
use proptest::sample::Index;
use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct T1 {
    #[strategy(any::<Index>())]
    #[map(|i: Index| i.index(10))]
    x: usize,
}

// `#[strategy(any::<Index>())]` can be omitted.
#[derive(Arbitrary, Debug)]
struct T2 {
    #[map(|i: Index| i.index(10))]
    x: usize,
}

#[filter]

By adding #[filter] , you can limit the values generated.

In the following examples, x is an even number.

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct TestInput {
    #[filter(#x % 2 == 0)]
    x: u32,
}

You can also use multiple variables in a predicate.

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
#[filter((#x + #y) % 2 == 0)]
struct T1 {
    x: u32,
    y: u32,
}

#[derive(Arbitrary, Debug)]
struct T2 {
    x: u32,
    #[filter((#x + #y) % 2 == 0)]
    y: u32,
}

You can use the value of a structure or enum in the filter by using #self.

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
#[filter((#self.x + #self.y) % 2 == 0)]
struct TestInput {
    x: u32,
    y: u32,
}

If the expression specified for #[filter] does not contain a variable named by appending # to its own field name, the expression is treated as a predicate function, rather than an expression that returns a bool.

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct TestInput {
    #[filter(is_even)]
    x: u32,
}
fn is_even(x: &u32) -> bool {
    x % 2 == 0
}

#[derive(Arbitrary, Debug)]
struct T2 {
    a: u32,

    // Since `#a` exists but `#b` does not, it is treated as a predicate function.
    #[filter(|&x| x > #a)]
    b: u32,
}

Similarly, an expression that does not contain #self in the #[filter(...)] that it attaches to a type is treated as a predicate function.

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
#[filter(is_even)]
struct T {
    x: u32,
}
fn is_even(t: &T) -> bool {
    t.x % 2 == 0
}

You can specify a filter name by passing two arguments to #[filter].

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct TestInput {
    #[filter("x is even", #x % 2 == 0)]
    x: u32,
}

#[by_ref]

By default, if you use a variable with #[strategy], #[any], #[map] or #[filter] with # attached to it, the cloned value is set.

Adding #[by_ref] to the field makes it use the reference instead of the cloned value.

use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct TestInput {
    #[by_ref]
    #[strategy(1..10u32)]
    x: u32,

    #[strategy(0..*#x)]
    y: u32,
}

#[arbitrary]

#[arbitrary(args = T)]

Specifies the type of Arbitrary::Parameters.

You can use the Rc value of this type in #[strategy], #[any], or #[filter] with the variable name args.

use test_strategy::Arbitrary;

#[derive(Debug, Default)]
struct TestInputArgs {
    x_max: u32,
}

#[derive(Arbitrary, Debug)]
#[arbitrary(args = TestInputArgs)]
struct TestInput {
    #[strategy(0..=args.x_max)]
    x: u32,
}

#[arbitrary(bound(T1, T2, ..))]

By default, if the type of field for which #[strategy] is not specified contains a generic parameter, that type is set to trait bounds.

Therefore, the following TestInputA and TestInputB are equivalent.

use proptest::{
    arbitrary::any, arbitrary::Arbitrary, strategy::BoxedStrategy, strategy::Strategy,
};
use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug)]
struct TestInputA<T> {
    x: T,
}

#[derive(Debug)]
struct TestInputB<T> {
    x: T,
}
impl<T: Arbitrary + 'static> Arbitrary for TestInputB<T> {
    type Parameters = ();
    type Strategy = BoxedStrategy<Self>;

    fn arbitrary_with(_args: Self::Parameters) -> Self::Strategy {
        any::<T>().prop_map(|x| Self { x }).boxed()
    }
}

Types of fields with #[strategy] do not set trait bounds automatically, so you need to set trait bound manually with #[arbitrary(bound(T))].

use proptest::arbitrary::any_with;
use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug, PartialEq)]
#[arbitrary(bound(T))]
struct TestInput<T> {
    #[strategy(any_with::<T>(Default::default()))]
    x: T,
}

You can also specify where predicate instead of type.

use proptest::arbitrary::{any_with, Arbitrary};
use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug, PartialEq)]
#[arbitrary(bound(T : Arbitrary + 'static))]
struct TestInput<T> {
    #[strategy(any_with::<T>(Default::default()))]
    x: T,
}

.. means automatically generated trait bounds.

The following example uses a manually specified trait bounds in addition to the automatically generated trait bounds.

use proptest::arbitrary::any_with;
use test_strategy::Arbitrary;

#[derive(Arbitrary, Debug, PartialEq)]
#[arbitrary(bound(T1, ..))]
struct TestInput<T1, T2> {
    #[strategy(any_with::<T1>(Default::default()))]
    x: T1,

    y: T2,
}

#[arbitrary(dump)]

Causes a compile error and outputs the code generated by #[derive(Arbitrary)] as an error message.

#[proptest]

#[proptest] is the attribute used instead of #[test] when defining a property test.

The following example defines a test that takes a variety of integers as input.

use test_strategy::proptest;

#[proptest]
fn my_test(_input: i32) {
    // ...
}

You can add #[strategy], #[any], #[filter], #[by_ref] to the parameter of the function with # [proptest].

use test_strategy::proptest;

#[proptest]
fn my_test2(#[strategy(10..20)] _input: i32) {
    // ...
}

You can change the configuration of a property test by setting the argument of #[proptest] attribute to a value of proptest::prelude::ProptestConfig type.

use proptest::prelude::ProptestConfig;
use test_strategy::proptest;


#[proptest(ProptestConfig { cases : 1000, ..ProptestConfig::default() })]
fn my_test_with_config(_input: i32) {
    // ...
}

As with #[any], you can also set only the value of the field to be changed from the default value.

The example below is equivalent to the one above.

use proptest::prelude::ProptestConfig;
use test_strategy::proptest;

#[proptest(ProptestConfig::default(), cases = 1000)]
fn my_test_with_config_2(_input: i32) {
    // ...
}

#[proptest(cases = 1000)]
fn my_test_with_config_3(_input: i32) {
    // ...
}

#[proptest(async = ...)]

Async functions can be tested by setting async = ... to the argument of #[proptest].

The following values are allowed after async =. The value specifies the asynchronous runtime used for the test.

• "tokio"

[dev-dependencies]
test-strategy = "0.4.0"
proptest = "1.5.0"
tokio = { version = "1.38.0", features = ["rt-multi-thread"] }

use test_strategy::proptest;
use proptest::prop_assert;

#[proptest(async = "tokio")]
async fn my_test_async() {
    async { }.await;
    prop_assert!(true);
}

#[proptest(dump)]

You can use #[proptest(dump)] and output the code generated by #[proptest] as an compile error message.

#[proptest(dump)]
fn my_test(_input: i32) {
    // ...
}

License

This project is dual licensed under Apache-2.0/MIT. See the two LICENSE-* files for details.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.