This crate provides two procedural macros, #[derive(Arbitrary)]
and #[proptest]
.
Each of these macros is an alternative to the following proptest's official macros.
test-strategy | proptest | proptest-derive |
---|---|---|
#[derive(Arbitrary)] |
#[derive(Arbitrary)] |
|
#[proptest] |
proptest ! { } |
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.
Add this to your Cargo.toml:
[dependencies]
test-strategy = "0.4.0"
proptest = "1.5.0"
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 can be written in the following positions.
attribute | function | struct | enum | variant | field | function parameter |
---|---|---|---|---|---|---|
#[strategy] |
✔ | ✔ | ||||
#[any] |
✔ | ✔ | ||||
#[weight] |
✔ | |||||
#[map] |
✔ | ✔ | ||||
#[filter] |
✔ | ✔ | ✔ | ✔ | ✔ | ✔ |
#[by_ref] |
✔ | ✔ | ||||
#[arbitrary(args = T)] |
✔ | ✔ | ||||
#[arbitrary(bound(...))] |
✔ | ✔ | ✔ | ✔ | ||
#[arbitrary(dump)] |
✔ | ✔ | ||||
#[proptest] |
✔ | |||||
#[proptest(async = ...)] |
✔ | |||||
#[proptest(dump)] |
✔ |
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()
}
}
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,
}
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,
}
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),
}
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,
}
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 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,
}
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,
}
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,
}
Causes a compile error and outputs the code generated by #[derive(Arbitrary)]
as an error message.
#[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) {
// ...
}
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);
}
You can use #[proptest(dump)]
and output the code generated by #[proptest]
as an compile error message.
#[proptest(dump)]
fn my_test(_input: i32) {
// ...
}
This project is dual licensed under Apache-2.0/MIT. See the two LICENSE-* files for details.
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.