Skip to content

Commit

Permalink
[book] document example of manual arbitrary impl
Browse files Browse the repository at this point in the history
  • Loading branch information
cameron1024 authored and matthew-russo committed Sep 23, 2024
1 parent 52d3a38 commit d8678b7
Showing 1 changed file with 82 additions and 1 deletion.
83 changes: 82 additions & 1 deletion book/src/proptest/tutorial/arbitrary.md
Original file line number Diff line number Diff line change
Expand Up @@ -10,5 +10,86 @@ In some cases, where it makes sense to define a canonical strategy, such as in
the [JSON AST example](recursive.md), it is a good idea to implement
`Arbitrary`.


## Deriving `Arbitrary`

The experimental [`proptest-derive` crate](../../proptest-derive/index.md) can
be used to automate implementing `Arbitrary` in common cases.
be used to automate implementing `Arbitrary` in common cases. For example, imagine we have a struct that represents a point in a 2-D coordinate space:
```rust
#[derive(Debug)]
struct Point {
x: i32,
y: i32,
}
```
This struct has the property that any pair of valid `i32`s can make a valid `Point`, so that is perfect for using `#[derive(Arbitrary)]`.

## Manual `Arbitrary` implementations

Sometimes, however, there are extra constraints that your type has, which the derive macro can't understand. In these cases, you'll need to implement `Arbitrary` for your type manually.

For example, consider this struct which represents a range (note, the derive API is can actually represent this case, it's just an example):
```rust
#[derive(Debug)]
struct Range {
lower: i32,
upper: i32,
}

impl Range {
pub fn new(lower: i32, upper: i32) -> Option<Self> {
if lower <= upper {
Some(Self { lower, upper })
} else {
None
}
}
}
```
This struct has an invariant: `lower <= upper`. However, if we derive an `Arbitrary` implementation naively, it might generate `Range { lower: 1, upper: 0 }`.

Instead, we can write a manual implementation:
```rust
impl Arbitrary for Range {
type Parameters = ();
type Strategy = FilterMap<StrategyFor<(i32, i32)>, fn((i32, i32)) -> Option<Self>>;

fn arbitrary_with(_parameters: Self::Parameters) -> Self::Strategy {
any::<(i32, i32)>() // generate 2 arbitrary i32s
.prop_map(|(a, b)| {
let (lower, upper) = if a < b {
(a, b)
} else {
(b, a)
};
Range::new(lower, upper).unwrap()
})
}
}
```
Here, there are three items we need to define:
- `type Parameters` - the type of any parameters to `arbitrary_with`. Here (and in many cases), we don't need this, so `()` is used.
- `type Strategy` - the type of the strategy produced
- `fn arbitrary_with` - the code that creates the canonical `Strategy` for this type

It's important to consider what type you want to use for `Strategy`. Here, we explicitly write the type out. This uses static dispatch, which is often faster and easier to optimize, but has a few downsides:
- you need to write out the type of the strategy. Even for this small function, it's a pretty lengthy function signature. In the worst case, it's impossible, since some types are unnameable (e.g. closures which capture their environment)
- it makes the implementation of `arbitrary_with` a part of your public API signature (if you expose `Arbitrary` impls in general from your crate). This means that changes to the implementation may require a breaking change.

There are a couple of ways around this:
- heap-allocate the strategy by:
- returning `BoxedStrategy<T>`
- calling `.boxed()` on the strategy before returning it
- use the nightly-only `#![feature(type_alias_impl_trait)]`:
```rust
type RangeStrategy = impl Strategy<Value = Range>;

impl Arbitrary for Range {
type Parameters = ();
type Strategy = RangeStrategy;
// ...
}
```

Using `BoxedStrategy` will incur some performance penalty relating to a heap allocation as well as dynamic dispatch, but it works on stable (as of November 2022).

0 comments on commit d8678b7

Please sign in to comment.