Fluent assertions for the Rust programming language.
-
Assertions that read like english:
assert_that("foobar").starts_with("foo").contains("ooba"); assert_that(vec![1, 2, 3]).has_length(3).contains(2); assert_that((Ok(42)).is_ok().is_equal_to(42); assert_that(Person { id: 42 }).has_debug_value("Person { id: 42 }");
-
Assertions to be defined on common traits as often as possible. Allowing, for example, all types implementing
Eq
to allowis_equal_to
,PartialOrd
types to allowis_greater_than
assertions and all types implementing theHasLength
trait to support thehas_length
assertion. -
No requirement to use macros for simple assertions. An
assert_that(...)
suffices to get into an assertion context. Useassert_that_ref(&val)
if you cant give up ownership and instead want to assert on a reference. -
One import should be enough to access all possible assertions through autocomplete.
use assertr::prelude::*;
-
Chainable assertions.
assert_that("foobar") .is_not_empty() .starts_with("foo") .ends_with("bar") .has_length(6);
-
Extensibility. Write assertions for your own types.
#[derive(Debug, PartialEq)] struct Person { age: u32, } trait PersonAssertions { fn has_age(self, expected: u32) -> Self; } impl<M: Mode> PersonAssertions for AssertThat<'_, Person, M> { fn has_age(self, expected: u32) -> Self { self.satisfies(|p| p.age, |age| { age.is_equal_to(expected); }) } } #[test] fn test() { assert_that(Person { age: 30 }) .has_age(30); }
-
Partial equality assertions (meaning that only some fields of a struct are compared, while some are ignored). Add the
AssertrEq
annotation to one of your struct to enable this.use assertr::prelude::*; use indoc::formatdoc; // Deriving `Debug` is necessary, as we want to actually use `Foo` in an assertion. #[derive(Debug, AssertrEq)] pub struct Foo { pub id: i32, pub name: String, pub data: (u32, u32), } fn main() { let foo = Foo { id: 1, name: "bob".to_string(), data: (42, 100), }; assert_that(foo).is_equal_to(FooAssertrEq { id: eq(1), name: eq("bob".to_string()), data: any(), }); }
One other style of assertions in Rust is the "individual macros" approach.
The standard library already comes with a few of them, like the assert_eq!
macro, many libraries provide a more
exhaustive list of macros specifically tailored for specific types and operations.
Let me point out a few benefits of fluent assertions compared to individual assert macros.
The fluent interface allows you to chain multiple assertions naturally, following the way we think about validating properties. Instead of writing multiple separate assertions, you can express related checks in a single, flowing statement that reads almost like natural language.
Additionally, having a concrete entrance into the assertion context using a function like assert_that
with assertions
coming after makes it totally obvious which value is the "actual" and which is the "expected" value. This provides
a clear schema for how assertions are written, compared to an assertion macro, like std's assert_eq!
, in which the
order of arguments can be chosen freely, making it non-obvious when coming into a new codebase which style
was chosen.
Fluent assertions can provide more detailed and structured error messages out of the box. Rather than just showing the values that didn't match, they can include context about what specific check failed within the chain and clearer descriptions of the expected vs actual values. Descriptive messages can be collected throughout the call chain.
With traditional assert macros, you often need to reference the same value multiple times:
let vec = vec![1, 2, 3];
assert_eq!(vec.len(), 3);
assert!(vec.contains(&2));
Versus the fluent style:
assert_that(vec![1, 2, 3]).has_length(3).contains(2);
Fluent assertions provide better IDE support through method chaining. The IDE can show you exactly what assertions are available for your specific type, making it harder to write invalid assertions and easier to discover available checks.
It's generally easier to add new assertion types in a fluent interface - you just need to implement new methods on the assertion type. With macros, you'd need to create entirely new macros for each assertion type, which can be more complex and harder to maintain.
-
Many assertions require
std::fmt::Debug
, limiting usability to types implementing Debug. Can we implement fallback rendering? Will probably require the currently unstable specialization feature. -
The differentiation between
assert_that
for owned values andassert_that_ref
for references is bad. Oneassert_that
function, not being macro, accepting both owned values and references would be much preferred. But that would also require the specialization feature to be able to detect the use of a reference type at compiletime.
- Derived assertions are not allowed to control whether the location is printed.
- Detail messages are collected from the current assertion upwards, taking the messages of all parents into account.
- Failures are stored at the root assertion.
- Failures can only be extracted from the root assertion.
To test all creates, run with --all when in root
cargo test --all
This crate uses features. Some tests are declared under conditional compilation.
Run all tests using
cargo test --all-features
Contributions are welcome. Feel free to open a PR with your suggested changes.
Midway through implementing this, I found out that "spectral" already exists, which uses a very similar style of
assertions.
After looking into it, I took the concept of generally relying on *Assertions
trait definitions instead of directly
implementing Assertions with multiple impl blocks on the AssertThat
type.