Assertion macros for Rust

March 13, 2020 4-minute read

I started writing integration tests for heim recently (better late than never) and the testing code started to bloat up really quick; it’s not very easy to check the logic correctness elegantly with what assertion macros we have in the Rust standard library, especially when we are working with the commonly used Result and Option types.

For example, let’s say that we have some function which returns io::Result<i32> and we want to test if it returns Ok(42) as expected:

fn foo() -> io::Result<i32> {
    Err(io::Error::new(io::ErrorKind::Other, "example purposes"))
}

#[test]
fn test_foo() {
    let result = foo();
}

We always can test if Result is Ok with the is_ok method:

assert!(result.is_ok());

// This assert will panic with the following message:
//
// thread 'main' panicked at 'assertion failed: r.is_ok()', examples/foo.rs:10:5

The panic message is not really helpful, because we have no idea what error caused the assertion failure, and in addition we still need to get value from the Ok variant somehow if this one assertion is correct. Maybe we should use .unwrap instead?

let value = result.unwrap();
assert_eq!(value, 42);

// `.unwrap()` will panic too:
//
// thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: 
// Custom { kind: Other, error: "example purposes" }', examples/foo.rs:11:17

That’s better! Should we shorten this quick test now?

#[test]
fn test_foo() {
	assert_eq!(do_foo().unwrap(), 42);
}

Okay, that kinda does the trick both for Result and Option types.

Can we shorten it a little bit more? Of course, let’s write our own assertion macro!

macro_rules! assert_ok_eq {
    ($cond:expr, $expected:expr) => {
        match $cond {
            Ok(t) => {
                assert_eq!(t, $expected);
            },
            e @ Err(..) => {
                panic!("assertion failed, expected Ok(..), got {:?}", e);
            }
        }
    };
}

It is a pretty simple macro, all it does is a match on the first macro argument and if is Ok(t), it compares that t with the second macro argument:

#[test]
fn test_foo() {
	assert_ok_eq!(foo(), 42);
}

// # If `foo()` returned an `Err`:
//
// thread 'main' panicked at 'assertion failed, expected Ok(..),
//     got Err(Custom { kind: Other, error: "example purposes" })', examples/foo.rs:8:5
//
// # And if it returned an `Ok(1)`:
//
// thread 'main' panicked at 'assertion failed: `(left == right)`
//     left: `1`,
//     right: `42`', examples/foo.rs:8:5

Amazing, not only we had reduced the visual noise in this line now, but also declared our expectations in the macro name — it should be an Ok variant and its value should be equal to the second argument.

More macros!

Once you have started making more abstractions, it is hard to stop, so I ended up with a separate crate called “claim”, which provides a lot of assertion macros to supplement what we already have in libstd:

for comparison: assert_ge, assert_gt, assert_le, and assert_lt
matching: assert_matches
Result: assert_ok, assert_err, and assert_ok_eq
Option: assert_some, assert_none, and assert_some_eq
and Poll: assert_pending, assert_ready, assert_ready_ok, assert_ready_err, and assert_ready_eq

Of course, other crates with the similar purpose already exists, there are assert2, spectral, more-asserts, totems, galvanic-assert, a lot of them!

Why I should use this one?

First of all, claim macros are quite conservative: only the most popular cases handled and for each one case there is a separate macro; for example, there is a assert_some_eq!(Some(42), 42) and not the assert_some!(Some(42), value == 42) (because where that value came from?!)
There is no “fluent” approach also, because I personally prefer not to use these half-assed DSLs, where each one is a unique and impossible to remember. I mean, why: it.expected_to_be_okay().and.equal_to(5)?
Great thing is that current macros system had not changed that much for a last few years, so almost all those macros can be used with any modern enough Rust version; CI checks compatibility back to Rust 1.10 version, but I expect that it should also work with 1.0 version too.
Yet, some types are just not available in older versions, for example, Poll was introduced in Rust 1.36 and macros syntax needed to implement matches! macro for all Rust editions was added in Rust 1.32. With the help of autocfg crate, claim automatically exposes corresponding macros if used Rust compiler can support them, so you don’t even need to enable any features manually!
It does not need std at all, meaning that it can be used in the #![no-std] environments too
And it also has no dependencies (except for a small build one)!

So, it’s a very tiny crate, which makes panic messages a bit better. It is not much, but can be useful in some cases.

More macros!

Why I should use this one?

Links