Clean tests in Rust.
expect(a + 2) {
when(a = 2) {
to equal(4)
}
}
- Introduction
- Installation
- Usage
- Assertions
- Supported libraries
- More examples
- Known issues and limitations
- Debugging
- License
How often when you see a Rust test you think to yourself "wow, this is a really beautifully written test"? Not often, right? Classic Rust tests do not provide any structure beyond the test function itself. This often results in a lot of boilerplate code, ad-hoc test structure and overall poor quality.
Tests are about verifying that a given piece of code run under certain conditions works as expected. A good testing framework embraces this way of thinking. It makes it easy to structure your code in a way that reflects it. Folks in other communities have been doing this for a long time with tools like RSpec and Jasmine.
If you want beautiful, high-quality tests that are a pleasure to read and write you need something else. Using Rust's procedural macros lets_expect
introduces
a syntax that let's you clearly state what you're testing, under what conditions and what is the expected result.
The outcome is:
- easy to read, DRY, TDD-friendly tests
- less boilerplate, less code
- nicer error messages
- more fun
expect(posts.create_post(title, category_id)) {
before { posts.push(Post {title: "Post 1" }) }
after { posts.clear() }
when(title = valid_title) {
when(category_id = valid_category) to create_a_post {
be_ok,
have(as_ref().unwrap().title) equal(valid_title),
change(posts.len()) { from(1), to(2) }
}
when(category_id = invalid_category) to return_an_error {
be_err,
have(as_ref().unwrap_err().message) equal("Invalid category"),
not_change(posts.len())
}
}
when(title = invalid_title, category_id = valid_category) to be_err
}
Now let's compare it to a classic Rust test that does the same thing:
fn run_setup<T>(test: T) -> ()
where T: FnOnce(&mut Posts) -> () + panic::UnwindSafe
{
let mut posts = Posts { posts: vec![] };
posts.push(Post { title: "Post 1" });
let posts = Mutex::new(posts);
let result = panic::catch_unwind(|| {
test(posts.try_lock().unwrap().deref_mut())
});
posts.try_lock().unwrap().clear();
assert!(result.is_ok());
}
#[test]
fn creates_a_post() {
run_setup(|posts: &mut Posts| {
let before_count = posts.len();
let result = posts.create_post(VALID_TITLE, VALID_CATEGORY);
let after_count = posts.len();
assert!(result.is_ok());
assert_eq!(VALID_TITLE, result.unwrap().title);
assert_eq!(after_count - before_count, 1);
})
}
#[test]
fn returns_an_error_when_category_is_invalid() {
run_setup(|posts: &mut Posts| {
let before_count = posts.len();
let result = posts.create_post(VALID_TITLE, INVALID_CATEGORY);
let after_count = posts.len();
assert!(result.is_err());
assert_eq!("Invalid category", result.unwrap_err().message);
assert_eq!(after_count, before_count);
})
}
#[test]
fn returns_an_error_when_title_is_empty() {
run_setup(|posts: &mut Posts| {
let result = posts.create_post("", VALID_CATEGORY);
assert!(result.is_err());
})
}
Add the following to your Cargo.toml
:
[dev-dependencies]
lets_expect = "0"
Under the hood lets_expect
generates a single classic test function for each to
block. It names those tests automatically based on what you're testing and
organizes those tests into modules. This means you can run those tests using cargo test
and you can use all cargo test
features. IDE extensions will
also work as expected.
lets_expect
tests need to be placed inside of a lets_expect!
macro, which in turn needs to be placed inside of a tests
module:
#[cfg(test)]
mod tests {
use super::*;
use lets_expect::lets_expect;
lets_expect! {
expect(subject) {
to expectation
}
}
}
It might be a good idea to define a code snippet in your IDE to avoid having to type this piece of boilerplate every time.
The examples here omit the macro for brevity.
expect
sets the subject of the test. It can be any Rust expression (including a block). to
introduces expectations. It can be followed
by a single expectation or a block of expectations. In the latter case you must provide a name for the test, which needs to be a valid Rust identifier.
expect(2) {
to equal(2)
}
If there are multiple assertions in a to
block they need to be separated by a comma.
expect({ 1 + 1 }) {
to be_actually_2 {
equal(2),
not_equal(3)
}
}
One to
block generates a single test. This means the subject will be executed once and then all the assertions inside that to
block will be run.
If you want to generate multiple tests you can use multiple to
blocks:
expect(files.create_file()) {
to make(files.try_to_remove_file()) be_true
to make(files.file_exists()) be_true
}
If your expect
contains a single item you can omit the braces:
expect(a + 2) when(a = 2) {
to equal(4)
}
Inside the top level lets_expect!
macro as well as expect
and when
blocks you can use let
to define variables.
expect(a) {
let a = 2;
to equal(2)
}
Variables can be overwritten in nested blocks. New definitions can use values from outer blocks.
expect(a) {
let a = 2;
when a_is_4 {
let a = a + 2;
to equal(4)
}
}
Variables don't have to be defined in the order they're used.
expect(sum) {
let sum = a + b;
let a = 2;
when b_is_three {
let b = 3;
to equal(5)
}
}
when
sets a value of one or more variables for a given block. This keyword is this library's secret sauce. It allows you to define values of variables
for multiples tests in a concise and readable way, without having to repeat it in every test.
expect(a + b + c) {
let a = 2;
when(c = 5) {
when(b = 3) {
to equal(10)
}
when(a = 10, b = 10) {
to equal(25)
}
}
}
You can use similar syntax as in let
to define variables. The only difference being the let
keyword itself is ommited.
expect(a += 1) {
when(mut a: i64 = 1) {
to change(a.clone()) { from(1), to(2) }
}
}
You can also use when
with an identifier. This will simply create a new context with the given identifier. No new variables are defined.
expect(login(username, password)) {
when credentials_are_invalid {
let username = "invalid";
let password = "invalid";
to be_false
}
}
If your when
contains only one item the braces can be ommited:
expect(a + 2) when(a = 2) to equal(4)
when
blocks do not have to be placed inside of expect
blocks. Their order can be reversed.
when(a = 2) {
expect(a + 2) to equal(4)
}
have
is used to test values of attributes or return values of methods of the subject.
let response = Response { status: 200, content: ResponseContent::new("admin", "123") };
expect(response) {
to be_valid {
have(status) equal(200),
have(is_ok()) be_true,
have(content) {
have(username) equal("admin".to_string()),
have(token) equal("123".to_string()),
}
}
}
Multiple assertions can be provided to have
by wrapping them in curly braces and separating them with commas.
make
is used to test values of arbitrary expressions.
expect(posts.push((user_id, "new post"))) {
let user_id = 1;
to make(user_has_posts(user_id)) be_true
}
Multiple assertions can be provided to make
by wrapping them in curly braces and separating them with commas.
change
is used to test if and how a value changes after subject is executed. The expression given as an argument to change
is evaluated twice. Once before the subject is executed and once after.
The two values are then provided to the assertions specified in the change
block.
expect(posts.create_post(title, category_id)) {
after { posts.clear() }
when(title = valid_title) {
when(category_id = valid_category) {
to change(posts.len()) { from(0), to(1) }
}
when(category_id = invalid_category) {
to not_change(posts.len())
}
}
}
The contents of the before
blocks are executed before the subject is evaluated, but after the let
bindings are executed. The contents of the after
blocks are executed
after the subject is evaluated and the assertions are verified.
before
blocks are run in the order they are defined. Parent before
blocks being run before child before
blocks. The reverse is true for after
blocks.
after
blocks are guaranteed to run even if assertions fail. They however will not run if the let statements, before blocks, subject evaluation or assertions panic.
let mut messages: Vec<&str> = Vec::new();
before {
messages.push("first message");
}
after {
messages.clear();
}
expect(messages.len()) { to equal(1) }
expect(messages.push("new message")) {
to change(messages.len()) { from(1), to(2) }
}
Because lets_expect
uses standard Rust tests under the hood it has to come up with a unique identifier for each test. To make those identifiers
readable lets_expect
uses the expressions in expect
and when
to generate the name. This works well for simple expressions but can get a bit
messy for more complex expressions. Sometimes it can also result in duplicated names. To solve those issues you can use the as
keyword to give
the test an explicit name:
expect(a + b + c) as sum_of_three {
when(a = 1, b = 1, c = 1) as everything_is_one to equal(3)
}
This will create a test_named:
expect_sum_of_three::when_everything_is_one::to_equal_three
instead of
expect_a_plus_b_plus_c::when_a_is_one_b_is_one_c_is_one::to_equal_three
lets_expect
promotes tests that only test one piece of code at a time. Up until this point all the test we've seen define a subject, run that subject and
verify the result. However there can be situations where we want to run and test multiple pieces of code in sequence. This could be for example because executing a piece
of code might be time consuming and we want to avoid doing it multiple times in multiple tests.
To address this lets_expect
provides the story
keyword. Stories are a bit more similar to classic tests in that they allow
arbitrary statements to be interleaved with assertions.
Please note that the expect
keyword inside stories has to be followed by to
and can't open a block.
story login_is_successful {
expect(page.logged_in) to be_false
let login_result = page.login(&invalid_user);
expect(&login_result) to be_err
expect(&login_result) to equal(Err(AuthenticationError { message: "Invalid credentials".to_string() }))
expect(page.logged_in) to be_false
let login_result = page.login(&valid_user);
expect(login_result) to be_ok
expect(page.logged_in) to be_true
}
NOTE: For now
expect
blocks can't be placed inside of loops or closures. They need to be top-level items in a story.
For some tests you may need to make the tested value mutable or you may need to pass a mutable reference to the assertions. In expect
, have
and make
you can
use the mut
keyword to do that.
expect(mut vec![1, 2, 3]) { // make the subject mutable
to have(remove(1)) equal(2)
}
expect(mut vec.iter()) { // pass a mutable reference to the iterator to the assertion
let vec = vec![1, 2, 3];
to all(be_greater_than(0))
}
expect(vec![1, 2, 3]) {
to have(mut iter()) all(be_greater_than(0)) // pass a mutable reference to the iterator to the assertion
}
let
and when
statements also support mut
.
expect(2 == 2) to be_true
expect(2 != 2) to be_false
expect(2) to be_actually_two {
equal(2),
not_equal(3)
}
expect(2.1) {
to be_close_to(2.0, 0.2)
to be_greater_than(2.0)
to be_less_or_equal_to(2.1)
}
match_pattern!
is used to test if a value matches a pattern. It's functionality is similar to matches!
macro.
expect(Response::UserCreated) {
to match_pattern!(Response::UserCreated)
}
expect(Response::ValidationFailed("email")) {
to match_email {
match_pattern!(Response::ValidationFailed("email")),
not_match_pattern!(Response::ValidationFailed("email2"))
}
}
lets_expect
provides a set of assertions for Option
and Result
types.
expect(Some(1u8) as Option<u8>) {
to be_some_and equal(1)
to be_some {
equal(Some(1)),
be_some
}
}
expect(None as Option<String>) {
to be_none {
equal(None),
be_none
}
}
expect(Ok(1u8) as Result<u8, ()>) {
to be_ok_and equal(1)
to be_ok {
be_ok,
equal(Ok(1)),
}
}
expect(Err(2) as Result<(), i32>) {
to be_err_and equal(2)
to be_err {
be_err,
equal(Err(2)),
}
}
expect(panic!("I panicked!")) {
to panic
}
expect(2) {
to not_panic
}
panic
and not_panic
assertions can be the only assertions present in a to
block.
expect(vec![1, 2, 3]) {
to have(mut iter()) all(be_greater_than(0))
to have(mut iter()) any(be_greater_than(2))
}
lets_expect
provides a way to define custom assertions. An assertion is a function that takes the reference to the
subject and returns an AssertionResult
.
Here's two custom assertions:
use lets_expect::*;
fn have_positive_coordinates(point: &Point) -> AssertionResult {
if point.x > 0 && point.y > 0 {
Ok(())
} else {
Err(AssertionError::new(vec![format!(
"Expected ({}, {}) to be positive coordinates",
point.x, point.y
)]))
}
}
fn have_x_coordinate_equal(x: i32) -> impl Fn(&Point) -> AssertionResult {
move |point: &Point| {
if point.x == x {
Ok(())
} else {
Err(AssertionError::new(vec![format!(
"Expected x coordinate to be {}, but it was {}",
x, point.x
)]))
}
}
}
And here's how to use them:
expect(Point { x: 2, y: 22 }) {
to have_valid_coordinates {
have_positive_coordinates,
have_x_coordinate_equal(2)
}
}
Remember to import your custom assertions in your test module.
Similarly custom change assertions can be defined:
use lets_expect::*;
fn by_multiplying_by(x: i32) -> impl Fn(&i32, &i32) -> AssertionResult {
move |before, after| {
if *after == *before * x {
Ok(())
} else {
Err(AssertionError::new(vec![format!(
"Expected {} to be multiplied by {} to be {}, but it was {} instead",
before,
x,
before * x,
after
)]))
}
}
}
And used like so:
expect(a *= 5) {
let mut a = 5;
to change(a.clone()) by_multiplying_by(5)
}
This library has fairly few builtin assertions compared to other similar ones. This is because the use of have
, make
and match_pattern!
allows for
expressive and flexible conditions without the need for a lot of different assertions.
The full list of assertions is available in the assertions module.
lets_expect
works with Tokio. To use Tokio in your tests you need to add the tokio
feature in your Cargo.toml
:
lets_expect = { version = "*", features = ["tokio"] }
Then whenever you want to use Tokio in your tests you need to add the tokio_test
attribute to your lets_expect!
macros like so:
lets_expect! { #tokio_test
}
This will make lets_expect
use #[tokio::test]
instead of #[test]
in generated tests.
Here's an example of a test using Tokio:
let value = 5;
let spawned = tokio::spawn(async move {
value
});
expect(spawned.await) {
to match_pattern!(Ok(5))
}
lets_expect
repository contains tests that might be useful as examples of using the library.
You can find them here.
- rust-analyzer's auto-import doesn't seem to work well from inside of macros. It might be necessary to manually add
use
statements for types from outside of the module. - Syntax highlighting doesn't work with
lets_expect
syntax. Currently there's no way for Rust macros to export their syntax to language tools. - Shared contexts (similar to RSpec) seem to be impossible to implement without eager macro expansion.
If you're having trouble with your tests you can use cargo-expand to see what code is generated by lets_expect
.
The generated code is not always easy to read and is not guaranteed to be stable between versions. Still it can be useful for debugging.
This project is licensed under the terms of the MIT license.