Galvanic-test: easier test setup for Rust

This crate provides a framework for creating test suites, managing their shared dependencies, and for writing parameterised tests.

• supports test fixtures to handle setup/tear-down of test dependencies
• test fixtures with parameters provide more flexibility in test setups
• parameterised tests---write your test once and verify it in different scenarios
• automatic dependency injection of test fixtures into your tests
• richer test design with galvanic-assert---express more specific properties of parameterised tests
• test suites integrate with galvanic-mock---no manual #[use_mocks] is needed
• other panic-based assertion and mocking frameworks of your choice can be used too

The crate is part of galvanic---a complete test framework for Rust. The framework is shipped in three parts, so you can choose to use only the parts you need.

A short introduction to galvanic-test

Galvanic-test simplifies the setup and tear-down of your test environments and helps you to organise your tests. Everything you already know about Rust testing should still apply.

Tests are organised in test suites which are either named or anonymous.

use galvanic_test::test_suite;

// test suites are only built when a test is executed, e.g., with `cargo test`
test_suite! {
    // for anonymous test suites remove the name directive
    name my_test_suite;

    // suites act as modules and may contain any item
    fn calc(a: i32, b: i32) -> i32 { a*b }

    // instead of `fn`, `test` defines a test item.
    test simple_first_test() {
        assert_eq!(3*2, 6);
    }

    // attributes can usually be applied as for functions, e.g., #[should_panic(expected = "...")] is curently not supported
    #[should_panic]
    test another_test() {
        assert_eq!(calc(3,2), 7);
    }
}

The most powerful part of galvanic-test are test fixtures to manage your test environments. A test fixture is a piece of code which setups one specific part of a test and makes sure that it's torn down after the test executed (even if it failed). If you know pytest you should feel at home. If you have experience with XUnit-style frameworks, e.g., JUnit, CPPUnit, ...; then you can think about fixtures as different before/after blocks which belong together.

use galvanic_test::test_suite;

test_suite! {
    use std::fs::{File, remove_file};
    use std::io::prelude::*;

    fixture bogus_number() -> i32 {
        setup(&mut self) {
            42
        }
    }

    fixture input_file(file_name: String, content: String) -> File {
        members {
            file_path: Option<String>
        }
        setup(&mut self) {
            let file_path = format!("/tmp/{}.txt", self.file_name);
            self.file_path = Some(file_path.clone());
            {
                let mut file = File::create(&file_path).expect("Could not create file.");
                file.write_all(self.content.as_bytes()).expect("Could not write input.");
            }
            File::open(&file_path).expect("Could not open file.")
        }
        // tear_down is optional
        tear_down(&self) {
            remove_file(self.file_path.as_ref().unwrap()).expect("Could not delete file.")
        }
    }

    // fixtures are arguments to the tests
    test a_test_using_a_fixture(bogus_number) {
        assert_eq!(21*2, bogus_number.val);
    }

    // fixtures with arguments must receive the required values
    test another_test_using_fixtures(input_file(String::from("my_file"), String::from("The stored number is: 42"))) {
        let mut read_content = String::new();
        input_file.val.read_to_string(&mut read_content).expect("Couldn't read 'my_file'");

        assert_eq!(&read_content, input_file.params.content);
    }
}

Test fixtures enable us also to run the same test code with different parameterisations. This can significantly reduce our work required for testing complex code with multiple execution paths.

test_suite! {
    fixture product(x: u32, y: u32) -> u32 {
        params {
            vec![(2,3), (2,4), (1,6), (1,5), (0,100)].into_iter()
        }
        setup(&mut self) {
            self.x * self.y
        }
    }

    test a_parameterised_test_case(product) {
        let wrong_product = (0 .. *product.params.y).fold(0, |p,_| p + product.params.x) - product.params.y%2;
        // fails for (2,3) & (1,5)
        assert_eq!(wrong_product, product.val)
    }
}

Documentation

Galvanic-test simplifies the setup of shared test environments, i.e., it helps us to create and reset the resources needed by our tests work properly.

It is recommended that you add galvanic-test as a dev-dependency in your Cargo.toml. Make sure to use an appropriate version specification. The crate follows semantic versioning.

For Rust edition 2018 use a version number of at least 0.2

[dev-dependencies]
galvanic-test = "0.2"

After specifying the dependency we can import the test_suite macro as follows.

use galvanic_test::test_suite;

When using galvanic-test as a dev-dependency make sure that the use statement is only reachable when your crate is compiled when tests are enabled, e.g., wrap it in a #[cfg(test)] annotated module.

Rust version before edition 2018

For using the crate with a Rust version before edition 2018 use a version number up to 0.1.5

[dev-dependencies]
galvanic-test = "0.1.5"

After specifying the dependency we include the library with enabled macros in our main.rs,lib.rs, and/or our integration tests in tests/.

#[cfg(test) #[macro_use] extern crate galvanic_test;

When using galvanic-test as a dev-dependency make sure that any macro of galvanic-test is only reachable when tests are enabled.

Creating test suites for grouping tests

Before we start writing tests we have a look at how to group them. Tests are organized in test suites. A test suite takes care of several things:

• They create a private module to group test cases and test fixtures.
• They are only built if tests are built, e.g., using cargo test.
• Test fixtures defined in the suite can be injected into its test cases.
• If the galvanic_mock_integration feature is enabled then the test suite uses an implicit #[use_mocks] directive. (nightly)

They come in two varieties: anonymous and named. To create a anonymous test suite we use the test_suite! macro.

test_suite! {
    // ...
}

For easier location of a failing test case it is recommended to name a test suite.

test_suite! {
    name some_identifer_naming_the_suite;
    // ...
}

Note that the name directive must occur as the first element of the suite.

Writing tests in test suites

Now that we have defined a test suite, we can fill it with test cases. A test case is defined as a test item.

test_suite! {
    test my_first_test_case() {
        // ... some assertions
        assert_eq!(1+1, 2);
    }
}

If we want to define a test which is expected to panic we can simply use the #[should_panic] attribute or if we need more fine grained control we may use galvanic-assert's assert_that!(..., panics); macro.

test_suite! {
    #[should_panic]
    test a_panicking_test_case() {
        // ... some failing assertion
        assert_eq!(1+1, 4);
    }
    test a_panicking_test_case_using_galvanic_assert() {
        assert_that(panic("No towels!"), panics);
    }
}

So far test cases behave similar to functions annotated with #[test] as in simple Rust unit tests. Though, test cases defined as a test item support automatic injection of test fixtures and parametrisation, as we will see later.

Adding test fixtures for test resource management

Tests often depend on some resources of the test environment, e.g., objects used by the test, files with input, etc. All those things must be created at the beginning of the test and torn down at the end of the test. If we forget or mess up one of those tasks we introduce errors in test code, which is actually not central to the test. Further if many parts of these environments are the same or similar for several test cases then the problem gets even worse.

To keep our tests clean we do not want to code setup and tear down tasks multiple times. Therefore we write a test fixture for each resource.

fixture a_number() -> i32 {
    setup(&mut self) {
        42
    }
    tear_down(&self) {
        println!("Cleaning up ...");
    }
}

Every fixture definition consists of the following parts:

• the fixture keyword
• a name: a_number in our example
• a list of typed arguments: none in our example
• the type of the resource managed by the fixture: i32 here
• a required setup block which receives the fixture (self) as a mutable borrow and must return a resource of the type specified by the fixture
• an optional tear_down block which receives the fixture (self) as an immutable borrow

To use our new fixture in a test it must be defined in the same test_suite!. The fixtures required by a test are given as parameters for test case by name. Before the test is executed, setup method is invoked. Its return value is then wrapped in a FixtureBinding and the binding is injected into test case. The return value can then be accessed by the binding's val member.

test a_test_using_a_fixture(a_number) {
    assert_eq!(a_number.val, 42);
}

Test fixtures with arguments

Often setting up exactly the same resource for several tests is not enough and we'd like to parameterise the setup/tear_down code. We can do so by specifying arguments for the fixture.

fixture offset_number(offset: i32) -> i32 {
    setup(&mut self) {
        self.offset + 42
    }
    tear_down(&self) {
        println!("Cleaning up a number with offset {} ...", self.offset);
    }
}

The arguments are then accessible as members of the fixture. A test can then specify the required arguments when requesting the fixture. The arguments passed to the fixture are accessible in the test case through the FixtureBinding's params member by the names used in the fixture definition.

test a_test_using_a_fixture(offset_number(8)) {
    assert_eq!(offset_number.val, 42 + offset_number.params.offset);
}

Sharing data between setup() and tear_down()

We've seen that fixture arguments are available both in the setup and tear_down blocks via self. However there are situations where we depend on some external input, e.g., the system time, a random number, both in our setup code and tear-down code to, e.g., create unique file names or other identifiers. With the facilities shown so far we have no (non-hacky) way to transfer the information.

To get around this issue we can define member variables for our fixtures. A member variable is accessible via self and is always an Option which is initialised with None. The setup block may then overwrite the members' values (therefore its &mut self).

To declare member variables we need to place a members block before the setup block and list our variable declarations ase we would in a struct.

fixture offset_number() -> i32 {
    members {
        some_identifier: Option<i32>
    }
    setup(&mut self) {
        self.some_identifier = Some(12)
        42
    }
    tear_down(&self) {
        println!("Cleaning up a fixture with identifier {} ...", self.some_identifier.as_ref().unwrap();
    }
}

Writing parameterised tests and fixtures

A very powerful feature of galvanic-test is the ability to parameterise tests. A parameterised test case is run with several different initialisations of its fixtures.

First we need a test case which accepts one or multiple fixtures. Let's write a test which calculates the product of two numbers (x,y) by summing x, y-times.

test parameterised_test(product) {
    let sum: u32 = (0..product.params.y).fold(0, |a,b| a + product.params.x);
    assert_eq!(sum, product.val);
}

We want to test this code snippet with different values to test the border cases and equivalence classes. For we create a product fixture with arguments x and y and let the setup block calculate the product of the two numbers. To make the fixture parameterised we add a params block at the beginning. The block must return an Iterator<R> where R is the type of the fixture's return value.

fixture product(x: u32, y: u32) -> u32 {
    params {
        vec![(2,3), (1,4), (0,100)].into_iter()
    }
    setup(&mut self) {
        self.x * self.y
    }
}

Now if we run our tests each test case with takes the product fixture as an argument without supplying parameters to the fixture will take the values from the params block instead. The setup and tear_down block will be executed before/after each parameterisation.

If the test case takes multiple parameterised fixtures then all possible combinations (the cross-product) will be evaluated. Again before/after each parameterisation all setup/tear_down blocks of the parameterised fixtures will be executed.

If on the other hand you provide parameters to a parameterised fixture, as shown below, then only that parameterisation will be considered for the fixture.

test parameterised_test(product(3,8)) {
    let sum: u32 = (0..product.params.y).fold(0, |a,b| a + product.params.x);
    assert_eq!(sum, 24);
}

Errors, #[should_panic], and parameterised fixtures

Let's see what happens if a test fails.

test failing_parameterised_test(product) {
    let sum: i32 = (0..*product.params.y).fold(0, |a,b| a + product.params.x);
    assert_eq!(sum, product.val - product.params.x%2)
}

The framework will show you all parameterisations which triggered an error so debugging will be easier.

...
running 1 test
thread 'test::parameterised_test' panicked at 'assertion failed: `(left == right)`
  left: `4`,
 right: `3`', src/main.rs:17:8
note: Run with `RUST_BACKTRACE=1` for a backtrace.
The above error occured with the following parameterisation of the test case:
    product { x: 1, y: 4 }

thread 'test::parameterised_test' panicked at 'Some parameterised test cases failed', src/main.rs:3:0
test test::parameterised_test ... FAILED
...

Be careful when applying #[should_panic] to a parameterised test case. In that case the test will succeed if any parameterisation fails. To assert that all parameterisation fail it's recommended to use assert_that!(..., panics) from the galvanic-assert crate to treat panicking like a regular behaviour.

Further #[should_panic(expected = "message")] currently is not supported for tests with fixtures as the test output is modified to include information about the failing fixture parameterision.

Enabling Galvanic-mock integration

If you want to use galvanic-mock integration (only available on nightly) then add

#[macro_use] extern crate galvanic_test;
#![feature(proc_macro)]
extern crate galvanic_mock;

and enable the galvanic_mock_integration feature in your Cargo.toml

[dev-dependencies]
galvanic-test = { version = "*", features = ["galvanic_mock_integration"] }
galvanic-mock = "*" # replace with the correct version

Afterwards each test suite will automatically apply the #[use_mocks] attribute so you can use fixtures to return actual mock objects.