I'm still working on learning Rust. Beyond syntax, learning a language requires familiarizing oneself with its idioms and ecosystem. I'm at a point where I want to explore testing in Rust.

The initial problem

We have used Dependency Injection a lot - for ages on the JVM. Even if you're not using a framework, Dependency Injection helps decouple components. Here's a basic example:

class Car(private val engine: Engine) {

    fun start() {
        engine.start()
    }
}

interface Engine {
    fun start()
}

class CarEngine(): Engine {
    override fun start() = ...
}

class TestEngine(): Engine {
    override fun start() = ...
}

In regular code:

val car = Car(CarEngine())

In test code:

val dummy = Car(TestEngine())

DI is about executing different code snippets depending on the context.

Testing in Rust

Rust differs on two main points from the Kotlin approach:

Most DI frameworks happen at runtime, Dagger 2 being the only exception I know about. Rust focuses on compile-time.
Rust is not Object-Oriented

For both of these reasons, we cannot duplicate the above approach. However, Rust excels at macros. For tests, it offers the test macro.

To change a function into a test function, add #[test] on the line before fn. When you run your tests with the cargo test command, Rust builds a test runner binary that runs the annotated functions and reports on whether each test function passes or fails.

-- The Anatomy of a Test Function

At its most basic level, it allows for defining test functions. These functions are only valid when calling cargo test:

fn main() {
    println!("{}", hello());
}

fn hello() -> &'static str {
    return "Hello world";
}

#[test]
fn test_hello() {
    assert_eq!(hello(), "Hello world");
}

cargo run yields the following:

Hello world

On the other hand, cargo run yields:

running 1 test
test test_hello ... ok

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.00s


running 0 tests

test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.01s

However, our main issue is different: we want to code depending on whether it's a testing context.

The test macro is not the solution we are looking for.

Playing with the `cfg` macro

Rust differentiates between "unit" tests and "integration" tests. I added double quotes because I believe the semantics can be misleading. Here's what they mean:

Unit tests are written in the same file as the main. You annotate them with the #[test] macro and call cargo test as seen above
Integration tests are external to the code to test. You annotate code to be part of integration tests with the #[cfg(test)] macro.

Enter the cfg macro:

Evaluates boolean combinations of configuration flags at compile-time.

In addition to the #[cfg] attribute, this macro is provided to allow boolean expression evaluation of configuration flags. This frequently leads to less duplicated code.

-- Macro std::cfg

The cfg macro offers lots of out-of-the-box configuration variables:

Variable	Description	Example
`target_arch`	Target's CPU architecture	`"x86"` `"arm"` `"aarch64"`
`target_feature`	Platform feature available for the current compilation target	`"rdrand"` `"sse"` `"se2"`
`target_os`	Target's operating system	`"windows"` `"macos"` `"linux"`
`target_family`	More generic description of a target, such as the family of the operating systems or architectures that the target generally falls into	`"windows"` `"unix"`
`target_env`	Further disambiguating information about the target platform with information about the ABI` or `libc used	`""` `"gnu"` `"musl"`
`target_endian`	`"big"` or `"little"`
`target_pointer_width`	Target's pointer width in bits	`"32"` `"64"`
`target_vendor`	Vendor of the target	`"apple"` `"pc"`
`test`	Enabled when compiling the test harness
`proc_macro`	When the crate compiled is being compiled with the `proc_macro`
`panic`	Depending on the panic strategy	`"abort"` `"unwind"`

You may have noticed the test flag among the many variables. To write an integration test, annotate the code with the #[cfg(test)] macro:

#[cfg(test)]
fn test_something() {
    // Whatever
}

One can also use the macro to provide alternative code in the test context:

fn hello() -> &'static str {
    return "Hello world";
}

#[cfg(test)]
fn hello() -> &'static str {
    return "Hello test";
}

The above snippet works during cargo run but not during cargo test. In the first case, the second function is ignored; in the second, it's not, and Rust tries to compile two functions with the same signature.

error[E0428]: the name `hello` is defined multiple times
  --> src/lib.rs:10:1
   |
5  | fn hello() -> &'static str {
   | -------------------------- previous definition of the value `hello` here
...
10 | fn hello() -> &'static str {
   | ^^^^^^^^^^^^^^^^^^^^^^^^^^ `hello` redefined here
   |
   = note: `hello` must be defined only once in the value namespace of this module

Fortunately, the cfg macro offers boolean logic. Hence we can negate the test config for the first function:

fn main() {
    println!("{}", hello());
}

#[cfg(not(test))]
fn hello() -> &'static str {
    return "Hello world";
}

#[cfg(test)]
fn hello() -> &'static str {
    return "Hello test";
}

#[test]
fn test_hello() {
    assert_eq!(hello(), "Hello test");
}

cargo run yields Hello world
cargo test compiles then executes the test successfully

While it solves our problem, it has obvious flaws:

It's binary - test context or not
It doesn't scale: after a specific size, the sheer number of annotations will make the project unmanageable

Refining the design

To refine the design, let's imagine a simple scenario that I've faced multiple times on the JVM:

during the regular run, code connects to the production database, e.g., Postgres
for integration testing, code uses a local database, e.g., SQLite
for unit testing, the code doesn't use a database but a mock

Here's the foundation for the design:

fn main() {
    // Get a database implementation                          // 1
    db.do_stuff();
}

trait Database {
    fn doStuff(self: Self);
}

struct MockDatabase {}
struct SqlitDatabase {}
struct PostgreSqlDatabase {}

impl Database for MockDatabase {
    fn doStuff(self: Self) {
        println!("Do mock stuff");
    }
}

impl Database for SqlitDatabase {
    fn doStuff(self: Self) {
        println!("Do stuff with SQLite");
    }
}


impl Database for PostgreSqlDatabase {
    fn doStuff(self: Self) {
        println!("Do stuff with PostgreSQL");
    }
}

How to get the correct implementation depending on the context?

We have three contexts, and cfg[test] only offers a boolean flag. It's time for a new approach.

Leveraging Cargo features

As I searched for a solution, I asked on the Rust Slack channel. William Dillon was kind enough to answer and proposed that I look at Cargo's features.

Cargo "features" provide a mechanism to express conditional compilation and optional dependencies. A package defines a set of named features in the [features] table of Cargo.toml, and each feature can either be enabled or disabled. Features for the package being built can be enabled on the command-line with flags such as --features. Features for dependencies can be enabled in the dependency declaration in Cargo.toml.

-- Features

Defining features

The first step is to define what features we will use. One configures them in the Cargo.toml file:

[features]
unit = []
it = []
prod = []

Using the features in the code

To use the feature, we leverage the cfg macro:

fn main() {
    #[cfg(feature = "unit")]                   // 1
    let db = MockDatabase {};
    #[cfg(feature = "it")]                     // 2
    let db = SqlitDatabase {};
    #[cfg(feature = "prod")]                   // 3
    let db = PostgreSqlDatabase {};
    db.do_stuff();
}

trait Database {
    fn do_stuff(self: Self);
}

#[cfg(feature = "unit")]                       // 1
struct MockDatabase {}

#[cfg(feature = "unit")]                       // 1
impl Database for MockDatabase {
    fn do_stuff(self: Self) {
        println!("Do mock stuff");
    }
}

// Abridged for brevity's sake                 // 2-3

Compiled only if the unit feature is activated
Compiled only if the it feature is activated
Compiled only if the prod feature is activated

Activating a feature

You must use the -F flag to activate a feature.

cargo run -F unit

Do mock stuff

Default feature

The "production" feature should be the most straightforward one. Hence, it's crucial to set it by default.

It has bitten me in the past: when your colleague is on leave, and you need to build/deploy, it's a mess to read the code to understand what flags are mandatory.

Rust allows setting default features. They don't need to be activated; they are on by default. The magic happens in the Cargo.toml file.

[features]
default = ["prod"]                             # 1
unit = []
it = []
prod = []

The prod feature is set as default

We can now run the program without explicitly setting the prod feature:

cargo run

Do stuff with PostgreSQL

Exclusive features

All three features are exclusive: you can activate only one at a time. To disable the default one(s), we need an additional flag:

cargo run --no-default-features -F unit

Do mock stuff

The documentation offers multiple approaches to avoid activating exclusive features at the same time:

There are rare cases where features may be mutually incompatible with one another. This should be avoided if at all possible, because it requires coordinating all uses of the package in the dependency graph to cooperate to avoid enabling them together. If it is not possible, consider adding a compile error to detect this scenario.

-- Mutually exclusive features

Let's add the code:

#[cfg(all(feature = "unit", feature = "it"))]
compile_error!("feature \"unit\" and feature \"it\" cannot be enabled at the same time");
#[cfg(all(feature = "unit", feature = "prod"))]
compile_error!("feature \"unit\" and feature \"prod\" cannot be enabled at the same time");
#[cfg(all(feature = "it", feature = "prod"))]
compile_error!("feature \"it\" and feature \"prod\" cannot be enabled at the same time");

If we try to run with the unit feature while the default prod feature is enabled:

cargo run -F unit

error: feature "unit" and feature "prod" cannot be enabled at the same time
 --> src/main.rs:4:1
  |
4 | compile_error!("feature \"unit\" and feature \"prod\" cannot be enabled at the same time");
  | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Fixing the above design

The above design is not so slightly misleading. In tests, the entry point is not the main function but the test functions themselves.

Let's re-add some tests as in the initial phase.

#[cfg(feature = "prod")]                            // 1
fn main() {
    let db = PostgreSqlDatabase {};
    println!("{}", db.do_stuff());
}

trait Database {
    fn do_stuff(self: Self) -> &'static str;        // 2
}

#[cfg(feature = "unit")]
struct MockDatabase {}
#[cfg(feature = "prod")]
struct PostgreSqlDatabase {}

#[cfg(feature = "unit")]
impl Database for MockDatabase {
    fn do_stuff(self: Self) -> &'static str {
        "Do mock stuff"
    }
}

#[cfg(feature = "prod")]
impl Database for PostgreSqlDatabase {
    fn do_stuff(self: Self) -> &'static str {
        "Do stuff with PostgreSQL"
    }
}

#[test]
#[cfg(feature = "unit")]
fn test_unit() {
    let db = MockDatabase {};
    assert_eq!(db.do_stuff(), "Do mock stuff");     // 3
}

// it omitted for brevity

The PostgreSqlDatabase struct is not available when any test feature is activated
Change the signature to be able to test
Test!

At this point, we can run the different commands:

cargo test --no-default-features -F unit            #1
cargo test --no-default-features -F it              #2
cargo run                                           #3

Run the unit test
Run the "integration test" test
Run the application

Conclusion

In this post, I described the problem caused by having different test suites, focusing on different scopes. The default test configuration variable is binary: either the scope is test or not. It's not enough when one needs to separate between unit and integration tests, each one requiring a different trait implementation.

Rust's features are a way to solve this issue. A feature allows guarding some code behind a label, which one can enable per run on the command line.

To be perfectly honest, I don't know if Rust features are the right way to implement different test scopes. In any case, it works and allows me to understand the Rust ecosystem better.

The complete source code for this post can be found on GitHub

ajavageek / rust-test-suites

To go further:

Originally published at A Java Geek on October 9^th, 2022

Different test scopes in Rust