One Test One Thing

One test, one thing — name it for the scenario, structure it AAA

Each test exercises one logical behaviour, is named after the exact scenario and expected outcome, and is laid out Arrange / Act / Assert. If the name needs _and_, it's two tests.

Why

A test name is the failure message in CI. When test_x_and_y goes red, you can't tell which behaviour broke without opening the file — and the test itself often half-passes through one branch and half-asserts the other, hiding regressions. One scenario per test means the name pins the regression, the AAA shape makes the intent scannable, and a future reader fixes the right thing instead of patching whichever assert tripped first.

How to apply

• Name = scenario + expected outcome. <unit>_<condition>_<expected>. If you wrote _and_, split the test.
• AAA visible. Pick blank-line separation OR // arrange / // act / // assert comments — either works, just make the three phases readable at a glance.
• One thing. One logical behaviour per test. Multiple assert! calls are fine when they all check the same behaviour from different angles (e.g. asserting result.foo and result.bar after one call). Multiple distinct scenarios → multiple tests.

Examples

The "and" smell (live case):

// BEFORE — one test, two scenarios, "and" in the name
#[test]
fn group_len_returns_expected_size_and_zero_for_unknown_group() { ... }

// AFTER — split into two
#[test]
fn group_len_returns_member_count_for_known_group() { ... }

#[test]
fn group_len_returns_zero_for_unknown_group() { ... }

Packed-scenarios smell (live case): created_group_flag_is_true_only_for_first_order_in_a_new_group exercised three states (new group, same group, different group) in one method — split into three named tests, one per scenario.

AAA shape on a small Rust test:

#[test]
fn group_len_returns_zero_for_unknown_group() {
    // arrange
    let store = GroupStore::new();

    // act
    let len = store.group_len(&GroupId::from(42));

    // assert
    assert_eq!(len, 0);
}