docs(test): make Given/When/Then the default BDD structure (WP-71)

bdd.mdx previously banned "Given/When/Then ceremony" outright, which
directly contradicted WP-70's own acceptance tests (Acceptance/
BesluitLifecycleTests.cs already used // Given/When/Then comments) and
the backend's organically-evolved PascalCase_snake_sentence convention,
which the doc gave zero guidance for. Reverses that rule: every test is
now structured Given -> When -> Then, with a genuinely empty phase
omitted rather than faked; present-tense declarative naming and the
one-behaviour-per-test rule are unchanged. ADR-0006 gets a cross-reference
so both documents agree everywhere, not just in acceptance tests.

Also closes out the doc's other named-but-unenforced rules found by the
audit: fixes the 5 files asserting rendered $localize copy instead of
the underlying tag/message-id (the compliant pattern already existed in
werkvoorraad-item-view.spec.ts), splits the multi-behaviour titles the
doc itself calls a smell (";", "and", "/"), and fixes bdd.mdx's own false
citation of registratie-wizard.machine.spec.ts as "one transition per
test" by actually splitting that test into one-transition-per-test.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
eho
2026-08-18 20:25:30 +02:00
co-authored by Claude Sonnet 5
parent 306d002221
commit 3652ff8d3f
9 changed files with 247 additions and 34 deletions
+81 -9
View File
@@ -12,22 +12,53 @@ test, by layer_); BDD owns _how each test is phrased and scoped_.
## Three rules
### 1. `describe` = the subject, `it` = one observable behaviour
### 1. `describe` = the subject, `it` = one observable behaviour, structured Given → When → Then
The `describe()` block names the unit under test; each `it()` states a single behaviour in
**declarative present tense** — the implicit subject is "it". No `should`, no
Given/When/Then ceremony: present-tense declaration already reads as a spec.
**declarative present tense** — the implicit subject is "it" (no `should`). Present-tense
naming and Given/When/Then structure are not in tension — the _title_ stays a declarative
one-liner; the _body_ is what's organised as Given → When → Then:
```ts
describe('parsePostcode', () => {
it('normalises to "1234 AB" (uppercase, single space, trimmed)', () => { … });
it('rejects malformed input', () => { … });
it('normalises to "1234 AB" (uppercase, single space, trimmed)', () => {
// Given a postcode with mixed case, extra whitespace, and no gap before the letters.
// When it is parsed...
const result = parsePostcode(' 1234ab ');
// Then it comes back normalised.
expect(result).toEqual(ok('1234 AB'));
});
it('rejects malformed input', () => {
// (no Given — the input itself IS the setup) When a non-postcode string is parsed...
// Then it is rejected.
expect(parsePostcode('nope').ok).toBe(false);
});
});
```
Read top-to-bottom it _is_ the spec: "parsePostcode — normalises to 1234 AB; rejects
malformed input."
**A genuinely empty phase is omitted, not faked with an empty comment.** The rejection test
above has no Given worth writing — the malformed literal passed to `parsePostcode` already
is the setup — so it degenerates straight to When/Then. Never write `// Given (nothing)` to
keep three comments lined up; an omitted phase is the correct, honest shape for a test that
doesn't need it. The three phases stay in order (Given before When before Then) whichever
of them are present.
**This reverses this doc's earlier advice** ("No … Given/When/Then ceremony") — the team
decided explicit G/W/T structure earns its keep as the default for every test, not just
acceptance tests. What doesn't change: no `should`, present-tense titles, one behaviour per
test, ubiquitous-language naming (rules 2–3 below).
**The Elm-machine naming style is a sanctioned form of rule-1 naming, not an exception to
it.** A store/reducer spec titled after the `Msg` tag it drives —
`it('BriefLoaded moves loading to loaded', …)` — names the domain event the same way the
reducer's own `switch (msg.tag)` does; the tag IS ubiquitous language for a state machine,
so this reads as a present-tense behaviour statement exactly like `'rejects malformed
input'` does, not as a violation of rule 3.
### 2. One behaviour per test
A test asserts **one behaviour**, not one `expect()`. Several assertions that pin down the
@@ -66,10 +97,51 @@ it('confirmed dutch proficiency requires taalvaardigheid proof', …);
richest specs; the wire boundary is tested as "rejects malformed input", the UI as
Storybook stories.
## C#/xUnit shape
The three rules above are language-agnostic; xUnit follows them with its own idiom rather
than Vitest's `describe`/`it` nesting:
- **The method name is the title, in `PascalCase_snake_sentence`** — the same present-tense,
ubiquitous-language behaviour statement as a `describe`+`it`, folded into one identifier
because xUnit has no nested-description syntax: `Only_open_statuses_are_decidable`,
`Afwijzen_requires_a_toelichting`, `A_terminal_besluit_is_frozen`.
- **`// Given` / `// When` / `// Then` comments mark the three phases inside the test body** —
the same structure as rule 1, made explicit because C# has no BDD framework layered on
xUnit here (see ADR-0006 — the language's own test framework plus the builder is enough,
deliberately not a Gherkin runner). As in TypeScript, an empty phase is omitted rather than
commented for its own sake.
- **Fixtures go through the `Given` type-state builder** (ADR-0006 §1), never a field-by-field
object initializer — keeping the Given phase itself honest about which states are
reachable.
```csharp
[Fact]
public void A_terminal_besluit_is_frozen()
{
// Given a case already decided Goedgekeurd — terminal, per BeoordelingRules.CanDecide.
var aanvraag = Given.Concept(type: "registratie").Submitted().Decided(Besluit.Goedkeuren).Build();
Persist(aanvraag);
// When a behandelaar tries to record a further besluit on it...
var (outcome, updated) = ApplicationStore.RecordBesluit(aanvraag.Id, Besluit.Afwijzen, "te laat", DateTimeOffset.UtcNow);
// Then the write is refused, and the original decision still stands.
Assert.Equal(ApplicationStore.RecordBesluitOutcome.Conflict, outcome);
Assert.Null(updated);
}
```
See `Acceptance/BesluitLifecycleTests.cs` for the canonical shape (it already does this) and
`AuthzTests.cs` for the truth-table naming convention this predates — a `[Theory]` row set
stays one behaviour (rule 2's "loop asserting one rule over many inputs"), so it doesn't need
per-row G/W/T comments, just one clear method name.
## Where to look
Canonical behaviour specs in the repo: `registratie/domain/value-objects/postcode.spec.ts`
(parser behaviour), `registratie/domain/registratie-wizard.machine.spec.ts` (one transition
per test), and backend `AuthzTests.cs` (rule truth-tables). The
[Testing strategy](?path=/docs/foundations-testing-strategy--docs) page maps which layer
gets which kind of test.
(parser behaviour), `registratie/domain/registratie-wizard.machine.spec.ts` (the
message-driven `describe` block — one reducer transition per test), and backend
`Acceptance/BesluitLifecycleTests.cs` (G/W/T-commented behaviour tests) and `AuthzTests.cs`
(rule truth-tables). The [Testing strategy](?path=/docs/foundations-testing-strategy--docs)
page maps which layer gets which kind of test.