Settings in tests
Is this package for you?
17 reasons, each a before and an after. Select the problems you have and see how many match.
Change one value without restating the other seven
A test that cares about one timeout should not have to name every other property to get there — and it should not break the day someone adds an eighth.
$this->app->singleton(AuthSettings::class, fn () =>
new AuthSettings(
domain: 'test.example.com',
redirect_url: 'http://test.example.com/cb',
timeout: 5, // the only one this test cares about
mfa_enabled: false,
));Leave one out and you get ArgumentCountError.
AuthSettings::fake(['timeout' => 5]);
// bound as the container singleton for you —
// the other properties keep their real values,
// and a new one added next month changes nothing hereThe test states its subject and nothing else.
The real cost of the left-hand column is later. Add a property to the class and every test that swapped it breaks at once — none of which cared about the new property. That is a tax on exactly the discipline this package exists to encourage.
Three ways in
Pick a call to see what it returns. Every value below came from running it against this demo's own
AiSettings.
AiSettings::resolve()what the app gets, honouring APP_ENV
| provider | openai | unchanged |
| text_model | gpt-4o | unchanged |
| max_tokens | 8000 | unchanged |
AiSettings::from('staging')any environment, whatever APP_ENV says
| provider | openai | unchanged |
| text_model | gpt-4o-mini | differs from resolve() |
| max_tokens | 2000 | differs from resolve() |
AiSettings::fake(['max_tokens' => 10])one value changed, the rest left alone
| provider | openai | unchanged |
| text_model | gpt-4o | unchanged |
| max_tokens | 10 | differs from resolve() |
AiSettings::fake(['provider' => 'azure'], 'development')start from another environment, then change one value
| provider | azure | differs from resolve() |
| text_model | llama3.2 | differs from resolve() |
| max_tokens | 1000 | differs from resolve() |
AiSettings::from('production')->with(['temperature' => 0.0])derive a copy from an instance you already hold
| provider | openai | unchanged |
| text_model | gpt-4o | unchanged |
| temperature | 0.0 | differs from resolve() |
What each one is for
from(string $env)
Resolve any environment on demand
Ignores APP_ENV entirely. Useful for asserting that production really does use the values you think it does — from a test running under testing.
fake(array $props, ?string $from)
A double with only the differences named
Starts from the current environment, or the one you name, replaces just the properties you list, and binds the result as the container singleton.
with(array $props)
Derive a copy from an instance
Called on an object you already hold rather than on the class, so you can layer a change onto anything — including the result of from().
They return the real class
Not a mock and not a stub — an actual AiSettings, fully typed,
so anything type-hinting the class accepts it and your assertions read normally.
They compose
from('production')->with([...]) reads as one sentence:
take production, change this. Each call returns a new instance rather than mutating.
In a test
it('stops retrying after the configured limit', function () {
PaymentSettings::fake(['retry_attempts' => 2]);
// mode, currency and webhook_url keep their real values
expect(attemptPayment())->toHaveBeenTried(2);
});