Variadic support for #[TestWith] attribute

[davidbyoung]

Loving PHPUnit 10, thank you so much for it! I've switched a lot of my @dataProvider annotations to use the #[TestWith] attribute, but I have found the fact that it takes an array of parameters rather than variadic parameters to be awkward. For example, I think this is cleaner:

#[TestWith(20, 21, false)]
#[TestWith(22, 21, true)]
public function testAgeRequirement(int $age, int $minimumAge, bool $expectedToPass): void
{
    $this->assertSame($expectedToPass, $age >= $minimumAge);
}

...than this:

#[TestWith([20, 21, false])]
#[TestWith([22, 21, true])]
public function testAgeRequirement(int $age, int $minimumAge, bool $expectedToPass): void
{
    $this->assertSame($expectedToPass, $age >= $minimumAge);
}

...because the former almost looks like an invocation of the test method, which to me feels intuitive. I feel the latter looks like you're passing in array as a parameter. Since TestWith::__construct() does not take in any other parameters, I think it would be a better developer experience to take in a variadic parameter instead. Obviously, this would be a breaking change, which might mean we should use a new attribute instead. Using a variadic parameter would also complicate supporting a feature such as named data sets. I'm curious what your thoughts are on this and whether we should support variadic parameters or introduce a new attribute to prevent the breaking change. If so, I'll contribute a pull request.

I should note that NUnit and xUnit both use the variadic parameter approach.

[sebastianbergmann]

I would consider a PR for this, but only if introduces a new/separate attribute for this and supports named data sets (first parameter of type string for the name followed by ...mixed will likely work). Right now I do not have a good name for the new attribute, though.

[davidbyoung]

Sounds good. I'll start a PR. Again, thanks for 10.0!

[davidbyoung]

One thing - I noticed that PHUnit typically uses the @no-named-arguments Psalm annotation. Are you open to using a named argument with the new attribute, eg:

#[TestData('foo', 'bar', name: 'Test with foo and bar')]
public function testSomething(string $a, string $b): void
{
    // ... 
}

Is this a non-starter and I should find another way, or should I continue with such an approach? I think it would be more declarative and reduce unintended side effects of always interpreting the first string parameter as a name.