Merge guessed argument properties and args described with GQL\Arg

[mathroc]

Q	A
Bug fix?	yes/no
New feature?	yes/no
BC breaks?	yes/no
Deprecations?	no
Tests pass?	yes/no
Documented?	no
Fixed tickets	#...
License	MIT

When a single argument is described with #[GQL\Arg] all other are not guessed anymore

This commit starts by guessing all arguments and override the ones that have #[GQL\Arg] attributes describing them

I'm not sure of a few things:

• should the type specified in the attribute override the one that was guessed?
• same for the default
• Is there cases where guessing could fail and adding the #[GQL\Arg] attribute was a way to prevent guessing failing / crashing?

in the future it would be nice if we could add the attribute on the argument directly instead of the method so we could make the name & type optional there (or forbidden?)

[Vincz]

@mathroc to answer your questions:

• Should the type specified in the attribute override the guessed one => YES. The developper should have the final word regarding the argument type. So, #[GQL\Arg] should be the priority.
• For the default as well.
• We auto-guess only if we don't have a #[GQL\Arg]

You are right, we could also explore the possibility to set attributes directly on the arguments.

[mathroc]

make the attributes have the final word on type and default value

[mathroc]

and for the last question, I can answer it by looking at the tests: yes sometimes the guessing can't work at all, e.g.: arrays

🤔

[Vincz]

Yes, and we want an error when we don't have a corresponding #[GQL\Arg] and the auto-guess failed.
To make this work, you first need to get arguments having a corresponding#[GQL\Arg] and auto-guess only the remaining ones (by passing the list of already resolved to the auto-guesser for example).

[mathroc]

How should the arguments be ordered? if it has any importance)

[Vincz]

@mathroc The GraphQL arguments order doesn't matter as they are named. But we need them in the right order when the PHP method is called.

[mathroc]

ok, I think the PR is fine then, let me know if there's anything to do. (should I squash the commits?)