Local type aliases

[jiripudil]

Resolves phpstan/phpstan#3001. Depends on phpstan/phpdoc-parser#62.

Since Psalm only seems to support class-scoped type aliases, I did the same here. As I've said before, "it seems to be the easiest direction to take and is imo a good first iteration" that neatly follows the 80/20 rule.

I think the implementation still has some rough edges, but it appears to work surprisingly well 😅 looking forward to your thoughts and remarks!

[ondrejmirtes]

Thank you, this looks very promising. I'm also doing everything with "80/20" in mind 😊

[ondrejmirtes]

Hi, I'm gonna rebase this myself as I just implemented nested generic bounds on master (e671cc0) and I caused quite a few conflicts for you. Stay tuned :)

[ondrejmirtes]

I rebased this and added one commit that fixes prefix priority. I really like this! :)

Another test idea: Please try to use an alias right after defining it/importing in class docblock, for example in a @method tag.

[ondrejmirtes]

Oh, wrong rebase, will fix the problem :)

And please, go through the review one-by-one and add each fix as a new commit :)

[ondrejmirtes]

Alright, giving the branch back to you :) I don't know about the memory limited exceeded in the few failing jobs - it's probably a coincidence that it happened in this branch.

[ondrejmirtes]

Maybe the memory is leaking somewhere after all...

[jiripudil]

So, I've fixed some of the issues from your review. I'd say the most notable change is 210cb62 in which I've wrapped every type alias declaration in a TypeAlias class along with its name scope, allowing it to be resolved lazily later. This works around a couple of issues with nested and imported aliases.

I'll try and take a look at the memory issues too, they don't seem to be just a coincidence 🤔

[ondrejmirtes]

Looking forward to this being finished, it looks awesome already :)

[ondrejmirtes]

While thinking about ObjectType and IdentifierTypeNode, I realized unwanted behaviour:

<?php

/** @phpstan-type int \stdClass */
class Fooooo
{

	/** @param int $a */
	public function doFoo($a)
	{
		\PHPStan\dumpType($a); // should be int, but current is stdClass
	}

}

I'm gonna solve it by deleting TypeAliasesTypeNodeResolverExtension and doing the logic in the right place in TypeNodeResolver::resolveIdentifierTypeNode() (after things that are never supposed to be aliased are handled). It should have performance benefits as well.

[ondrejmirtes]

Is this fine? @jiripudil

[jiripudil]

Yes, looks fine.

[ondrejmirtes]

Alright, I think we're green :)

I have two more things to ask from you:

1. Please add another condition checked in the rule - that @phpstan-type int ShouldNotHappen isn't allowed. I think you could pass the int portion to the TypeNodeResolver::resolve and it should result in ObjectType.
2. Please send a PR that edits https://github.com/phpstan/phpstan/edit/master/website/src/writing-php-code/phpdoc-types.md and adds documentation about local type aliases :)

Thank you.

[jiripudil]

Glad to hear that :) I'll add the check and the docs.

Thank you for your help!