support removing class-strings from GenericClassStringType

[staabm]

refs phpstan/phpstan#7698 (comment)

[ondrejmirtes]

Please also add a regression test for the match expression, there's also some example with $obj::class I think.

[Dgame]

Here is a smaller example: https://phpstan.org/r/e4476215-99ea-434f-adae-3564ef93a5a5

If this works I really have to take a look how the semantics work, I'm not sure I understand the NeverType there.

[Dgame]

Not sure if this is related in this case, but have another test case there: phpstan/phpstan#7721

[staabm]

just added the match-test. its still failling though.

I think Dgame is right, that there is a missing piece in MatchExpressionRule. I need to invest more time, to get an idea how this match-arm analysis is working...
it seems MatchExpressionRule is successfully removing all match-arm expressions from the match condition, but instead of a NeverType it leads to a empty ClassStringType atm

[staabm]

ok, after sleeping a night over it, I know understood the basic building blocks of the arm analysis.

we more or less rewrite the AST in

phpstan-src/src/Rules/Comparison/MatchExpressionRule.php

Lines 54 to 57 in 1537424

	$armConditionExpr = new Node\Expr\BinaryOp\Identical(
	$matchCondition,
	$armCondition->getCondition(),
	);

which adds similar IF-identical expressions we already have in https://github.com/phpstan/phpstan-src/pull/1590/files#diff-9ca94a9a28f6d393c869ef34bde57e4bb0d53dcc9a8cb5fc865f074ef0df5348R30-R32

with this knowledge I was able to fix the test with commit 21e5bcf.

---

still, I think the match rule contains a bug which makes it only work when the match condition is a variable.
as soon as you put directly a expression into the match, it does not work atm:

works:

	$t = new Test(new A());
	$class = $t->value::class;
	echo match ($class) {
		A::class => 'A',
		B::class => 'B'
	};

doesn't work:

	$t = new Test(new A());
	echo match ($t->value::class) {
		A::class => 'A',
		B::class => 'B'
	};

I am still figuring out the details.

[staabm]

I put my finding into a new issue, as I think this one can be merged as is?

phpstan/phpstan#7746

[Dgame]

Thanks @staabm! That a good step forward. 🙂 Could we call tryRemove in the MatchExpressionRule in case the match-condition is an expression?

[staabm]

issue phpstan/phpstan#7746 is not related to the MatchRule. its a more general problem, see e.g. https://phpstan.org/r/be1b13ce-b8d9-445e-b6ed-dc9e87f2879b

[Dgame]

@staabm Here is a more reduced example which does not work: https://phpstan.org/r/0a06314d-d25e-4435-8aa1-d60bb41b8fda
It does not seems to work if I store the class-name in a variable.

[staabm]

phpstan.org/try does not yet contain the patch, as the PR is not yet merged.
taking your example and running it against phpstan with the PR applied works for me as expected locally (no errors)

[ondrejmirtes]

TypeWithClassName pls

[staabm]

I tried that while PR creation and it made some tests fail.

Therefore I went back to ObjectType

[ondrejmirtes]

That's a red flag, we should investigate that

[staabm]

ok - seems to work now. maybe it was some intermediate step which made it fail…

[ondrejmirtes]

Thank you.

[Dgame]

Thanks a lot @staabm !