Fix array_filter with non-empty result

[enumag]

phpstan/phpstan#3752

[ondrejmirtes]

This isn't what I suggested. Keep the test but try to fix ConstantArrayType::generalize() implementation instead. It just needs to do TypeCombinator::intersect(new ArrayType(...), new NonEmptyArrayType()); instead on the last line.

[enumag]

When looking more into it I realized that your suggestion is actually incorrect. I added a test which would break in your case:

function dummy4(?DateTimeInterface $date): void { assertType('DateTimeInterface|false', min(array_filter([$date]))); }

[ondrejmirtes]

My suggestion is necessary, non empty constant array needs to be generalized into a non empty general array. But it might not be sufficient to fix this bug and something else needs to be done in the extension too.

[enumag]

Alright. Check this commit please if it is what you want.

It will most likely break the ArrayFilterFunctionReturnTypeReturnTypeExtension again so I'll look into it. Just wanna know if the ConstantArrayType fix is correct.

[enumag]

It broke a bunch of other tests though...

[enumag]

Ah. I get it now. Sorry, I was doing it in a hurry. I'll get back to it later.

[ondrejmirtes]

Yeah, so the actual fix in ConstantArrayType::generalize() surfaces some other bugs that have to be looked into. Unfortunately I don't have time to look into this now so if no one is able to fix this on their own to get to a green build, it will have to wait until it's a priority for me. I want to look into the backlog of Easy fixes before 1.0 comes out.

[enumag]

In that case would you consider merging my changes to ArrayFilterFunctionReturnTypeReturnTypeExtension without fixing the ->generalize() here?

[enumag]

I tried but... I'm afraid that I can't fix the ConstantArrayType::generalize() method and make everything green. I'm fairly confident that my first commit alone is fine though. And it does fix the issue I want to fix.

[enumag]

Simply put, the ArrayFilterFunctionReturnTypeReturnTypeExtension needs to use the old implementation of generalize() because the ->isIterableAtLeastOnce() result is not reliable and the extension needs to decide it on it's own. But when I do change the generalize() method I'm unsure how to fix the extension.

[ondrejmirtes]

Superseded by: d67dae3