Fix 130

[herndlm]

Closes https://github.com/phpstan/phpstan-webmozart-assert/issues/118 Closes https://github.com/phpstan/phpstan-webmozart-assert/issues/119 Closes https://github.com/phpstan/phpstan-webmozart-assert/issues/130

Looks like the rootExpr approach is potentially making problems if the same thing is asserted twice in a row, where previously it would detect the second one as always evaluating to true/false. But this could be related to the all* implementation here too.

[ondrejmirtes]

Looks like the rootExpr approach is potentially making problems

Can you show the false positive here?

[herndlm]

It's just that https://github.com/phpstan/phpstan-webmozart-assert/blob/1.1.2/tests/Type/WebMozartAssert/data/impossible-check.php#L26 is not reported anymore, but maybe I need to adapt something else here

[herndlm]

@ondrejmirtes I took another look to find out why that assertion is not reported anymore. The step debugging revealed the following for the second assertion:

What baffles me is the scope state. $b is already correct, the array_filter(... is correct (the same as $b), but the $b === array_filter(.. expression resolves still to a BooleanType instead of a ConstantBooleanType. Any idea where I could/should look maybe? In MutatingScope I guess? :) UPDATE: ah nevermind, looks like I got it :)

[herndlm]

It's coming from https://github.com/phpstan/phpstan-src/blob/1.6.4/src/Analyser/MutatingScope.php#L2345 but that makes sense. Since phpstan only knows that both left and right are array<mixed, Bar> it cannot know if they are equal or identical. Hmm, stuck again :/ Maybe we have to take a look at the specified types after all in some cases with $rootExpr given

[herndlm]

Came up with something new for the all* handling. Instead of the typewise-complex array_filter solution I switched to a more simple check of the first array element. Combined with the already existing arrayOrIterable helper that seems to be working fine. And helps solving the annoying impossible type checks. What do you think?

Also added 2 other regression tests, but the bugs behind them were most likely fixed by the recent typespecifier adaptions in phpstan itself 🎉

[herndlm]

I think this one's good enough, did not came up with anything better yet and it's not too weird either IMO.

[ondrejmirtes]

Thank you.