Retain left and right types in equal expression

[herndlm]

This is not really changing much.. It's an adaption very similar to #1021

Equal expressions are, among other things, converted to Identical expressions if possible for the left and right types. If not, then no types at all are specified. This change modifies that, so that at least the left and right types are retained.

The first new test case was already working (because $int == (int) $int) is safe to be converted to an Identical expression. The second new test case is affected by this change, previously no types at all were specified. I'm mostly doing this so that we have more type information and strict impossible checks, as reported in e.g. phpstan/phpstan-webmozart-assert#32, do not fail

[ondrejmirtes]

Maybe I'm too dumb today but I need to see what this changes via a NodeScopeResolverTest case :)

[herndlm]

I'm also not sure yet if this makes sense, I'm still playing around with it and need some more time. Should have waited with the PR a bit longer

[ondrejmirtes]

Alright, why does this work? 😂 Where's the exception to treat == and === differently for objects?

[herndlm]

Yeah, let me get back to that at a later time, I can't figure this out today I'm afraid
maybe it is an indication that this all is horribly wrong? :D

[ondrejmirtes]

Maybe first open a PR that just adds the test cases that already work on master (and comment those that do not work). Then the situation will be more transparent maybe.

[herndlm]

I'm not quite sure I understand what the problem is? What do you think should be different here in regad to == vs ===? To me it still makes sense, but maybe I'm overlooking something.
The current difference to master are those 2 *NEVER* per file.

[ondrejmirtes]

It makes sense but I have no idea why it works.

[herndlm]

AFAIK $a === $a and $a == $a return no sure and sureNotTypes on master at all. But here it returns at least that $a is stdClass. Which is already known in the scope of course, but now it can be used in a falsey scope for the if and turned into sureNotTypes. I think. I'd need to double-check with the debugger again. But I'm pretty sure I saw something like that.

[ondrejmirtes]

Ah, so the same expression on left/right creates a contradiction inside the SpecifiedTypes leading to NeverType in the else branch?

[ondrejmirtes]

That'd make sense :)

[herndlm]

I did confirm this now.

For the true scope / if branch we get the stdClass ObjectType as left and right sureTypes. And those are unionized. No surprises there.

For the false scope / else branch we get the stdClass ObjectType as left and right sureNotTypes. And those are normalized to NeverTypes (using your improved normalization which makes use of the scope and removes stdClass from stdClass) which are then intersected.

[ondrejmirtes]

Great, I'll merge this if you mark it as ready :)

[ondrejmirtes]

Thank you.