Use root expression when checking impossible types

[herndlm]

This is the implementation of the idea that Ondrej mentioned in phpstan/phpstan-webmozart-assert#130 (comment)

I'm not quite sure about details yet, it looks like it makes phpstan already smarter though. It also looks like phpstan now remembers more specified types as stated in https://phpstan.org/blog/remembering-and-forgetting-returned-values.

This PR is mostly to find out how much will break 👀

[herndlm]

Tbh, I was even expecting more errors. But it looks like it's doing too much now, not sure. E.g. it complains in https://github.com/phpstan/phpstan-doctrine/blob/1.3.3/tests/Type/Doctrine/Query/QueryResultTypeWalkerTest.php#L126 with

  ------ --------------------------------------------------------------------- 
  Line   tests/Type/Doctrine/Query/QueryResultTypeWalkerTest.php              
 ------ --------------------------------------------------------------------- 
  126    Call to method                                                       
         Doctrine\Common\Collections\ArrayCollection<*NEVER*,*NEVER*>::add()  
         with QueryResult\Entities\Many will always evaluate to true.         
 ------ --------------------------------------------------------------------- 

Not sure if this should even be looked at in the first place?

[ondrejmirtes]

Some jobs didn't run because I misunderstood the documentation of GitHub Actions. Hopefully it's gonna be better now: 3bf2958

[ondrejmirtes]

This looks clean and promising. Could you please try to simulate the situation that I outlined here? phpstan/phpstan-webmozart-assert#130 (comment)

$str !== '' && SOME_FAUX_FUNCTION($str)

With this improvement we'll be able to write extensions for functions like "ipv4"...

[herndlm]

the 10 webmozart failures are actual fixes, or respectively cases were phpstan got smarter :)
not sure about the remaining composer failures, I'm still looking at those. UPDATE: they might be mostly OK too, not sure about some details though. some things actually surprised me that phpstan understood them :D

[herndlm]

This looks clean and promising. Could you please try to simulate the situation that I outlined here? phpstan/phpstan-webmozart-assert#130 (comment)

$str !== '' && SOME_FAUX_FUNCTION($str)

With this improvement we'll be able to write extensions for functions like "ipv4"...

I think it's not yet enough, I tried with #1068 here and with phpstan/phpstan-webmozart-assert#130 itself. I was running out of time, but I think such an expression would still resolve to a BooleanType and then my adaptions don't trigger anything and it falls back to the old behaviour. I'll continue later

[ondrejmirtes]

BTW I'm trying afew more commits above your PR: #1256

[herndlm]

Thx for the help Ondrej! I rebased on latest 1.6.x already, apparently ImpossibleCheckTypeMethodCallRuleEqualsTest still needs fixing, but we're getting closer.

[herndlm]

OK, a first local check looks like you did it. great job! :)
I'll add the new things it found and push when it's ready local. UPDATE: you did that too xD ok then, I'll cherry-pick that commit too

[ondrejmirtes]

Please let's not step on each other's toes right now, I'm very close to being green here and merging it :)

[herndlm]

just cherry-picking your changes xD but yeah, sorry for hogging CI, every push I thought you're done

[herndlm]

there's a new legit webmozart thing it found, but a couple other loosely-equal checks are now not reported anymore. is that fine? Let me know what to adapt

[ondrejmirtes]

Thank you :) I'm gonna wait for the issue bot and then go to bed :) And fix the integration tests in phpstan/phpstan tommorrow.

[ondrejmirtes]

a couple other loosely-equal checks are now not reported anymore

A bigger mystery for me was that they actually were reported before. There's no logic in PHPStan to say that 1 == true is true.

[herndlm]

I think I thought the same while adding them. weird. ok then

[ondrejmirtes]

So it's fine to remove those ::eq asserts-reported errors from the tests.

[herndlm]

than that shall be my last action for the day

[herndlm]

sorry, one more thing, it looks like true == true is also not reported any more. does that one need fixing?

[ondrejmirtes]

Sure, should be covered here :)

phpstan-src/src/Analyser/MutatingScope.php

Lines 2353 to 2358 in 6bf335c

	if (
	($stringType->isSuperTypeOf($leftType)->yes() && $stringType->isSuperTypeOf($rightType)->yes())
	|| ($integerType->isSuperTypeOf($leftType)->yes() && $integerType->isSuperTypeOf($rightType)->yes())
	|| ($floatType->isSuperTypeOf($leftType)->yes() && $floatType->isSuperTypeOf($rightType)->yes())
	) {
	return $this->resolveIdenticalType($leftType, $rightType);

[ondrejmirtes]

Wow, I'm shocked, the issue bot haven't found a single change in the issues :) But at least we haven't broken anything :)

[herndlm]

I hope this opens the door for furhter type specifier cleanups at least. or features

[ondrejmirtes]

Yeah, I think so, you can try it out in webmozart/assert extension :) Also this one should be possible now #1068