Prevent "unable to resolve template type T" in conditional branches

[rvanvelzen]

Part of phpstan/phpstan#3853

This prevents the error for template types that are not part of both conditional branches.

It isn't perfect, but I couldn't figure out how to decide which branch would be taken. There is still the error regarding the incorrect passed type in many cases (as shown in the added test case), so it might not be a big deal.

[ondrejmirtes]

I have an idea, I'm gonna try something :)

[ondrejmirtes]

So I tried something - I thought that we should use the same logic to resolve the conditional type as it's currently used for other reasons. The problem was that we're also resolving template types at the same type which caused the logic in FunctionCallParametersCheck to not see what was originally there.

I divided this in two steps - first resolve the conditionals (that's what the "original parameters acceptor" contains), and only then resolve the template types.

[ondrejmirtes]

I've attempted to do an additional commit - a cleanup which I will push now, but it breaks some cases - probably because the conditional is resolved at the wrong time and the template type needed for that is not yet known.

[ondrejmirtes]

Could you please try to fix what's wrong with 478aa27?diff=unified&w=1? Or just tell me it's impossible and the previous commit is fine. But I don't like that this code is duplicated and executed twice:

			if ($type instanceof ConditionalTypeForParameter || $type instanceof ConditionalType) {
				$type = $traverse($type);

				if ($type instanceof ConditionalTypeForParameter && array_key_exists($type->getParameterName(), $passedArgs)) {
					$type = $type->toConditional($passedArgs[$type->getParameterName()]);
				}

				if ($type instanceof ConditionalType) {
					return $type->resolve();
				}

				return $type;
			}

[rvanvelzen]

Yeah, the issue there is that the conditional needs the template types to be resolved to know which branch to take. However, ConditionalTypeForParameter will always be resolved after that (or not be resolvable at all), so only ConditionalType needs to be resolved. I agree that it's not nice though

[rvanvelzen]

I have some ideas, will update soon

[rvanvelzen]

Naming aside this seems like a reasonable way forward to me

[ondrejmirtes]

I really like this! Please mark this as ready, I'll merge it :)

[ondrejmirtes]

Thank you!