fix iterator_to_array, iterator_count for PHP 8.2

[schlndh]

Fixes phpstan/phpstan#9793 (and the example given in phpstan/phpstan#7760, though that issue is specified more broadly).

The problem happens here:

phpstan-src/src/Reflection/SignatureMap/Php8SignatureMapProvider.php

Line 198 in 02ea303

if ($this->functionSignatureMapProvider->hasFunctionSignature($functionName)) {

. We get the correct signature from php8-stubs, but the signature in functionMap.php was outdated and it overwrote the the correct signature.

[ondrejmirtes]

We could have done this, but it feels just like a workaround.

The facts:

• php-8-stubs have the correct native parameter type for PHP 8.2: https://github.com/phpstan/php-8-stubs/blob/main/stubs/ext/spl/iterator_count.php
• Types from functionMap.php are considered as PHPDoc types, and in case of a clash (PHPDoc type is not a subtype of the native type), the PHPDoc type is thrown away. This logic is in TypehintHelper::decideType() that's called in Php8SignatureMapProvider.
• There are other functions that have the wrong type in functionMap, the correct native type in php-8-stubs, and they work correctly. For example curl_init: https://phpstan.org/r/436b8a3d-4f12-401e-8974-266ea83eec3e

This is why we need the actual fix in the logic and not a workaround :) Because more functions might be impacted by this bug, if not now, then certainly in the future.

[ondrejmirtes]

Thinking about the actual reason, I think the problem might be that functionMap has Traversable but the correct type is iterable. But iterable->isSuperTypeOf(Traversable) is yes(). At the same time PHPDoc types can actually be used to narrow down the type. Think of native int with PHPDoc's positive-int.

iterable is actually array|Traversable, but in PHPStan there's a special IterableType implementation.

So maybe we need to special-case this problem in TypehintHelper. If we have iterable in the native type and Traversable in the PHPDoc, iterable should still win.

[schlndh]

@ondrejmirtes Ok, thanks for the pointers. I'll try look into the underlying problem. But I guess it wouldn't hurt to fix the functionMap.php regardless.

[ondrejmirtes]

I'm looking forward to this one! The reflection golden test should make this a breeze, we'll see all the implication of the fix :)

[schlndh]

I tried adding a hack to decideType instead of fixing the signature map. According the the reflection golden test the only two currently affected native functions/methods are these two. So it doesn't seem worth it to me to add such a weird edge-case to decideType. Also, the change broke existing test - i.e. it could have negative impact on non-native code.

You can see the experiment here: schlndh#3

[ondrejmirtes]

Awesome, thank you! It's great we can now do these changes in informed way.