Add "DynamicFunctionReturnTypeExtension" for "array_replace" + fix for "array_merge"

[voku]

This Extension is a copy of "ArrayMergeFunctionDynamicReturnTypeExtension"

⇾ requested here: #622

INFO: Currently we are getting wrong types from "ArrayMergeFunctionDynamicReturnTypeExtension" because PHP handles array-keys different here: https://3v4l.org/YfKYq

• if the key is numeric, then the value will be added
• if the key is non-numeric, then the value will be replaced

[ondrejmirtes]

How is this different to ArrayMergeFunctionDynamicReturnTypeExtension? Are they going to differ in the future?

[ondrejmirtes]

I don't understand this part - the last inner type in the union will win in the $argType variable?

[ondrejmirtes]

I know the same code is in ArrayMergeFunctionDynamicReturnTypeExtension - it also puzzles me there.

[voku]

It makes no sense to me. Maybe this https://github.com/phpstan/phpstan-src/pull/622/files#diff-2b90d7b4bf7114d779c5be8175b43e99ebb849c71781c4285c6cc47bb6cba5d3R40 is more correct?

[ondrejmirtes]

I don't know, you tell me :(

[ondrejmirtes]

:)

[voku]

In the last pull request we already tried to process every parameter. I could re-add it, so we can see if it works, but the performance was bad, so that I will not add anything related to array shapes, ok?

[ondrejmirtes]

Let's try and see what happens :)

[voku]

I re-added the fixes and added a limit for the array. I also added the fix for "array_merge" here, so we can see the differences between the implementations in one diff.

[voku]

How is this different to ArrayMergeFunctionDynamicReturnTypeExtension? Are they going to differ in the future?

If we check the type of the key, then we could implement the correct behavior in the future.

[ondrejmirtes]

Please add a regression test for phpstan/phpstan#5846 here and I'll take a look :) Thank you.

[voku]

Please add a regression test for phpstan/phpstan#5846 here and I'll take a look :) Thank you.

I re-added some changes, even more tests and added some checks to prevent performance issues, but maybe we need to adjust the consts "MAX_*" settings? I already tested the performance with a private project and blackfire.io and for me, it looks good, but as airways it depends on the tested code.