PHP 8.1: Report missing typehints in overridden native methods

[danog]

Allows detection of ReturnTypeWIllChange warnings, also fixes #7362

[weirdan]

@danog, is there any reason not to add this to 4.x line?

[danog]

Not really, no, I just wanted to switch directly to psalm v5 :D

[danog]

Hmm, that's weird, I rebased on top of 4.x...

[orklah]

There must have been something wrong. isEmpty was only removed in master

[danog]

All fixed!

[orklah]

Is this related to #6395 (comment) ?

[danog]

Yep, it's related precisely to that.

[orklah]

Yep, it's related precisely to that.

but then, in your PR, you doesn't seem to distinguish phpdoc from real types in signature: https://github.com/vimeo/psalm/pull/7363/files#diff-c70a8399e3804f1bb260231f6af972513c1ca05d21e3f4221ae35ad5a9d4e6c0R121

Matt's branch was all about adding types in signature to be able to make the difference I believe?

[danog]

Matt's branch was all about adding types in signature to be able to make the difference I believe?
Yes, but not all native classes are declared as stubs: for example, the JsonSerializable class I used in the tests is populated at runtime using reflection.

I could've changed the reflection populator to populate signature_return_type instead of return_type (and fix existing stubs), but I decided it would be easier (safer?) to just create an explicit check that simply relies on the interaction of a native parent class and a user-defined child class.

Actually, do you think it should be made a separate issue, given that it signals a rather important PHP 8.1 BC semi-break?

[orklah]

A different issue could be interesting indeed. About the signature_return_type vs return_type, did PHP add signatures types for every internal function/class/interface or did they do it only for certain cases?

[danog]

About the signature_return_type vs return_type, did PHP add signatures types for every internal function/class/interface or did they do it only for certain cases?

The documentation says Most non-final internal methods now require overriding methods to declare a compatible return type, so I just assumed all of them did.

[orklah]

@danog is this ready?

[danog]

Yep!

[orklah]

Thanks!