Fix arginfo generation from stubs for namespaced functions

[alcaeus]

This PR fixes two issues:
The first commit ensures that the @deprecated and @alias annotations are respected for namespaced functions as well as for those without a namespace.
The second commit fixes an issue where having functions with the same name in multiple namespaces would generate two identical ZEND_FUNCTION entries. Note that this also significantly refactors the code used to generate function arginfo to reduce duplication.

Changes don't affect any existing arginfo files in core, as none of the default extensions use namespaced functions. However, the changes could cause issues in downstream code, namely if an extension uses stub files to generate arginfo for namespaces functions. I'm not sure how to best deal with this and would like some feedback as to what kind of BC breaks we can make in the build scripts.

Last but not least, there is one remaining edge case which I deliberately did not solve. Consider the following stub file:

namespace Foo {
    function Bar_baz(): void {}
}

namespace Foo\Bar {
    function baz(): void {}
}

In this case, we would once again create two identical ZEND_FUNCTION entries in the arginfo file. This could also happen with classes as long as they are defined in the same stub file, as the script always uses a single underscore to replace a backslash in the fully qualified name without treating the namespace and member name differently. However, due to the extensive refactoring involved to fix this as well as the low probability of that happening, I decided to blatantly ignore that issue for the time being.

[kocsismate]

Regarding the issue with Foo\Bar\Baz and Foo\Bar_Baz: I also found this edge case, and I also decided not to fix it: c345b6e#r71259703

[alcaeus]

Yes - barring using characters not allowed in function names there will always be the chance of conflict. Not worth it IMO 🤷‍♂️

[kocsismate]

I think this a very nice and useful fix. The BC break is unfortunate, but it's for the good, and probably it's not difficult to address in other 3rd party code.

[jmikola]

I think this a very nice and useful fix. The BC break is unfortunate, but it's for the good, and probably it's not difficult to address in other 3rd party code.

Does this PR warrant an update to UPGRADING.INTERNALS, or is that typically done later on and closer to release?

[kocsismate]

Does this PR warrant an update to UPGRADING.INTERNALS, or is that typically done later on and closer to release?

Yes, it does, and the UPGRADING.INTERNALS file is updated continuously.

[kocsismate]

I'll wait with the merge for the opinion of other reviewers until next Tuesday the latest (before RC1).

[alcaeus]

Yes, it does, and the UPGRADING.INTERNALS file is updated continuously.

Added the following upgrade note:

Identifier names for namespaced functions generated from stub files through gen_stub.php have been changed. This will require changes in extensions using namespaced funtions.

Any wording suggestions are highly appreciated.

[kocsismate]

Added the following upgrade note:

FYI: generally, any upgrading note is added by the one who applies the commit in order to prevent conflicts due to simultaneous updates. I think though that any other change in the upgrading file is highly unlikely at this point, so I'm fine with having it in the PR :)

[alcaeus]

FYI: generally, any upgrading note is added by the one who applies the commit in order to prevent conflicts due to simultaneous updates.

TIL - I'll consider it target practice should I find myself in that situation at some point in time ;)

[cmb69]

Thanks for the PR. The implementation looks good, and apparently this is a necessary evil; still might cause quite some PITA for extension maintainers.

@remicollet may want to review (although he's AFK, AFAIK).

[kocsismate]

I think though that any other change in the upgrading file is highly unlikely at this point, so I'm fine with having it in the PR

Reality has just kicked in :D so please fix the conflict.

[kocsismate]

I fixed the conflict and slightly adjusted the wording. Feel free to change it should you find any issues with it.

[kocsismate]

There was a problem with the aliases. If you run ./build/gen_stub.php --verify --force-regeneration then gen_stub.php complains about the inexistent alias functions (zend_test_alias, zend_test_deprecated_alias etc.). So I fixed them in ef21bbe