gh-104683: Argument Clinic: Extract parse_function_names() helper

[erlend-aasland]

Reduce duplicate code in state_modulename_name()

• Issue: Modernise code in Tools/clinic/ #104683

[erlend-aasland]

This refactor helped me discover a regression of gh-107885; Argument Clinic does not fail() for these cases:

/*[clinic input]
class C "void *" ""
C.meth
  a: int
[clinic start generated code]*/
/*[clinic input]
# This should not be allowed!
C.__new__ = C.meth
[clinic start generated code]*/
/*[clinic input]
# Neither should this
@classmethod
C.__init__ = C.meth
[clinic start generated code]*/

We should bake those assertions into the parse_names helper.

[erlend-aasland]

Post this, I want to further extract helpers from this stage of the state machine; with this refactor, the remaining code is clearer (simply, since it is fewer lines of code), and it is now obvious we need to do something with the _module_and_class() calls :)

[AlexWaygood]

it is now obvious we need to do something with the _module_and_class() calls :)

Off-topic, but I'm pretty sure this block of code is unreachable FYI:

cpython/Tools/clinic/clinic.py

Lines 2440 to 2441 in 13c36dc

	if not hasattr(parent, 'classes'):
	return module, cls

It's not currently covered by any of our tests. If our type annotations are correct, moreover (and I believe they are), parent should only ever be an instance of Clinic, Module or Class. Instances of Clinic, Module and Class all have a .classes attribute. (Mypy likely doesn't catch this because it doesn't have a great understanding of hasattr(), which is a bit too dynamic for it.)

[erlend-aasland]

There are some corner cases which I think should be fixed:

• The regression from gh-107880: Teach Argument Clinic to clone __init__ and __new__ methods #107885
• gh-104683: Argument Clinic: Extract parse_function_names() helper #107964 (comment)

I suggest we create an issue for this, and retarget this PR as a fix for these issues.

[AlexWaygood]

The knot of code in #107964 (comment) is such a complex conditional (and, it appears, apparently buggy as well) that I think I'd prefer to see a dedicated PR untangling just that bit, and fixing the bug, before we go ahead with a change that factors out the common logic into a helper function. Would that be okay?

[AlexWaygood]

Thanks!

[AlexWaygood]

Oh, one optional other suggestion:

[erlend-aasland]

Thanks for the review!

[AlexWaygood]

🎉

[erlend-aasland]

I'll follow-up with further refactorings in this part of the source code; there's still a lot of duplicated code there.