bpo-36974: Make tp_call=PyVectorcall_Call work for inherited types

[encukou]

When inheriting a heap subclass from a vectorcall class that sets .tp_call=PyVectorcall_Call (as recommended), the subclass does not inherit _Py_TPFLAGS_HAVE_VECTORCALL, and thus PyVectorcall_Call does not work for it.

This attempts to solve the issue by:

• always inheriting tp_vectorcall_offset unless tp_call is overridden in the subclass
• inheriting _Py_TPFLAGS_HAVE_VECTORCALL for static types, unless tp_call is overridden
• making PyVectorcall_Call ignore _Py_TPFLAGS_HAVE_VECTORCALL

This means it'll be ever more important to only call PyVectorcall_Call on classes that support vectorcall. In PyVectorcall_Call's intended role as tp_call filler, that's not a problem.

cc @jdemeyer (not Mark, as he seemed to want to stay out of subclassing – but he's nosy on the issue)

https://bugs.python.org/issue36974

[jdemeyer]

I don't have much time to look into this, but your idea sounds reasonable.

[jdemeyer]

Reviewing this now after the fact... looks mostly good to me.

One potential issue: it is safer to inherit tp_vectorcall_offset unconditionally, even if tp_call is overridden. This allows a custom tp_call implementation falling back to the tp_call of the base class. Essentially the C equivalent of

def __call__(self, *args, **kwds):
    # do something interesting here
    return super().__call__(*args, **kwds)

Do you agree?

And one minor detail: in _PyVectorcall_Function(), the check Py_TYPE(callable)->tp_call is an assertion assert(PyCallable_Check(callable)); so I wonder why you made it a real check here.

[encukou]

it is safer to inherit tp_vectorcall_offset unconditionally

Yes, that's a good point. Do you want to write the test for that?

in _PyVectorcall_Function(), the check Py_TYPE(callable)->tp_call is an assertion assert(PyCallable_Check(callable));

Ah, that's a remnant of my investigations that I didn't adjust back to match _PyVectorcall_Function again. I don't feel strongly about this.