bpo-33012: Add _Py_CAST_FUNC() to cast function ptr

[vstinner]

Use _Py_CAST_FUNC() to cast functions pointers to getter, setter,
destructor, etc.

Change mostly written in gvim using the command:

:%s!(\(inquiry\|destructor\|reprfunc\|getter\|setter
 \|releasebufferproc\|getbufferproc\))
 \([A-Za-z0-9_]\+\)
 !_Py_CAST_FUNC(\1, \2)!g

https://bugs.python.org/issue33012

[serhiy-storchaka]

The problem is not in warnings, but in incorrect function signatures. This PR just silences valid compiler warnings. See #10746.

[vstinner]

The problem is not in warnings, but in incorrect function signatures.

What do you mean by "incorrect function signature"? Example:

static PyObject *
method_get_doc(PyMethodDescrObject *descr, void *closure)
{
    return _PyType_GetDocFromInternalDoc(descr->d_method->ml_name, descr->d_method->ml_doc);
}

static PyGetSetDef method_getset[] = {
    {"__doc__", (getter)method_get_doc},
    ...
};

Here the warning is that PyGetSetDef expects PyObject* as the first argument of the getter, whereas method_getset uses PyMethodDescrObject*. Is it an invalid signature? Would you prefer to write:

static PyObject *
method_get_doc(PyObject *descr_obj, void *closure)
{
    PyMethodDescrObject *descr = (PyMethodDescrObject *)descr_obj;
    return _PyType_GetDocFromInternalDoc(descr->d_method->ml_name, descr->d_method->ml_doc);
}

which is the "correct" signature? The additional "PyMethodDescrObject *descr = (PyMethodDescrObject *)descr_obj;" doesn't seem to provide any value to the compiler nor to the developer who read/maintain the code.

The problem is not in warnings, but in incorrect function signatures. This PR just silences valid compiler warnings. See #10746.

Right, that's a good fix, but I was too lazy to check each function pointer.

[serhiy-storchaka]

I did not get a warning for method_get_doc. Could you please attach the output of the compiler to bpo-33012?

[vstinner]

@serhiy-storchaka: Oh wow, I don't understand how it's possible, but almost all GCC warnings are gone. It seems like I didn't try on your latest compiler warning fix. So this PR is completely useless!