Bug report

Bug description:

The following Python function

def foo(*args): pass

Has vectorcall enabled (PyVectorcall_Function() returns a non-NULL function), but because it collects args, Python needs to convert the arguments into a tuple, and so vectorcall won't be any faster than allocating a tuple directly.

This is not a problem if we know the number of arguments beforehand, but if we allocate the array for vectorcall, we will allocate twice needlessly.

Avoiding allocating for vectorcall also does not always provide the best performance, because vectorcall can have an edge with bound methods with the flag PY_VECTORCALL_ARGUMENTS_OFFSET.

Per the docs:

A class should not implement vectorcall if that would be slower than tp_call. For example, if the callee needs to convert the arguments to an args tuple and kwargs dict anyway, then there is no point in implementing vectorcall.

As such, I believe this function (and similar functions) should not implement vectorcall.

Originally reported on Stack Overflow.

CPython versions tested on:

3.12

Operating systems tested on:

Windows