Bug report

Bug description:

The comment here

Indicates that it needs to be checked every time. However, once it removes vectorcall in partial_vectorcall_fallback there is no mechanism to set it back.

def func(a, b, c=0):
    return a - b - c
p2 = partial(func, 1, c=1)
In [13]: print(p2(2))    # -1
vectorcall
call
-2
In [14]:
In [14]: print(p2(2))    # -1
call
-2
In [15]: del p2.keywords['c']
In [16]: print(p2(2))    # -1
call
-1

I see 3 possibilities:

1. Set additional flag in partial_setvectorcall(pto->hasvcall) storing state whether vectorcall is supported. Then implement fallback in partial_call in the same manner as it looks now in partial_vectorcall
2. Re-setting vectorcall after _PyObject_MakeTpCall in partial_vectorcall_fallback. So that partial_vectorcall will stay as a primary entry in all cases. This adds 4ns overhead compared to 1.
3. Leave it as it is, but remove confusion that this needs to be checked more than 1 time.

CPython versions tested on:

3.11

Operating systems tested on:

macOS

Linked PRs

• gh-119109: functools.partial.fallforward fix #119125
• gh-119109: functool.partial vectorcall supports pto->kw & fallback to tp_call removed #120783
• gh-119109: improve functools.partial vectorcall with keywords #124584