Improve performance of inherited mapping assignment slot calls

Feature or enhancement

Proposal:

__setitem__ and __delitem__ share the mp_ass_subscript slot. If a Python subclass of a C mapping overrides only one of them, slot_mp_ass_subscript() is used for both operations.

For example, collections.Counter overrides __delitem__ but inherits dict.__setitem__. Assigning an item looks up the inherited wrapper_descriptor and calls it through the generic tp_call path. Since PyWrapperDescr_Type does not support vectorcall, _PyObject_MakeTpCall() and wrapperdescr_call() each create a temporary argument tuple before reaching dict's C slot.

Reproducer:

from collections import Counter
from time import perf_counter_ns

obj = Counter()

start = perf_counter_ns()
for index in range(20_000_000):
    obj[0] = index

print((perf_counter_ns() - start) / 1_000_000)

On macOS 15.7.1 arm64 with a release build of main (d856402e7176e5a03c5c578f0efc1be968be3257), using two warmups and nine alternating-order samples:

main:      1402.07 ms median (1376.93–1445.12 ms)
prototype:  859.97 ms median (846.24–902.74 ms)

The prototype calls the inherited C slot directly when lookup returns the expected unbound mapping slot wrapper, and otherwise keeps the existing call path. This reduced the benchmark time by 38.66%.

The relevant call-path files are unchanged at b062727097e997bcb900e11503d3248daac903da.

Has this already been discussed elsewhere?

This is a minor feature, which does not need previous discussion elsewhere

Links to previous discussion of this feature:

No response

Linked PRs

• gh-156258

Beginne damit, die im Issue beschriebenen Pfade slot_mp_ass_subscript(), _PyObject_MakeTpCall() und wrapperdescr_call() zu lesen, und prüfe dann den verknüpften PR gh-156258. Fertig ist die Arbeit, wenn der geerbte Pfad für die Zuweisung eines Mappings temporäre Argument-Tupel vermeidet, dabei aber den bestehenden Fallback beibehält, und der bereitgestellte Counter-Benchmark die gemeldete Verbesserung zeigt.