Feature or enhancement

list_vectorcall() currently allocates with PyType_GenericAlloc() so list() and list(iterable) always allocate new. [] and PyList_New() reuse an object when possible from the existing list freelist instead. Can reuse the freelist in the tp_vectorcall slot which only applies for the exact type.

I found this while profiling builtin constructor calls with callgrind (instructions per function). For list(t), about a third of the cost was PyType_GenericAlloc() plus the matching malloc/free, which [] never pays.

Microbenchmarks (pyperf, PGO+LTO, Linux x86-64, interleaved A/B, main vs patch):

Untouched controls ([], a list subclass, x + 1, len(l), l[0], tuple(l)) stay within the A/A noise (≤ 1.07x).

import pyperf

runner = pyperf.Runner()
for stmt in ["list()", "list(t)", "list(l)", "list(r)", "list(d)", "L(t)", "[]",
             "len(l)", "l[0]", "x + 1", "tuple(l)"]:
    runner.timeit(stmt, stmt,
                  "t = (1, 2, 3); l = [1, 2, 3]; r = range(3); d = dict.fromkeys(t); x = 12345\n"
                  "class L(list): pass")

./python bm_list.py --affinity=10-13 -o main.json      # on main
./python bm_list.py --affinity=10-13 -o patch.json     # on the patch
python -m pyperf compare_to main.json patch.json --table -G

• The list freelist was added in 2004 by Apply freelist technique to empty lists #40215..
• list_vectorcall() was added in bpo-37207: Use vertorcall for list() #18928 and kept using the existing generic alloc.

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-160858: Use the list freelist in list() #160859