Speed up print() by looking up file.write once per call #160725

Feature or enhancement

print() currently looks up file.write, creating a new bound method, for every argument, separator, and line ending. Currently print('a', 1, 2.5) does six lookups. The default sep and end also go through PyFile_WriteString(), which converts " " and "\n" from C strings with strlen() and the UTF-8 decoder each time, only to return CPython's cached one-character strings. Looking write up once per call and passing the cached strings directly makes print() 1.20x faster (geometric mean), and write() is still called exactly as before.

Measurements

Time: pyperf, -p 10, release build without PGO, Linux x86_64. n.s. = not significant. Instructions: per print() call, measured with valgrind (cachegrind) and PYTHONHASHSEED=0, with a noise floor of about ±1%.

Behavior change

write is looked up once per call. If an argument's __str__ rebinds file.write, the new method is used from the next print() call on. A file without a write method still raises AttributeError before any argument is converted.

Benchmark

import pyperf

SETUP = """
import io, os, tempfile
devnull = open(os.devnull, 'w')
sio = io.StringIO()
disk = tempfile.TemporaryFile('w')
ten_ints = tuple(range(10))
big = 'x' * 2**20
big_list = list(range(1000))
"""

CASES = {
    "one_str[devnull]": "print('hello world', file=devnull)",
    "one_str[sio]": "print('hello world', file=sio)",
    "three_args[devnull]": "print('a', 1, 2.5, file=devnull)",
    "ten_ints[devnull]": "print(*ten_ints, file=devnull)",
    "empty[devnull]": "print(file=devnull)",
    "str_1mib[devnull]": "print(big, file=devnull)",
    "list_1000[devnull]": "print(big_list, file=devnull)",
    "flush[disk]": "print('hello world', file=disk, flush=True)",
}

runner = pyperf.Runner()
for name, stmt in CASES.items():
    runner.timeit(name, stmt, SETUP)

./python-main bench_print.py -p 10 -o main.json
./python-patched bench_print.py -p 10 -o patched.json
python -m pyperf compare_to main.json patched.json --table

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-160725: Speed up print() by looking up write() once per call #160727