Use non-capturing groups for IPv4 address detection.

[elliotwutingfeng]

Since we are not using the captured groups, we can replace them with non-capturing groups as a micro-op. Benchmarks indicate about 10-20% improvement in parsing speed for positive matches.

Using the re.ASCII flag appears to have slight speed boost for CPython, but makes things slower on PyPy for certain inputs.

Results are mixed for bracketless-IPv6 addresses.

import timeit

stmts = []
stmts.append(
"""
IP_RE.fullmatch("1.1.1.1")
"""
)
stmts.append(
"""
IP_RE.fullmatch("aBcD:ef01:2345:6789:aBcD:ef01:127.0.0.1")
"""
)
stmts.append(
"""
IP_RE.fullmatch("255.255.255.com")
"""
)
stmts.append(
"""
IP_RE.fullmatch("a.b.c.e.co.uk")
"""
)

setups = []
setups.append(
"""
import re
IP_RE = re.compile(
    r"^(([0-9]|[1-9][0-9]|1[0-9]{2}|2[0-4][0-9]|25[0-5])\.)"
    r"{3}([0-9]|[1-9][0-9]|1[0-9]{2}|2[0-4][0-9]|25[0-5])$"
)
"""
)
setups.append(
"""
import re
IP_RE = re.compile(
    r"^(?:(?:[0-9]|[1-9][0-9]|1[0-9]{2}|2[0-4][0-9]|25[0-5])\.)"
    r"{3}(?:[0-9]|[1-9][0-9]|1[0-9]{2}|2[0-4][0-9]|25[0-5])$"
)
"""
)
setups.append(
"""
import re
IP_RE = re.compile(
    r"^(?:(?:[0-9]|[1-9][0-9]|1[0-9]{2}|2[0-4][0-9]|25[0-5])\.)"
    r"{3}(?:[0-9]|[1-9][0-9]|1[0-9]{2}|2[0-4][0-9]|25[0-5])$", re.ASCII
)
"""
)

messages = ["original", "non-capturing groups", "non-capturing groups + re.ASCII"]

for (setup, message) in zip(setups, messages):
    print(message)
    for stmt in stmts:
        timer = timeit.Timer(stmt=stmt, setup=setup)
        print(timer.timeit(number=10_000_000))

CPython 3.11

original
2.1482726680005726
1.1063692410007206
3.2284626149994438
1.1412295229993106
non-capturing groups
1.8773814500000299
1.148869920000834
2.7006006390001858
1.1598876739999469
non-capturing groups + re.ASCII
1.7322393260001263
1.1551464869999108
2.7154349570000704
1.1148897359998955

PyPy 3.9

original
0.42392663299960986
0.21313040499990166
0.48231843799931085
0.21305666600073891
non-capturing groups
0.3258821990002616
0.23411077200034924
0.41092471499996464
0.2272209719994862
non-capturing groups + re.ASCII
0.34160923099989304
0.21923503299967706
0.42936557000030007
0.21578892799971072

[john-kurkowski]

Love it! Thanks for the thorough stats!