Bytes performance regression in python3.3 vs python3.2

[Lothiraldan]

BPO	13622
Nosy	@vstinner, @Lothiraldan
Files	stringbench_log_cpython3.2: Stringbenchmark log for cpython3.2 stringbench_log_cpython3.3: String benchmark log for cpython3.3 compare.py: Script used to compute diff between two runs

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2011-12-17.18:42:50.566>
created_at = <Date 2011-12-17.17:53:49.651>
labels = ['invalid', 'performance']
title = 'Bytes performance regression in python3.3 vs python3.2'
updated_at = <Date 2011-12-17.18:42:50.565>
user = 'https://github.com/Lothiraldan'

bugs.python.org fields:

activity = <Date 2011-12-17.18:42:50.565>
actor = 'vstinner'
assignee = 'collinwinter'
closed = True
closed_date = <Date 2011-12-17.18:42:50.566>
closer = 'vstinner'
components = ['Benchmarks']
creation = <Date 2011-12-17.17:53:49.651>
creator = 'Boris.FELD'
dependencies = []
files = ['23995', '23996', '23997']
hgrepos = []
issue_num = 13622
keywords = []
message_count = 4.0
messages = ['149683', '149685', '149687', '149692']
nosy_count = 3.0
nosy_names = ['collinwinter', 'vstinner', 'Boris.FELD']
pr_nums = []
priority = 'normal'
resolution = 'not a bug'
stage = None
status = 'closed'
superseder = None
type = 'performance'
url = 'https://bugs.python.org/issue13622'
versions = ['Python 3.2', 'Python 3.3']

[Lothiraldan]

Hello everyone, I juste tried to launch the stringbench on python3.2 and python3.3 dev versions and some bytes tests run slower in python3.3 than in python3.2.

I cc the two raw output of both runs. I also extracted most interesting data (all the tests with more than 20% of performance regression):

• (b"A"*1000).find(b"B") (*1000): -30.379747%
• b"Hello\t \t".rstrip() (*1000): -33.333333%
• b"this\nis\na\ntest\n".rsplit(b"\n") (*1000): -23.437500%
• b"\nHello!\n".strip() (*1000): -33.333333%
• dna.split(b"ACTAT") (*10): -21.066667%
• b"Andrew".endswith(b"w") (*1000): -23.529412%
• b"...text.with.2000.lines...replace(b"\n", b" ") (*10): -37.668161%
• b"\t \tHello".rstrip() (*1000): -33.333333%
• (b"A"*1000).rpartition(b"A") (*1000): -21.212121%
• (b"Here are some words. "*2).split() (*1000): -22.105263%
• b"this\nis\na\ntest\n".split(b"\n") (*1000): -23.437500%
• b"Hello!\n".rstrip() (*1000): -35.714286%
• b"B" in b"A"*1000 (*1000): -32.089552%
• b"Hello!\n".strip() (*1000): -35.714286%
• b"\nHello!".strip() (*1000): -28.571429%
• b"Andrew".startswith(b"A") (*1000): -20.588235%
• b"\nHello!".rstrip() (*1000): -35.714286%
• b"Andrew".endswith(b"Andrew") (*1000): -22.857143%
• b"Andrew".endswith(b"Anders") (*1000): -23.529412%
• b"Andrew".startswith(b"Anders") (*1000): -23.529412%
• b"this--is--a--test--of--the--emergency--broadcast--system".split(b"--") (*1000): -22.429907%
• b"Andrew"+b"Dalke" (*1000): -23.076923%

Hope it help

[vstinner]

Sorted and grouped results. "replace", "find" and "concat" should be easy to fix, "strip" and "split" depend on "find" performance.

replace:

• b"...text.with.2000.lines...replace(b"\n", b" ") (*10): -37.668161%

find:

• (b"A"*1000).find(b"B") (*1000): -30.379747%

• b"Andrew".startswith(b"A") (*1000): -20.588235%

• b"Andrew".startswith(b"Anders") (*1000): -23.529412%

• b"Andrew".endswith(b"w") (*1000): -23.529412%

• b"Andrew".endswith(b"Andrew") (*1000): -22.857143%

• b"Andrew".endswith(b"Anders") (*1000): -23.529412%

• b"B" in b"A"*1000 (*1000): -32.089552%

concat:

• b"Andrew"+b"Dalke" (*1000): -23.076923%

strip:

• b"\nHello!\n".strip() (*1000): -33.333333%

• b"Hello!\n".strip() (*1000): -35.714286%

• b"\nHello!".strip() (*1000): -28.571429%

• b"Hello\t \t".rstrip() (*1000): -33.333333%

• b"\t \tHello".rstrip() (*1000): -33.333333%

• b"Hello!\n".rstrip() (*1000): -35.714286%

• b"\nHello!".rstrip() (*1000): -35.714286%

split:

• dna.split(b"ACTAT") (*10): -21.066667%

• (b"Here are some words. "*2).split() (*1000): -22.105263%

• b"this\nis\na\ntest\n".split(b"\n") (*1000): -23.437500%

• b"this--is--a--test--of--the--emergency--broadcast--system".split(b"--") (*1000): -22.429907%

• b"this\nis\na\ntest\n".rsplit(b"\n") (*1000): -23.437500%

• (b"A"*1000).rpartition(b"A") (*1000): -21.212121%

[Lothiraldan]

Forgot to describe my environment:
Mac OS X 10.6.8
GCC i686-apple-darwin10-gcc-4.2.1 (GCC) 4.2.1 (Apple Inc. build 5666) (dot 3)
CPython3.3 revision ea421c534305
CPython3.2 revision 0b86da9d6964

[vstinner]

Boris.FELD told me that there was a bug in compare.py: all numbers are related to Unicode (see bpo-13621), not bytes.