Optimize textio write buffering

[methane]

BPO	36748
Nosy	@methane
PRs	bpo-36748: optimize TextIOWrapper.write() for ASCII string #13002 bpo-36927: Improve the docstring and Doc of traceback. #13359
Files	bm_textio.py

^{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 2019-05-16.06:20:43.711>
created_at = <Date 2019-04-29.06:12:38.657>
labels = ['3.8', 'expert-IO', 'performance']
title = 'Optimize textio write buffering'
updated_at = <Date 2019-05-16.19:34:02.847>
user = 'https://github.com/methane'

bugs.python.org fields:

activity = <Date 2019-05-16.19:34:02.847>
actor = 'mangrisano'
assignee = 'none'
closed = True
closed_date = <Date 2019-05-16.06:20:43.711>
closer = 'methane'
components = ['IO']
creation = <Date 2019-04-29.06:12:38.657>
creator = 'methane'
dependencies = []
files = ['48289']
hgrepos = []
issue_num = 36748
keywords = ['patch']
message_count = 2.0
messages = ['341044', '342627']
nosy_count = 1.0
nosy_names = ['methane']
pr_nums = ['13002', '13359']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = 'performance'
url = 'https://bugs.python.org/issue36748'
versions = ['Python 3.8']

[methane]

textio uses list of textio for internal buffering. There are two inefficiency:

• When textio is line buffered and all written strings are line (it's very common), list object is allocated and freed.
• We convert texts into bytes, and call b''.join(list_of_bytes). But when texts are ASCII and codecs are ASCII-compat, we can skip temporary bytes objects.

Attached patch is benchmark for buffered and line buffered write.

Faster (6):

• write_ascii_32k: 101 ns +- 1 ns -> 73.1 ns +- 0.4 ns: 1.39x faster (-28%)
• write_ascii_8k: 102 ns +- 1 ns -> 73.4 ns +- 0.4 ns: 1.38x faster (-28%)
• write_ascii_linebuffered: 815 ns +- 12 ns -> 731 ns +- 3 ns: 1.12x faster (-10%)
• write_unicode_linebuffered: 840 ns +- 11 ns -> 789 ns +- 15 ns: 1.06x faster (-6%)
• write_unicode_8k: 124 ns +- 1 ns -> 122 ns +- 1 ns: 1.01x faster (-1%)
• write_unicode_32k: 124 ns +- 1 ns -> 122 ns +- 1 ns: 1.01x faster (-1%)

[methane]

New changeset bfba8c3 by Inada Naoki in branch 'master':
bpo-36748: optimize TextIOWrapper.write() for ASCII string (GH-13002)
bfba8c3