Reduce number of modules imported by runpy

[vstinner]

BPO	41006
Nosy	@vstinner, @shihai1991
PRs	#20938 #20939 #20940 #20953 #20954 #20955
Files	mod.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 2020-06-17.21:36:12.127>
created_at = <Date 2020-06-17.13:58:21.130>
labels = ['library', '3.10']
title = 'Reduce number of modules imported by runpy'
updated_at = <Date 2020-06-17.23:20:57.312>
user = 'https://github.com/vstinner'

bugs.python.org fields:

activity = <Date 2020-06-17.23:20:57.312>
actor = 'vstinner'
assignee = 'none'
closed = True
closed_date = <Date 2020-06-17.21:36:12.127>
closer = 'vstinner'
components = ['Library (Lib)']
creation = <Date 2020-06-17.13:58:21.130>
creator = 'vstinner'
dependencies = []
files = ['49240']
hgrepos = []
issue_num = 41006
keywords = ['patch']
message_count = 12.0
messages = ['371741', '371742', '371746', '371750', '371761', '371765', '371766', '371774', '371776', '371777', '371781', '371782']
nosy_count = 2.0
nosy_names = ['vstinner', 'shihai1991']
pr_nums = ['20938', '20939', '20940', '20953', '20954', '20955']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue41006'
versions = ['Python 3.10']

[vstinner]

Currently, the runpy module imports many modules. runpy is used by "python3 -m module". I propose to attempt to reduce the number of imports to reduce Python startup time.

With my local changes, I reduce Python startup time from 24 ms to 18 ms:

Mean +- std dev: [ref] 24.3 ms +- 0.2 ms -> [patch] 18.0 ms +- 0.3 ms: 1.35x faster (-26%)

Timing measured by:

./python -m venv env
python -m pyperf command -v -o patch.json -- env/bin/python -m empty

Currently, runpy imports +21 modules:

• ./python mod.py: Total 33
• ./python -m mod: Total 54 (+21)

Example with attached mod.py:

$ ./python -m mod
['__main__',
 '_abc',
 '_codecs',
 '_collections',
 '_collections_abc',
 '_frozen_importlib',
 '_frozen_importlib_external',
 '_functools',
 '_heapq',
 '_imp',
 '_io',
 '_locale',
 '_operator',
 '_signal',
 '_sitebuiltins',
 '_sre',
 '_stat',
 '_thread',
 '_warnings',
 '_weakref',
 '_weakrefset',
 'abc',
 'builtins',
 'codecs',
 'collections',
 'collections.abc',
 'contextlib',
 'copyreg',
 'encodings',
 'encodings.aliases',
 'encodings.ascii',
 'encodings.latin_1',
 'encodings.utf_8',
 'enum',
 'functools',
 'genericpath',
 'heapq',
 'importlib',
 'importlib._bootstrap',
 'importlib._bootstrap_external',
 'importlib.abc',
 'importlib.machinery',
 'importlib.util',
 'io',
 'itertools',
 'keyword',
 'marshal',
 'operator',
 'os',
 'os.path',
 'pkgutil',
 'posix',
 'posixpath',
 're',
 'reprlib',
 'runpy',
 'site',
 'sre_compile',
 'sre_constants',
 'sre_parse',
 'stat',
 'sys',
 'time',
 'types',
 'typing',
 'typing.io',
 'typing.re',
 'warnings',
 'weakref',
 'zipimport']
Total 70

[vstinner]

My local changes removed the following imports from runpy:

• '_heapq',
• '_locale',
• '_sre',
• 'collections.abc',
• 'copyreg',
• 'enum',
• 'heapq',
• 'importlib.abc',
• 'itertools',
• 're',
• 'sre_compile',
• 'sre_constants',
• 'sre_parse',
• 'typing',
• 'typing.io',
• 'typing.re',

[vstinner]

Currently, runpy imports +21 modules:

• ./python mod.py: Total 33
• ./python -m mod: Total 54 (+21)

Oops sorry, that's with my local changes!

Currently, runpy imports not less than 37 modules:

• ./python mod.py: Total 33
• ./python -m mod: Total 70 (+37)

[vstinner]

See also bpo-40275: "test.support has way too many imports".

[vstinner]

I created 3 PRs. I have a few more local branches to avoid types and itertools imports. I may create PRs for these ones as well.

[vstinner]

New changeset 7824cc0 by Victor Stinner in branch 'master':
bpo-41006: collections imports lazily heap (GH-20940)
7824cc0

[vstinner]

New changeset 98ce7b1 by Victor Stinner in branch 'master':
bpo-41006: pkgutil imports lazily re (GH-20939)
98ce7b1

[vstinner]

New changeset 9e09849 by Victor Stinner in branch 'master':
bpo-41006: importlib.util no longer imports typing (GH-20938)
9e09849

[vstinner]

I close the issue. Making more imports lazy doesn't bring much benefit.

--

With the 3 changes, runpy now imports 55 modules, instead of 70.

The startup time is 1.3x faster, 18 ms instead of 24 ms:

Mean +- std dev: [before] 23.7 ms +- 0.4 ms -> [after] 17.8 ms +- 0.6 ms: 1.33x faster (-25%)

--

Avoiding itertools and types doesn't bring much benefit:

Mean +- std dev: [after] 17.8 ms +- 0.6 ms -> [WIP] 17.2 ms +- 0.4 ms: 1.03x faster (-3%)

[vstinner]

New changeset 4c18fc8 by Victor Stinner in branch 'master':
bpo-41006: Document the runpy optimization (GH-20953)
4c18fc8

[vstinner]

New changeset 1bf7959 by Victor Stinner in branch 'master':
bpo-41006: Remove init_sys_streams() hack (GH-20954)
1bf7959

[vstinner]

New changeset 2c2a4f3 by Victor Stinner in branch 'master':
bpo-41006: What's New: less => fewer modules (GH-20955)
2c2a4f3