Conversation
notatallshaw
force-pushed
the
add-fast-path-for-parsing-simple-versions
branch
from
c9db103 to
1f64976
Compare
Member
Author
|
@henryiii I think you mentioned you have some script that parses every version from PyPI? I expect this branch to be about ~2x faster, will be interesting to see real world trade off of simple vs. non-simple versions. |
notatallshaw
force-pushed
the
add-fast-path-for-parsing-simple-versions
branch
4 times, most recently
from
b507aaf to
cdeda42
Compare
Member
Author
|
I realized for the fast path simple version we can drop the regex and just directly do: tuple(map(int, version.strip().split(".")))My quick benchmarking on Python 3.14 shows doing this and catching the exception if it fails speeds up simple version parsing by 4.2x [faster], and slows down non-simple parsing 0.73x [slower], compared to main. So I came up with a fairly fast check using a frozenset that limited the overhead cost for non-simple versions, with this simple version parsing is sped up by 3.15x [faster], but the slow down in non-simple parsing is only 0.95x [slower], compared to main! |
dnicolodi
reviewed
Feb 10, 2026
Contributor
|
FYI, I tried pretty much exactly this (probably still in a stash), and did see it was faster, but decided to work on over all performance of the regex and didn't come back to it, largely due to worries about the non-fast case getting slower. But it looks like you have that impact pretty small (for a nice fast-path improvement). I can check on the full version list soon(ish). (Edit: "this" being the 4.2x, 0.73x one) |
Contributor
$ uv run tasks/benchmark_versions.py
Counter({'release': 7086247, 'pre': 532193, 'dev': 451268, 'post': 97508, 'invalid': 5422, 'epoch': 1155, 'local': 6})
Loaded 8,168,377 versions
Time: 9.9128 seconds
Per version: 0.242712358 µs
$ gh co 1082
$ uv run tasks/benchmark_versions.py
Counter({'release': 7086247, 'pre': 532193, 'dev': 451268, 'post': 97508, 'invalid': 5422, 'epoch': 1155, 'local': 6})
Loaded 8,168,377 versions
Time: 4.3251 seconds
Per version: 0.105899591 µs
$ git stash apply
$ uv run tasks/benchmark_versions.py
Counter({'release': 7086247, 'pre': 532193, 'dev': 451268, 'post': 97508, 'invalid': 5422, 'epoch': 1155, 'local': 6})
Loaded 8,168,377 versions
Time: 4.3296 seconds
Per version: 0.106007648 µsFinal version is with the diff above, using |
Contributor
|
>>> a = '٠١٢.٣٤٥.٦٧٨٩'
>>> a.replace(".", "").isdecimal()
True
>>> tuple(map(int, a.split(".")))
(12, 345, 6789) |
Member
Author
|
@henryiii is that testing distinct versions or all versions for every project regardless of duplication? Either way I think this approach catches a nice balance, all your work on speeding up the general case isn't too badly impacted because which ever of these tests we choose then overhead is low enough. I want to benchmark the overhead across a number of Python versions to make sure there's not an unexpected regression. |
Contributor
|
It's every version from PyPI, not deduplicated. |
Member
Author
|
I think this is ready, I tried the performance impact Python versions 3.8 to 3.15. I found there was variation, but generally this was a significant speed up, and that Python is trending faster on both main and this PR. |
dnicolodi
reviewed
Feb 11, 2026
Contributor
uv run asv continuous main pull/1082/head --sort default
SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY. |
Contributor
|
FYI, I tried always stripping, and dropping the additional whitespace from the regex: diff --git a/src/packaging/version.py b/src/packaging/version.py
index 95c8e8d..f159323 100644
--- a/src/packaging/version.py
+++ b/src/packaging/version.py
@@ -343,7 +343,7 @@ class Version(_BaseVersion):
__slots__ = ("_dev", "_epoch", "_key_cache", "_local", "_post", "_pre", "_release")
__match_args__ = ("_str",)
- _regex = re.compile(r"\s*" + VERSION_PATTERN + r"\s*", re.VERBOSE | re.IGNORECASE)
+ _regex = re.compile(VERSION_PATTERN, re.VERBOSE | re.IGNORECASE)
_epoch: int
_release: tuple[int, ...]
@@ -364,6 +364,7 @@ class Version(_BaseVersion):
If the ``version`` does not conform to PEP 440 in any way then this
exception will be raised.
"""
+ version = version.strip()
if _SIMPLE_VERSION_INDICATORS.issuperset(version):
try:
self._release = tuple(map(int, version.split(".")))$ uv run asv continuous add-fast-path-for-parsing-simple-versions^ add-fast-path-for-parsing-simple-versions --sort default --no-only-changed
...
BENCHMARKS NOT SIGNIFICANTLY CHANGED.Looking at the results, it seems to me it is a hair (1-2%) slower on average. By the way, I had to do that with #1059, the separate repo would have requirement me to push it somewhere with an open PR. Table
|
henryiii
reviewed
Feb 11, 2026
henryiii
approved these changes
Feb 11, 2026
Co-authored-by: Henry Schreiner <HenrySchreinerIII@gmail.com>
henryiii
force-pushed
the
add-fast-path-for-parsing-simple-versions
branch
from
984e2de to
d6c0a36
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This is a performance trade off, it fast paths versions only containing digits and dots (e.g. 1.2.3), at a slight cost to versions which are more complex (include any dev/pre/post/local components).
Testing on Python 3.14 on my machine and using these versions:
Simple: 1.0.0, 2.1.3, 0.0.1, 1.2.3, 2021.1.1, 1.2.3.4.5.6.7.8, 3.8.0, 1.0, 2.0, 10.5.3
Non-simple: 1.0.0a1, 2.0.0b2, 3.0.0rc1, 1.0.0.post1, 1.0.0.dev1, 1.0.0a1.dev1
I see simple versions go from an median of 1.16 µs to 0.49 µs (2.37x [faster]) and non-simple versions go from a median of 1.16 µs to 1.38 µs (0.84x [slower]).
Probably could do with a wider range of performance testing before accepting..