perf: use version.__replace__ in specifier comparison by notatallshaw · Pull Request #999 · pypa/packaging

notatallshaw · 2025-11-27T08:14:45Z

Takes advantage of being able to strip parts of a Version and get a Version without having to revalidate a version string.

Not had chance to benchmark this yet, but should significantly improve performance of specifier comparison when local, pre, post, or dev parts are involved.

henryiii · 2025-11-30T04:48:47Z

Would this also make sense to add to this PR?

diff --git a/src/packaging/specifiers.py b/src/packaging/specifiers.py
index f027aeb..2ba5541 100644
--- a/src/packaging/specifiers.py
+++ b/src/packaging/specifiers.py
@@ -420,7 +420,7 @@ class Specifier(BaseSpecifier):
         if spec.endswith(".*"):
             # In the case of prefix matching we want to ignore local segment.
             normalized_prospective = canonicalize_version(
-                prospective.public, strip_trailing_zero=False
+                _public_version(prospective), strip_trailing_zero=False
             )
             # Get the normalized version string ignoring the trailing .*
             normalized_spec = canonicalize_version(spec[:-2], strip_trailing_zero=False)

Also _canonical_spec access self._spec[1] directly and converts it to a version inside canonicalize_version, maybe that should use _get_spec_version instead? Also maybe _spec[1] could be cached separately, as _get_spec_version will change the cache to a different value if a different version is passed into it? Unless it's always called with the same value. I also see it used directly in a few other places, like .prerelease.

Full patch:

diff --git a/src/packaging/specifiers.py b/src/packaging/specifiers.py
index f027aeb..18edd34 100644
--- a/src/packaging/specifiers.py
+++ b/src/packaging/specifiers.py
@@ -285,11 +285,13 @@ class Specifier(BaseSpecifier):
             # The == specifier can include a trailing .*, if it does we
             # want to remove before parsing.
             if operator == "==" and version.endswith(".*"):
-                version = version[:-2]
+                version_object = Version(version[:-2])
+            else:
+                version_object = self._get_spec_version(version)

             # Parse the version, and if it is a pre-release than this
             # specifier allows pre-releases.
-            if Version(version).is_prerelease:
+            if version_object.is_prerelease:
                 return True

         return False
@@ -351,7 +353,7 @@ class Specifier(BaseSpecifier):
             return operator, version

         canonical_version = canonicalize_version(
-            version,
+            version if version.endswith(".*") else self._get_spec_version(version),
             strip_trailing_zero=(operator != "~="),
         )

@@ -420,7 +422,7 @@ class Specifier(BaseSpecifier):
         if spec.endswith(".*"):
             # In the case of prefix matching we want to ignore local segment.
             normalized_prospective = canonicalize_version(
-                prospective.public, strip_trailing_zero=False
+                _public_version(prospective), strip_trailing_zero=False
             )
             # Get the normalized version string ignoring the trailing .*
             normalized_spec = canonicalize_version(spec[:-2], strip_trailing_zero=False)

notatallshaw · 2025-11-30T07:01:55Z

Would this also make sense to add to this PR?

Yes, turns out there were two places I was missing public and base version calls where I could have been using them.

Also _canonical_spec access self._spec[1] directly and converts it to a version inside canonicalize_version, maybe that should use _get_spec_version instead? Also maybe _spec[1] could be cached separately, as _get_spec_version will change the cache to a different value if a different version is passed into it? Unless it's always called with the same value. I also see it used directly in a few other places, like .prerelease.

I would need to carefully go over this, the one-element cache is always called with the same value at the moment (assuming you don't mutate the specifier, in which case the cache invalidates itself when you call _get_spec_version), I need to validate that it would always be the same version under _canonical_spec.

notatallshaw mentioned this pull request Nov 27, 2025

feat: support __replace__ for Version #1003

Merged

notatallshaw added 2 commits November 29, 2025 20:55

Use Version.replace in specifiers

3fb0165

Switch to __replace__

eddb254

notatallshaw force-pushed the use-Version.replace-in-specifiers branch from 001fbe4 to eddb254 Compare November 30, 2025 01:59

notatallshaw changed the title ~~perf: use version.replace in specifiers~~ perf: use version.__replace__ in specifier comparison Nov 30, 2025

notatallshaw added 2 commits November 30, 2025 01:53

Additional place _public_version can be used

213028f

Additional place base version can be used

975bd14

notatallshaw marked this pull request as ready for review November 30, 2025 07:02

henryiii approved these changes Nov 30, 2025

View reviewed changes

Merge branch 'main' into use-Version.replace-in-specifiers

9f1626e

notatallshaw enabled auto-merge (squash) November 30, 2025 19:55

notatallshaw merged commit 5bf7018 into pypa:main Nov 30, 2025
37 checks passed

notatallshaw mentioned this pull request Nov 30, 2025

perf: use _get_spec_version in more places in Specifier #1005

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

perf: use version.replace in specifier comparison#999

perf: use version.replace in specifier comparison#999
notatallshaw merged 5 commits intopypa:mainfrom
notatallshaw:use-Version.replace-in-specifiers

notatallshaw commented Nov 27, 2025 •

edited

Loading

Uh oh!

henryiii commented Nov 30, 2025 •

edited

Loading

Uh oh!

notatallshaw commented Nov 30, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

notatallshaw commented Nov 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

henryiii commented Nov 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

notatallshaw commented Nov 30, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

notatallshaw commented Nov 27, 2025 •

edited

Loading

henryiii commented Nov 30, 2025 •

edited

Loading