Do not use@contextmanger (that is relatively slow) forlocal_type_map, since it appears in multiple hot paths.
Do not show name suggestions for import errors in third party packages (since those errors are ignored anyway). It callsdifflib that can be extremely slow with large modules.

Btw the second will probably not affect self-check, although it did affectmy self-check, since apparentlypytest depends onnumpy. Well, they don't specify it as a package dependency, buthttps://github.com/pytest-dev/pytest/blob/main/src/_pytest/python_api.py#L17-L18

ifTYPE_CHECKING:fromnumpyimportndarray

(and I have numpy installed in all my environments, LOL)

Two more micro-optimizations

024a9d8

ilevkivskyi requested review fromJukkaL,hauntsaninja andsterliakov

August 9, 2025 10:41

Copy link

Contributor

github-actionsbot commentedAug 9, 2025

According tomypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅

Copy link

MemberAuthor

ilevkivskyi commentedAug 9, 2025

Btw just wanted to brag about how stable is my local bench:

master                    3.810s (0.0%)  | stdev 0.008s twop-more-opts            3.756s (-1.4%) | stdev 0.008s

(you read that right, standard deviation is just 8 milliseconds)

sterliakov approved these changes

Aug 9, 2025

View reviewed changes

Copy link

Collaborator

sterliakov left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others.Learn more.

Nice one! Please post this to plugin API announcements - I expect some plugins in wild might use this functionality (yes, undocumented, yes, private, let's just do that and let people know something might break).