GH-132380: Add optimization for non-interned type lookup. #132652
Closed
Conversation
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
colesbury
reviewed
Apr 21, 2025
Comment on lines
+5743
to
+5745
| /* Keep a strong reference to mro because type->tp_mro can be replaced | ||
| during dict lookup, e.g. when comparing to non-string keys. */ | ||
| mro = _PyType_GetMRO(type); |
There was a problem hiding this comment.
Won't this lead to reference count contention?
I don't think we need to skip the cache in this case. The names are exact unicode objects, they're just not interned. If the string isn't interned can we use a different loop in the lookup code that uses something like PyUnicode_Compare?
There was a problem hiding this comment.
I implemented something that uses unicode compare in gh-133669. More notes there.
|
Closing since this is an approach we don't want to take. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.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.
Add an optimized implementation of type lookup for names that are non-interned strings. This is only enabled for free-threaded builds since that kind of lookup can cause contention for the global type lock.
I benchmarked this with LibCST running on the numpy source code. Summary of results:
_PyType_LookupRef#132380The backport to 3.13 of this is non-trivial and so I made a separate PR for it (GH-132651). It is basically the same logic though. This approach to avoid the contention might be preferred to GH-132381 since I think it should be more obviously safe.
Benchmark results from pyperformance