Enable lto=fat #20863
Merged
Enable lto=fat #20863
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
CodSpeed Performance ReportMerging #20863 will degrade performances by 5.28%Comparing Summary
Benchmarks breakdown
|
Contributor
|
MichaReiser
force-pushed
the
micha/fat-lto
branch
2 times, most recently
from
b264e60 to
c2f89bb
Compare
MichaReiser
force-pushed
the
micha/fat-lto
branch
from
c2f89bb to
e4a4964
Compare
Member
Author
|
I'm not sure why that one instrumented benchmark regresses. But a 10% improvement on our walltime benchmarks is very convincing |
Co-authored-by: Brent Westbrook <36778786+ntBre@users.noreply.github.com>
This was referenced Oct 15, 2025
dcreager
added a commit
that referenced
this pull request
Oct 15, 2025
…rable * origin/main: [ty] Add (unused) `inferable` parameter to type property methods (#20865) Run macos tests on macos (#20889) Remove `release` CI job (#20887) [ty] CI: Faster ecosystem analysis (#20886) Remove `strip` from release profile (#20885) [ty] Sync vendored typeshed stubs (#20876) [ty] Add some completion ranking improvements (#20807) Improved error recovery for unclosed strings (including f- and t-strings) (#20848) Enable lto=fat (#20863) [`pyupgrade`] Extend `UP019` to detect `typing_extensions.Text` (`UP019`) (#20825) [`flake8-bugbear`] Omit annotation in preview fix for `B006` (#20877) fix(docs): Fix typo in `RUF015` description (#20873) [ty] Improve and extend tests for instance attributes redeclared in subclasses (#20866) [ty] Ignore slow seeds as a temporary measure (#20870) Remove parentheses around multiple exception types on Python 3.14+ (#20768) Update Black tests (#20794)
dcreager
added a commit
that referenced
this pull request
Oct 15, 2025
…nt-sets * dcreager/non-non-inferable: (174 commits) [ty] Add (unused) `inferable` parameter to type property methods (#20865) Run macos tests on macos (#20889) Remove `release` CI job (#20887) [ty] CI: Faster ecosystem analysis (#20886) Remove `strip` from release profile (#20885) [ty] Sync vendored typeshed stubs (#20876) [ty] Add some completion ranking improvements (#20807) Improved error recovery for unclosed strings (including f- and t-strings) (#20848) Enable lto=fat (#20863) [`pyupgrade`] Extend `UP019` to detect `typing_extensions.Text` (`UP019`) (#20825) [`flake8-bugbear`] Omit annotation in preview fix for `B006` (#20877) fix(docs): Fix typo in `RUF015` description (#20873) [ty] Improve and extend tests for instance attributes redeclared in subclasses (#20866) [ty] Ignore slow seeds as a temporary measure (#20870) use existing method Remove parentheses around multiple exception types on Python 3.14+ (#20768) Update Black tests (#20794) just the api parts [ty] Fix further issues in `super()` inference logic (#20843) [ty] Document when a rule was added (#20859) ...
Member
|
Yeah I think I've slowly become okay with this sort of change. I even recently did the same for ripgrep. I still don't like that we have different compilation settings for profiling versus release, but I haven't been bitten too hard by it yet. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Summary
This PR switches from thin to fat LTO optimization.
The main motivation is that using lto=fat fixes the binary size increase that we've seen after switching to inventory for ingredient registration. The regression is mainly due to symbols not being removed when using lto=thin, but they're successfully removed when using lto=fat (from 19MB to 17MB).
Using lto=fat also results in a significant performance improvement, which I think is worth it on its own.
Unfortunately, using lto=flat does have the downside that release builds take significantly longer. One of the main motivations for using
lto=thinwhen we switched fromfattothinin #9031 was to improve performance. To mitigate this, I changed the--profilingprofile to only uselto=thin, similar to what we do in uv (where we disable lto entirely). This PR is also likely to make the mypy primer and benchmark jobs slower (or when running mypy primer locally), because of a significant increase in compile time.This setup now mirrors uv's (with the exception that
profilinguselto=fat).I do feel a bit bad about making this change while @BurntSushi is out, because I know he feels the most vocal about this.
Clean build timing:
Fixes #20845
Relevant discussions: