Skip to content

Conversation

@BSnelling
Copy link

Fix for #43847

I first implemented the fix proposed on the issue but as expected this was ambiguous. I'm not sure if my proposal is as general as the initial proposal but it is not ambiguous and results in desired behaviour in a test.

Branwen Snelling and others added 12 commits February 3, 2022 10:57
These stem from julia#15276, i.e.,
https://docs.julialang.org/en/v1/manual/performance-tips/#man-performance-captured
They were identified by scanning all compiled MethodInstances with
hasbox in the newly-released MethodAnalysis 0.4.5.

Core.Box often causes "follow-on" inference problems,
but for these cases there were relatively few, which may be why
these didn't show up earlier during the Great Invalidation Hunt.
Still, there doesn't seem to be any particular reason not to fix them.

This doesn't eliminate all Core.Box cases from Base, but only a
handful remain. The most common remaining case stems from inner
functions calling themselves.

It also prevents a couple of Test invalidations

Co-authored-by: Jameson Nash <vtjnash@gmail.com>
@timholy
Copy link
Member

timholy commented Feb 7, 2022

There seem to be some unrelated commits here. Can you rebase on current master?

@BSnelling
Copy link
Author

I think I must have merged when I should have rebased and now I cannot figure out how to fix that commit history. Since my changes are just a few lines I'm going to replace this with #44061

@BSnelling BSnelling closed this Feb 7, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants