preference to disable initial cache of fieldtypes(#991)#992
preference to disable initial cache of fieldtypes(#991)#992dpinol wants to merge 1 commit intotimholy:masterfrom
Conversation
aviatesk
left a comment
aviatesk
left a comment
There was a problem hiding this comment.
Can we document this somewhere?
dpinol
force-pushed
the
precache_fieldtypes-pref
branch
from
5d6190f to
a6b2081
Compare
done, thanks |
|
I'm starting to make steps towards "coming back," but I'm hesitating a bit on this. Once we merge this we need to stick with it until the next breaking release. I think there's some question whether the better thing to do is have an option to disable type revision. Without a pre-populated cache, type revision is so slow that it's often faster to just quit the session. If we just do it this way, then if you change any types it will start on the slow task of populating the cache. |
|
I don't have a strong opinion on whether the preference should be on the type cache or type revision entirely, but I'm definitely +1 on having some option that ends up disables the caching. It's kinda annoying to have Revise hog a thread for 30s every time Julia starts. |
|
+1 for allowing to disable struct vision all together |
This addresses the segfault issue reported in JuliaLang/julia#60721. The future of this feature is still undecided (#992), but this change prioritizes preventing segfaults over performance for now.
This addresses the segfault issue reported in JuliaLang/julia#60721. The future of this feature is still undecided (#992), but this change prioritizes preventing segfaults over performance for now.
|
We had a segfault issue with this part on v1.12 and Apple Silicon environments, as reported in JuliaLang/julia#60721 (due to a problem on the Julia side, just to be clear). For now, wIve added a fix in 3.12.2 to disable this pre cache-population when it's v1.12 and |
|
I think it's still good to have an option to disable struct revision until the performance issues are resolved |
|
Gentle bump, would be great to have an option for this. |
|
I think it's okay to add this option, as far as we put an explicit "experimental" warning on this preference, and make it clear that this configuration might change or even be removed without announcement, depending on our plans for struct revision-related issues. |
|
Should this be an opt-in instead, given that the performance issues can be fairly severe? |
6 participants
Fixes #991 adding a Preference.jl optional to disable the initial fieltypes cache population