<@U051221NF> I'm I remembering correctly that you ...
# development
c
@happy-kitchen-89482 I'm I remembering correctly that you had a local (NG?) branch where you had used the dep batching from https://github.com/pantsbuild/pants/pull/23075 ? I had played around with trying to batch by directory at one point, and https://github.com/pantsbuild/pants/pull/23434/changes has few "mega-batches". I'm trying to figure out how to coordinate the right approach
h
Yes, NG uses this, I'll find a link. I am actually looking into how to use #23075 to speed up
dependents
computation even in the existing backend. It's not trivial because the entire dep inference edifice expects to operate on a single target at a time.
Oh, hah, that is exactly what you're doing
So is this just assuming everything is python and ignoring other dep inference implementations? How would it work with custom dep inference?
w
I thought @worried-painter-31382 also had a PR for something like this?
w
Mine is a deeper change, allowing rules and cache to be implemented in rust and therefore gil free
Would be net same speedup most likely, but a crapton more work
c
So is this just assuming everything is python and ignoring other dep inference implementations? How would it work with custom dep inference?
If 'this' is 23434, that's @gentle-flower-25372 . If `this was my "dir at a time" branch... I have not managed to find it again yet ;-/
h
OK sorry, I got the wrong end of the stick here, I thought this was your PR. @gentle-flower-25372 and I have been chatting about this problem for a while, so I will follow up with him.