* David Hildenbrand <david@xxxxxxxxxx> [241209 17:02]: > > > > > > Maybe there is a better way to split up things or rework the code; likely > > > we'd want this code that works on VMAs to have a clean interface with the > > > core vma logic in vma.c, if the current way of handling it is a problem for > > > you. > > > > I think we probably need a compromise for the time being, and as I was > > saying to Jann I don't think it makes sense really to separate the VMA and > > page table bits from these files _except_ when we finally have a shared > > page table interface that we've spoken about before and perhaps we will > > collaborate on in future :) > > In fork.c we split out the page table bits (into mm/memory.c), but left the > VMA bits in there ... :) > > > > > The key point is so we avoid stepping on landmines when we discover > > something was merged in file X which we weren't cc'd on that breaks core > > feature Y we have been working on in the VMA part of the code. > > In my experience, tracking files is not particularly helpful. People will > still not CC you, or do nasty stuff in files you are not tracking. > > What I do is (well, besides screening most of linux-mm), is have a list of > magic names/keywords, and tag the mails. > > This way, when someone uses the magic "COW" keyword, for example, or calls > mapcount functions, I get them put into a separate folder where I can just > briefly sanity check them. Maybe I can explain where this came from a little bit and see what you think. We have found files changed after a release without hitting the mm branch at all. Not just once, but a few times. We have regular push-back and continued conversations for development after a nack on threads with ideas that need rethinking. Now we have a user visible undocumented change that is actively being used that will need to be supported, forever. It's horrible and we need to do something - anything - to try and change what is happening. I'm happy for lei and mailbox rules to help lighten the load, but some of these things need more weight behind them. I also don't think that a new tool means we should abandon the existing infrastructure. Maybe all the tools together will be more effective. My hope is that having names against the files will have more weight when someone decides that they can just take this patch through their own branch, or mailing list, or whatever. That is, an explicit statement to stop doing things behind our backs. We can also, hopefully, get more reviewers Cc'ed on these areas. Would a different topic with you, us, and others make sense instead of under VMA? I would think MEMORY MAPPING might work for a combined area, but I have no strong opinion on what we call it; this one already exists for mmap.c. Thanks, Liam
