Re: [PATCH RFC 2/9] timekeeping: new interfaces for multigrain timestamp handing

[Linus Torvalds]

On Fri, 20 Oct 2023 at 13:06, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> So on reboot, the inode would count as "changed" as far any remote
> user is concerned. [..]

Obviously, not just reboot would do that. Any kind of "it's no longer
cached on the server and gets read back from disk" would do the same
thing.

Again, that may not work for the intended purpose, but if the use-case
is a "same version number means no changes", it might be acceptable?
Even if you then could get spurious version changes when the file
hasn't been accessed in a long time?

Maybe all this together with with some ctime filtering ("old ctime
clealy means that the version number is irrelevant"). After all, the
whole point of fine-grained timestamps was to distinguish *frequent*
changes. An in-memory counter certainly does that even without any
on-disk representation..

               Linus