Re: [PATCH] Perform minimal stat comparison when some stat fields are not set

[Robin Rosenberg <robin.rosenberg@dewire.com>]

----- Ursprungligt meddelande -----
> Am 12/6/2012 2:09, schrieb Robin Rosenberg:
> >> Robin Rosenberg <robin.rosenberg@dewire.com> writes:
> >>> At least JGit does sets uid, gid, ctime, ino and dev fields to
> >>> zero
> >>> on update. To Git this looks like the stat data does not match
> >>> and
> >>> a full file compare will be forced even it size and mtime match.
> >>> This
> >>> is in practice unnecessary. Sense JGit's presence by checking if
> >>> ino
> >>> and dev is zero.
> 
> Is this meant to better support C git and JGit working on the same
> repository?
> 
> MinGW git sets these two stat fields to zero as well. But we have
> less of
> an interoparability problem between different git implementations in
> practice on Windows, I think.

It is purely for performance in some situations.

> >> Besides, is it sane to do this unconditionally to affect people
> >> who
> >> do not use JGit?
> > 
> > Would a config option like core.minstat be better? The name would
> > imply no dynamic detection.
> 
> A configuration option is the way to go. We already have
> core.trustctime,
> core.symlinks, core.filemode, core.ignoreCygwinFSTricks.
> 
> But your new mode is not "minimal". In some implementations or on
> some
> filesystems, even more bits of stat information could be meaningless
> (think of atime, rdev, nlink, uid, gid). Perhaps core.trustdevandino?

I already excluded uid and gid so the only thing left is mtime and size.
I can't see any reason for anyone to look at atime (somebody read the file,
so what?), ok for rdev and nlink, but we don not look at them my patch
does not avoid looking at them.

> Or
> an enumeration core.ignoreCacheStat=ctime,dev,ino?

That would mean only one configuration option. Good.

-- robin