Pavel Machek wrote:
> > So run with -noleaf, or compile it with -noleaf as the default. Those of
> > us who use AFS are used to that anyway.
>
> Notice: they want that to go into 2.4.X. Recompiling all distros all
> there is *not* option. But returning nlink==1 is option, and I like
> it. (You should not need -noleaf if AFS returns nlink==1 correctly...)
-noleaf slows it down noticably on local disks. I've just checked the
source for findutils-4.1, and nlink==1 (or nlink==0) will work just fine
though the logic doesn't explicitly check for that. (It's fine provided
you have <INT_MAX-1 directories :-).
I think all the filesystems in the Linux kernel tree always return a
reasonable link count: either 2+#subdirs or 0 or 1.
I just tried AFS on a few systems, some Linux and some not. It's idea
of i_nlink for directories is weird. Sometimes it's right, sometimes it's
a lower count that bears no obvious relationship to anything. I haven't
tried ARLA, the free AFS clone.
So AFS is broken :-)
I'm tempted to add an "is it AFS" check to treescan.
enjoy,
-- Jamie
