On Wed, Aug 15, 2001 at 03:16:27PM -0400, Paul Jarc allegedly wrote:
> "MarkD" <[EMAIL PROTECTED]> wrote:
> > And one wonders why inode isn't suggested instead of count to
> > completely avoid collisions.
>
> Suppose you tar up a maildir, and then untar it later. The inodes
> will change, so new deliveries might collide with old messages.
Only if your clock goes backwards as well. But count is just as likely
to fail then too, isn't it?
I wasn't trying to suggest that inode become the sole identifier, just
that it's a good "uniquifier" in conjunction with time+pid+FQDN which
works well for short-lived programs as well as long-standing
daemons. It's not a strong suggestion though, more thinking out load.
Regards.
