> I'm actually curious why djb specified this as the process start time,
> rather than the current time. I see no obvious advantage to using the
> start time, and as the PID and count are encoded, it would not appear to
> have any affect on the possibility of collisions.
Indeed. And one wonders why inode isn't suggested instead of count to
completely avoid collisions. Given that the file is first created in
tmp/ there is ample opportunity to determine the inode prior to
linking to new/. This should work across NFS.
> Personally, I would lean to using the current time instead of start
> time, and would not bother with resetting count; it has no harmful
> effects.
It could be argued that you're duplicating information. The process id
and start time give you information about the creating process, while
mtime tells you when the message was created (modulo subsequent
updates of course). If you use current time you're kinda duplicating
mtime and losing process identity (given that pids recycle). I'm not
sure that this is a big deal but it's worth point out.
Given the vagueness of the hostname component (in the sense of
interpreting it to have meaning beyond uniqueness across multiple
systems), one could trivially add material in this part of the
filename giving something like:
starttime.count.inode.currenttime.host.domain
Regards.
