"Jason R. Mastaler" <[EMAIL PROTECTED]> writes:

> Tim Legant <[EMAIL PROTECTED]> writes:
> 
> > Once I have the module I mentioned above in place, I will take a
> > look at the CGI code to see if I can figure out how to integrate the
> > virtual domain stuff better with that app.
> 
> Ok, but I assume this is longer term (i.e, not before the next
> release)?

Yes.  Don't hold up the release waiting for this!

> > I saw that.  Looks like the fallback algorithm, after you don't use
> > all the parts that you should use but aren't implemented on various
> > systems, is pretty similar to what maildir writers have been using
> > all along.  :)
> 
> Not really. The unix_* calls he mentions are fictitious (things that
> djb thinks OSs _should_ have),

Right.  Basically, services using his naming conventions.  Some
systems have acceptable substitutes (/dev/urandom, for instance).  But
either you have to test, at compile time or run time, if each one is
available or, you can just ignore the ones that aren't universal.

I think that leaves you, off the top of my head, with inodes and
device number plus the the three things maildirs have used all along.

> and inode/device numbers are available on any POSIX-compliant NFS
> implementation.

And I think that's a reasonable limitation to impose on ourselves (a
POSIX-compliant NFS).  If your OS doesn't have a modern NFS
implementation, it's time for a new OS!

> I think I'll mimic the new Postfix algorithm, which looks like:
> 
>   1. Create    tmp/time.P<pid>.hostname
>   2. Rename to new/time.V<device>I<inode>.hostname
> 
> IEEE Std 1003.1-2001 claims that st_ino and st_dev together uniquely
> identify a file within a filesystem.

That looks good to me.  Relying on PID uniqueness can apparently fail
on some systems that reuse PIDS immediately.

> > A final note: tmda-ofmipd's -A/--authprog flag is thoroughly broken.
> 
> [...]
> 
> > I'm working on fixing that today, but I might not get it done.  If
> > you want to release without this fix, that's fine by me.
> 
> No rush, I can wait for this fix. Thanks.

Ok.


Tim
_________________________________________________
tmda-workers mailing list ([EMAIL PROTECTED])
http://tmda.net/lists/listinfo/tmda-workers

Reply via email to