On Thu, Oct 03, 2013 at 06:07:42AM -0300, Rogério Brito wrote:
> It is a common method, indeed. I don't know how/why mtr uses a
> pseudo-random generator, though, without having read the code (will
> read that later, if I still have sufficient interest).
mtr sends out (crafted, non-regular) packets and looks at what comes
back as error messages. At some point in the past, close to the start
of the development of IP (i.e. 1980ies), someone made a mistake that
makes it difficult to determine WHAT packet the returned error
corresponds to. So we have to trick things a bit to make sure that we
associate the response with the correct probe. If we use a
deterministic (e.g. next = cur+1) method, two mtr's running on the
same machine starting out with similar "seeds" (the first "cur") will
continue to recieve eachother's responses. By using a pseudorandom
sequence they might occasionally match a response to the other mtr to
something they sent themselves. But that is very rare. (I think we
have about 15 or 16 bits, so 20 (average # of hops)/30000 might be a
good guess).
> > If that works, it's a bug that has been fixed in the main codebase:
> > either work around it by supplying a hostname, or upgrade if you want.
>
> Yes, passing an argument works fine. Perhaps we need the newest version of
> mtr packaged in Debian...
I haven't released the "next" version yet. I think Robert is up-to-date
with the latest release. This is first on my platter...
(next = 0.86, I should have a script that just makes a release, but
that script isn't perfect yet, and needs some hours of debugging which
I can only do when I actually have a release to do AND have the time
to debug the script)
> > (I personally think the default of "localhost" is untidy: cleanlyness
> > would require mtr to say: "no host to trace provided. Bye!" (or just
> > exit wihtout doing anything as it does in your case). However that
> > means that starting mtr from a menu in a gui without arguments would
> > make it close immediately.)
> I don't really think that it is *that* unclean of a solution. OTOH,
> just quitting without any message violates the spirit of Unix of
> being noisy when errors occur, while being silent when everything is
> fine.
It isn't an error: You specified nothing to do, so I do nothing.
But agreed, rm without arguments also considers this an error. On the
other hand, in C:
for (i=0;i<n;i++) {
...
}
will run the loop once with n=1, and silently no times at all when you
specify n=0. (which in fact is close to what happened inside mtr
without arguments).
> If mtr issued a message that I missed an argument, then I would not have
> sent this bugreport in the first place. :)
Agreed. But the "default is localhost" is a very old feature. It was
intended to keep on working. Some feature (multiple hosts) was added
that broke this. noone has wanted to change the old default behaviour.
> Oh, I have a feature request: can you put a button to toggle the name
> resolution in the GUI. :) This way, if name resolution is getting in the
> way, then we can turn it off (or on) very easily.
Yes. that'd be nice.
Could you paste that into a feature request at
https://launchpad.net/mtr/+bugs
Roger.
--
** r.e.wo...@bitwizard.nl ** http://www.BitWizard.nl/ ** +31-15-2600998 **
** Delftechpark 26 2628 XH Delft, The Netherlands. KVK: 27239233 **
*-- BitWizard writes Linux device drivers for any device you may have! --*
The plan was simple, like my brother-in-law Phil. But unlike
Phil, this plan just might work.
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
