Wireshark · Ethereal-dev: Re: [Ethereal-dev] hanging graphs

[Guy Harris <guy@xxxxxxxxxx>]

On Wed, Dec 18, 2002 at 11:40:37PM +0100, Martin Regner wrote:
> I noticed that the time stamps are incorrect for certain frames,
> resulting in negative time differences between packets.

That's what I suspected the problem would be.

I've checked in a change to "gtk/tcp_graph.c" not to assume that the
time stamp of the last frame is the largest time stamp in the packet;
instead, the loop that computes the maximum Y-axis value now also
computes the largest time stamp.

That appears to clear the hanging problem, although, as you note:

> I guess that even if the hanging problem is solved you will get very
> strange results for the TCP analysis if you don't get rid of the
> timestamp problem.

...the effect of the change will probably be to replace a hang with a
not-entirely-sensible graph.

> I also noticed that I got hangings for Statistics/IO/IO-Stat with the
> original capture, but not with the modified capture.

Hmm.  I don't see that with the current CVS version (I didn't modify the
io_stat.c code) on Solaris, but I do get a hang with 0.9.8 on Windows.
I don't get one if I build the current CVS version with the 0.9.8
version of io_stat.c replacing the current CVS one, and there's nothing
obvious in the changes since 0.9.8 that would fix the problem.