Ethereal-dev: Re: [Ethereal-dev] [patch] add TCP desegmentation to laplink dissector

Note: This archive is from the project's previous web site, ethereal.com. This list is no longer active.

From: "Ronnie Sahlberg" <ronnie_sahlberg@xxxxxxxxxxxxxx>
Date: Wed, 22 Oct 2003 20:18:40 +1000
From: "Brad Hards"
Sent: Sunday, October 19, 2003 6:51 PM
Subject: [Ethereal-dev] [patch] add TCP desegmentation to laplink dissector


> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> The attached patch adds desegmentation support to the laplink dissector.
>
> I had a lot of grief testing it, courtesy the default-off reassembly
option
> for the underlying TCP dissector.  Any reason why it should default to
off?
> If not, I'd like to change it to default on.
>
> Brad

The discussion of the default setting for TCP reassembly kind of died out so
lets retart it.

Reassembly is in general disabled since it consumes so much memory at
runtime, something that
really hurts people that deal with very large datasets (another thing that
hurts them is the very slow speed
of ethereal but that is a different discussion/project).

It is unfortunate and a failure that this setting is not intuitive to use.


I think that reassembly should still be disabled by default but maybe it
could be made more intuitive to the end user if
1, we added "This option requires TCP desegmentation to be enabled" to all
tooltips for dissectors doing tcp reassembly.
2, we change the TCP desegmentation option to be enabled as default and we
change all subdissectors using TCP desegmentation to be disabled by default.
Then people that wants reassembly of protocol x to work just has to enable
it at the preferences for protocol x and can forget about tcp.
or
3, remove the option to enable/disable TCP desegmentation completely and
hardcode it as enabled, then change all subdissectors
so they default to disabled?

With 3 the option to control reassembly at the tcp layer would be redundant
and the option could be removed.


comments?