Ethereal-dev: Re: [Ethereal-dev] shared libraries

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 This list is no longer active.

From: Ed Warnicke <hagbard@xxxxxxxxxxxxxxxxxxx>
Date: Mon, 19 Mar 2001 23:41:59 -0500 (EST)
I got a bit carried away in suggesting that libui, libdfilter, and 
libftypes all be shared libraries, I withdrawl that part of my suggestion.
I have noted that wiretap currently seems to stand on it's own.  The 
dependencies in libethereal appear at a cursory examination to be

packet.c:	depends on 
		packet-atalk.h
		packet-frame.h
		packet-ipv6.h
		packet-sna.h
		packet-vines.h
		packet-osi.h

resolv.c:	depends on 
		packet-ipv6.h
		packet-ipx.h


proto.c:	depends on 
		packet-ipv6.h
		register.h

I suspect that all of the dependencies on packet-*.h are simply 
utility functions related to those protocols and that those utility 
functions can be abstracted out into an libethereal itself (
util-ipv6.{c,h} etc. ) 

The dependency on register.h is a little more basic, as it 
speaks to how we handle dissectors.  I've considered a number of 
possible ways around this problem.  

1)	Convert the existing dissector base into individual plugins.  
	
	This has the advantage of being highly 
	modular.  It has the disadvantage that we are likely
	to run into some unhappy dependency issues that will have 
	to be resolved along the way, it will be a fair bit of 
	work, and that in almost any large undertaking there 
	is the risk of unpleasant unforseen consequences.

2)	Convert the existing dissctor base into one jumbo plugin.
	I don't see anything in the plugin mechanism that would 
	prevent the existing dissectors from being 
	rolled into one big plugin that registers all of the 
	protocols contained in it when plugin_init is called and 
	registers all of it's handoffs when it's plugin_reg_handoff is 
	called.  

	This has the advantage of being quick and easy.  It has the 
	disadvantage of being ugly as hell and extremely non-modular.


3)	Start with solution 2, as it will allow us to cleave libethereal
	off now ( at least internally to ethereal) and make it a shared
	object.  Proceed to move towards 1, as it is the right thing 
	to do in the long run.  


I favor the third way, but am quite interested in other opinions.

Ed

On Mon, 19 Mar 2001, Gilbert Ramirez wrote:

> Ed Warnicke wrote:
> > 
> > Is there a reason that libethereal, libdfilter, libftypes, libwiretap,
> > and libui aren't being built and installed as shared libraries?
> > 
> > I know there is some mention of doing this in the epan directory,
> > but I was wondering if there was a good reason this hadn't been done yet.
> > 
> > Ed
> 
> 
> There's no reason not to ship wiretap as a shared libary.
> Libethereal, however, still needs a little cleanup. There are some
> functions
> that it uses that are in the dissectors code. I tried it one time, but
> I don't remember exactly which functions. It had to do with address
> types
> used in packet.c that exist in the dissectors.
> 
> Hmm, maybe some better documentation on how to use wiretap would be
> good.
> Once its a shared library, there's a better chance that other projects
> will
> want to use it.
> 
> libdfilter and libftypes were meant to always be archives. When
> libethereal
> is a shared library, libftypes and libdfilter will be archives of PIC
> objects
> that are linked into libethereal.
> 
> Libui was also never meant to be a shared library.
> 
> --gilbert
>