Re: truss issue

Don Lewis wrote on Mon, Dec 15, 2003 at 01:28:50PM -0800: 
> On 15 Dec, Martin Cracauer wrote:
> > Don Lewis wrote on Mon, Dec 15, 2003 at 01:13:58PM -0800: 
> 
> >> Hmn, I wonder if it would be cleaner to exec() the executable to be
> >> traced in the parent process and run truss in the child ...
> > 
> > I think I misunderstand.  The parent is usually your login shell, you
> > don't want that one to exec() anything.
> 
> If you type "truss foo" at the shell prompt, the shell will fork and
> exec truss, which forks and execs foo.  When truss forks, the child
> process is the one that execs foo, and the parent process watches what
> the child process does.  My suggestion is to swap the roles of the
> parent and child truss processes.  The parent truss process would be the
> one that calls exec(), and the child process would be the one doing the
> monitoring.  When the process being traced exits, the shell would
> automagically get the correct exit status.

Ah, OK.

That would also have the advantage that the pid that the fork() in the
starting process gets will continue to stay correct for the child it
expects.

E.g.
truss foo &
pid=$!
dosomethingwith $pid

pid will point to the actual process and not foo, so you can safely
insert a truss prefix where you want.

I've seen quite some things blow up when breaking the assumption that
the child is directly what is expected, usually by people trying to be
cute and sneak in a shellscript wrapper around a real program and not
exec'ing the wrapper away.

Martin
-- 
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Martin Cracauer <cracauer_at_cons.org>   http://www.cons.org/cracauer/
 No warranty.    This email is probably produced by one of my cats 
 stepping on the keys. No, I don't have an infinite number of cats.