Re: idl-compiler creates 0xdeadbeaf generating skels inorbit-stable-0-5

[Sebastian Wilhelmi <wilhelmi ira uka de>]

Hi Dietmar and Martin,
 
> Martin Baulig wrote:
> 
> > Dietmar Maurer <dietmar ximian com> writes:
> >
> > > Martin Baulig wrote:
> > >
> > >           My problem is the following:
> > >
> > >          _impl_report_activation_succeeded(_ORBIT_servant, &(result), ev);
> > >      -   OAF_ActivationResult__free(&result, NULL, CORBA_FALSE); +
> > >      OAF_ActivationResult__free(&result, NULL, CORBA_TRUE);
> > >
> > > This seems to be the same error we have with strings - the sequence buffer
> > > point into the receive buffer:
> > >
> > >                result.res._u.res_shlib._buffer[_ORBIT_tmpvar_7] =
> > >                   (void *) _ORBIT_curptr;
> > >
> > > so we free it twice :-(
> > >
> > > But unfortunately I do not understand the idl compiler enough to fix that
> > > bug.
> >
> > Fix that bug ?
> >
> > This bug appeared somewhere between ORBit 0.5.6 and 0.5.7, to be more exact
> > between Jan 9 2001 and today. So someone should look at all changes in the
> > last two months to find out what went from.
> >
> > Unfortunately, ORBit is a module under Sopwith policy - there are people
> > who decide not to try to look at the problem when reading ``Do not touch
> > this file. Run 'rcs2log' or use bonsai'' in a ChangeLog.
> 
> We have changed the IDL compiler to remove some serious memory leaks. I turns
> out that this leads into more problems. But I think Sebastian is already
> working to fix.

Yeah, I have a fix for all the string_free problems, but Elliot doesn't seem
to want to include that. So I've reverted that change until Elliot approves my
patch or fixes the idl-compiler to not leak.

So we now have some more leaks again.

Bye,
Sebastian
-- 
Sebastian Wilhelmi
mailto:wilhelmi ira uka de
http://goethe.ira.uka.de/~wilhelmi