[orbitcpp-list] Re: passing a reference to another process

[<MHL...@t-...>]

On 11/22/2000, 03:11:36 Phil Dawes wrote:
> Hi Martin,
> 
> Glad you solved your problem - sorry about my reply arriving too
> late.
> 
> I agree that it`s unfortunate that this sort of mistake results in
> such a nebulus error - can you think of a way we could trap this in
> orbitcpp?.
> 

The problem is that all variables xxx_ptr that get initialized from
the same reference xxx_var point to a "profile_list". This profile_list
is freed when xxx_var is destroyed (unless the reference is duplicated,
of course). Now can you look at a xxx_ptr and say whether
profile_list still points to a valid memory block or not ?
I found my bug because "next" in profile_list was 0x0 but that doesn't
necessarily have to be the case.

Hmm, perhaps we could define a macro like

#ifdef DEBUG
#define REFERENCE_IS_VALID(obj) \
	if(obj->profile_list) obj->profile_list->data->profile_type = \
		obj->profile_list->data->profile_type;
#else
#define REFERENCE_IS_VALID(obj)
#endif

and use it in the cpp-stubs before passing obj to the c-stub which
in turn passes it to liborbit where the segfault happens.
I think this macro should a segfault?! If yes the users attendence
should be drawn on the right track!

Anyway I think it would be good to add a few
comments to client.cc in the test everything. People that start using a new
library will first have a look at the examples. Maybe you could modify the
program a little so that it stores a refence to a Factory and use it later
in another function.


Cheers,
  Martin.