On 12/31/2014 12:52 PM, Bill Meier wrote:
On 12/31/2014 11:31 AM, Evan Huus wrote:
This is an init routine, which can be called when no file is in scope,
so wmem_file_scope() is incorrect (and se_* was also incorrect).
I'm actually not sure what this routine is doing, since it deals with
conversations but there will never be any conversations e.g. on
startup when the init routine is actually called...
I realized (after I did the commit) that there must be a reason why a
just a few ep/se calls remained (all to related to the use of UATs ?). :)
This thing looks like it's setting up conversations based upon UAT info
prior to dissection (presumably so that any subsequent dissection will
be able to take advantage of info stored in the conversation structs).
Taking a quick look at things I see that in packet.c the init-routines
are called after setting up file scope and initializing the conversation
stuff, so it would seem that using file_scope might be OK.
se_free_all();
wmem_enter_file_scope();
...
epan_conversation_init();
...
g_slist_foreach(init_routines, &call_init_routine, NULL);
Am I missing something ?
Are there other cases where the init routines are called out of file
scope ?
A problem I do see: if the UAT is changed while a file is open, there's
no update of the conversation table from the UAT. I don't know if
there's something in the UAT info which might change the dissection.
Maybe this is why the code checks to see if there'san existing
conversation (altho currently there can't be any since the code is only
exec'd as an init.
Oops: I see that cleanup_dissection() also calls the init routines
(after doing conversation_cleanup()), so I guess that trying to add info
to the conversation structures at this point might not work too well.
I'll have to dig a little deeper to see what really happens now and what
might be done for all of this.
Bill