6 Mar
2009
6 Mar
'09
7:44 p.m.
On 03/05/2009 12:14 PM, Klaus Darilion wrote:
Eric PTAK schrieb:
Seems the right way to do it. You can commit on trunk and let it there
for testing, then backport.
Cheers,
Daniel
--
Daniel-Constantin Mierla
http://www.asipto.com
Hi all for a new debug day ;)
You're right Klaus, from my understanding :
- the expires value is taken from the dialog struct, or from
override_lifetime if it's set.
==> at this point, dialog->lifetime = 0, so if override_lifetime isn't
set the expires will equal 0 too. so we need to go more deeply.
(not in call order)
- the dialog struct is given via the callback and filled by the dialog
module
- the dialog module calls get_dlg_timeout function which return the avp
lifetime or default_timeout if the avp doesn't exist
- the dialog module calls run_create_callbacks which then calls
registered callbacks like pua_dialoginfo's one
the weird thing, from dialog/dlg_handlers.c line 470 to 510, the dialog
struct is built but the lifetime field is filled after the
run_create_callbacks call...
I remember I have come across this before. Currently the order is:
/* first INVITE seen (dialog created, unconfirmed) */
run_create_callbacks( dlg, msg);
if ( seq_match_mode!=SEQ_MATCH_NO_ID &&
add_dlg_rr_param( msg, dlg->h_entry, dlg->h_id)<0 ) {
LM_ERR("failed to add RR param\n");
goto error;
}
if ( d_tmb.register_tmcb( msg, t,
TMCB_RESPONSE_OUT|TMCB_RESPONSE_FWDED,
dlg_onreply, (void*)dlg, unref_new_dialog)<0 ) {
LM_ERR("failed to register TMCB\n");
goto error;
}
dlg->lifetime = get_dlg_timeout(msg);
I wonder if it should be changed. Shouldn't run_create_callbacks()
executed later? E.g. If there is an error creating the dialog IMO it
makes no sense to call the callback at all.
Daniel, what do you think?