as per daniel's suggestion, i wrote a new tmx function t_reuse_branch(), which can currently be called from branch-failure route:
/** * Creates new "main" branch by making copy of branch-failure branch. * Currently the following branch attributes are included: * request-uri, ruid, path, instance, and branch flags. */ static int w_t_reuse_branch(struct sip_msg* msg, char *p1, char *p2) { struct cell *t; int branch;
if (msg == NULL) return -1;
/* first get the transaction */ if (_tmx_tmb.t_check(msg, 0) == -1) return -1; if ((t = _tmx_tmb.t_gett()) == 0) { LM_ERR("no transaction\n"); return -1; } switch (get_route_type()) { case BRANCH_FAILURE_ROUTE: /* use the reason of the winning reply */ if ((branch = _tmx_tmb.t_get_picked_branch()) < 0) { LM_CRIT("no picked branch (%d) for a final response" " in MODE_ONFAILURE\n", branch); return -1; } rewrite_uri(msg, &(t->uac[branch].uri)); set_ruid(msg, &(t->uac[branch].ruid)); if (t->uac[branch].path.len) { set_path_vector(msg, &(t->uac[branch].path)); } else { reset_path_vector(msg); } setbflagsval(0, t->uac[branch].flags); set_instance(msg, &(t->uac[branch].instance)); return 1; default: LM_ERR("unsupported route_type %d\n", get_route_type()); return -1; } }
the new "main" branch does not include destination uri and socket, because i didn't find them in struct ua_client. unless someone adds them to struct ua_clients, they thus need to be set separately from htable after calling t_reuse_branch().
is it ok that i commit the above function to tmx module?
in addition, i would like to add $T_reply_bf (branch flags) pv, since i need to test some of the branch flags in other parts of branch-failure route.
-- juha
Hello,
the function can be committed.
For the variable, I would go for a new class, $T_branch(attr). You can implement the part for flags as $T_branch(flags). Other attributes can be implemented later under the same class.
Cheers, Daniel
On 19/04/14 14:57, Juha Heinanen wrote:
as per daniel's suggestion, i wrote a new tmx function t_reuse_branch(), which can currently be called from branch-failure route:
/**
- Creates new "main" branch by making copy of branch-failure branch.
- Currently the following branch attributes are included:
- request-uri, ruid, path, instance, and branch flags.
*/ static int w_t_reuse_branch(struct sip_msg* msg, char *p1, char *p2) { struct cell *t; int branch;
if (msg == NULL) return -1;
/* first get the transaction */ if (_tmx_tmb.t_check(msg, 0) == -1) return -1; if ((t = _tmx_tmb.t_gett()) == 0) { LM_ERR("no transaction\n"); return -1; } switch (get_route_type()) { case BRANCH_FAILURE_ROUTE: /* use the reason of the winning reply */ if ((branch = _tmx_tmb.t_get_picked_branch()) < 0) { LM_CRIT("no picked branch (%d) for a final response" " in MODE_ONFAILURE\n", branch); return -1; } rewrite_uri(msg, &(t->uac[branch].uri)); set_ruid(msg, &(t->uac[branch].ruid)); if (t->uac[branch].path.len) { set_path_vector(msg, &(t->uac[branch].path)); } else { reset_path_vector(msg); } setbflagsval(0, t->uac[branch].flags); set_instance(msg, &(t->uac[branch].instance)); return 1; default: LM_ERR("unsupported route_type %d\n", get_route_type()); return -1; } }
the new "main" branch does not include destination uri and socket, because i didn't find them in struct ua_client. unless someone adds them to struct ua_clients, they thus need to be set separately from htable after calling t_reuse_branch().
is it ok that i commit the above function to tmx module?
in addition, i would like to add $T_reply_bf (branch flags) pv, since i need to test some of the branch flags in other parts of branch-failure route.
-- juha
sr-dev mailing list sr-dev@lists.sip-router.org http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-dev
Daniel-Constantin Mierla writes:
For the variable, I would go for a new class, $T_branch(attr). You can implement the part for flags as $T_branch(flags). Other attributes can be implemented later under the same class.
i looked at implementation of current classes $T_req(pv), $T_rpl(pv), etc. looks like they make a copy of the whole environment and then access the pv using normal pv access functions. this sounds very heavy as compared to implementation of item functions $T_reply_ruid, $T_branch_idx, etc. i would thus prefer to implement $T_reply_flags the same way as the item functions is done.
-- juha
On 20/04/14 12:19, Juha Heinanen wrote:
Daniel-Constantin Mierla writes:
For the variable, I would go for a new class, $T_branch(attr). You can implement the part for flags as $T_branch(flags). Other attributes can be implemented later under the same class.
i looked at implementation of current classes $T_req(pv), $T_rpl(pv), etc. looks like they make a copy of the whole environment and then access the pv using normal pv access functions. this sounds very heavy
Those classes of variables needed to clone the buffers in order to parse them. But for $T_branch(flags) it is not need, you can look at how $T(key) is implemented, or any other class with inner names ($time(key), $branch(key)).
It is better structured as there is going to be $T_branch(uri), ... and other attributes.
as compared to implementation of item functions $T_reply_ruid, $T_branch_idx, etc. i would thus prefer to implement $T_reply_flags the same way as the item functions is done.
I don't get the relation between reply and branch flags to have a reason to call this T_reply_flags. Cheers, Daniel
as per daniel's suggestion, i added $T_branch(flags) pv. should $T_branch_idx and $T_reply_ruid be now renamed to $T_branch(idx) and $T_branch(ruid)?
-- juha
On 06/05/14 16:48, Juha Heinanen wrote:
Juha Heinanen writes:
as per daniel's suggestion, i added $T_branch(flags) pv. should $T_branch_idx and $T_reply_ruid be now renamed to $T_branch(idx) and $T_branch(ruid)?
i haven;t heard anything about this. can i just go head with renaming?
Not for $T_branch_index, in failure route it returns the index of the new branch, not the winning failed branch, so there are different things. Reply ruid should be renamed as it has nothing to do with the reply, but with the branch -- however, I haven't looked at it and what is doing in failure route and other routing blocks.
Cheers, Daniel
Daniel-Constantin Mierla writes:
On 06/05/14 16:48, Juha Heinanen wrote:
Juha Heinanen writes:
as per daniel's suggestion, i added $T_branch(flags) pv. should $T_branch_idx and $T_reply_ruid be now renamed to $T_branch(idx) and $T_branch(ruid)?
i haven;t heard anything about this. can i just go head with renaming?
Not for $T_branch_index, in failure route it returns the index of the new branch, not the winning failed branch, so there are different things. Reply ruid should be renamed as it has nothing to do with the reply, but with the branch -- however, I haven't looked at it and what is doing in failure route and other routing blocks.
ok, i'll pass, since it is more than just branch failure issue
-- juha
-
Daniel-Constantin Mierla -
Juha Heinanen