Hi all,

I have observed when looping a request through SER (using t_relay_to("127.0.0.1", 5060)) that sometimes the via branch parameter value generated when the request goes out is kept equal to the value in the previous via (when this request was looped), and two equal vias go out. This causes problems with on_reply routes. On_reply routes are then called only once, instead of two. Since these are 2 independent transactions, this should happen twice.

Browsing through tm code, it seems that this is caused by the usage of syn_branch == 1 in main.c (the default), leading to the following function calls when an initial INVITE loops:

tm/t_fwd.c: t_forward_nonack add_uac print_uac_request

tm/t_msgbuilder.c: t_calc_branch

msg_translator.c: branch_builder <-- here, when syn_branch==1, the transaction 'label' value is used for the branch value, but according to tm/h_table.h, this value is "sequence number within hash collision slot".

Is this the reason why some times the branch value is the same in two transactions looped through SER? The label value can perfectly be the same across 2 transactions not in the same hash slot... right? Maybe this has nothing to do with the request being looped, or it just makes this situation more probable? Anyone else can replicate this behaviour?

Thanks,

João Filipe Plácido -- PT Inovação