On Sat, Oct 17, 2009 at 9:27 AM, Juha Heinanen jh@tutpro.com wrote:

Jan Janak writes:

> I wrote a very simple sip-router script to test serial forking based  > on q values with t_load_contacts and t_next_contacts. I noticed that  > when I use t_next_contacts before the transaction exists, it uses the  > value of fr_inv_timer_next even for the first branch.  >  > Is this itendended behavior? Shouldn't the function use the value of  > fr_inv_timer for the first branch and then fr_inv_timer_next for  > subsequent branches?

jan,

it is the intended behavior and it has always been like that.  it has even been documented in the readme:

Function t_next_contacts() sets fr_inv_timer to fr_inv_timer_next value   if, after t_next_contacts() is called, there are still lower qvalue   contacts available, and to fr_inv_timer value if there are not.

the idea is that using fr_inv_timer_next, one can shorten the "ringing" if there are still other contacts to try.  if the set of contacts to try is that last one (or the only one), "ringing" time is normal, i.e., specified by fr_inv_timer.  if you would wait for normal time already on the first set, the call would timeout before you have had a change to try the remaining contacts.

Ohh, I see. I did not get that from the documentation and the code. So, if I understand it correctly, what you are saying is that fr_inv_timer_next is used if serial forking takes place and in that case it is used for all branches except those that are tried last. The last set of parallel branches in serial forking uses the usual fr_inv_timer.

I think we might need to clear this up a little in the readme because I did not understand it from the documentation. In fact, I was wondering why t_next_contacts resets the timer to fr_inv_timeout at the end and it makes sense now. Thanks.

-- Jan