> 
> I think this is how it work.
> 
> Using the STREAMS API, you:
> 
> - use t_open to create a socket in non-blocking mode
> 
> - use t_bind and t_listen to start waiting for incoming connections

Use t_bind() to assign a local address and say that you want incoming
connections (the CONIND_number parameter).
 
> - use t_look to wait for a T_CONNECT event

Then/or use t_listen to give you the incoming call message.

> At this point, you have the option of accepting the connection (using  
> t_accept) or disconnecting (using t_snddis).

From the XTI/TLI/TPI point of view yes.

> t_accept will send a  
> TCP_SYN and completion the open, while t_snddis will send a TCP_RST  
> (since the connection is not yet open).

Neither the Lachman STREAMS TCP stack (in SVR4), nor any stack Sun have
issued [1] the SYN-ACK in response to the t_accept().

> The Mac OS STREAMS implementation (based on Mentat's code) made the  
> t_look stuff a little easier by allowing you to specify a callback  
> when the event came in, so it was a little easier to handle.

Actually it is almost easier to use the TPI and getmsg/putmsg directly!
Additionally that removes all the problems the XTI library has in
supporting t_sync() and t_getstate() - both of which are actually
impossible.

	David

[1] To my knowledge
oh and see: www.opengroup.org/onlinepubs/9694979699/front.htm#tagcjh_01_03

-- 
David Laight: david@l8s.co.uk