Interconnect Software Consorium Socket API Extensions Working Group.
--------------------------------------------------------------------

Meeting:
	Thursday 12th September 2002
	20:00GMT (21:00BST, 04:00p EDT, 03:00p CDT, 01:00p PDT,
		 22:00CEST, 05:00JST+) 

	Toll Free (US & Canada):          1-888-742-8686
	International:                    +1-303-928-2600
	Conference ID:                    3378790

	Alternate teleconference numbers:
	Toll Free (US & Canada):	  1 877 214 5010
	International:			  +1 (504) 588 9772
	Code:				  769497

--------------------------------

Attendees:
	Vish Subramanian, HP
	David Edmondson, Sun
	Satya Sharma, IBM
	Masanori Itoh, Fujitsu
	Juliana Hsu, IBM

Apologies:
	Louis Laborde, HP
	Jeff Messing, IBM
	Annie Foong, Intel

--------------------------------

Agenda:

1. Agenda bashing.
   dme had no idea.

2. Review action items.
   No change.

3. Approve Last Minutes.

4. Strawman API proposal review (Vish)

   See version 0.2 posted by Louis on the 5th September.

   Event queue discussion is complete, need to start with the async
   initiation calls.

   * Each call has a "MANYPARAMS" version.  Is there preference for
     one or the other ?

     dme: prefers MANYPARAMS as it is more like the existing sync
          calls.
     Satya: no preference.
     Juliana: perhaps the version with the structure would allow
     	      easier extensibility.

     dme: leave it with both for now, wait to see if a compelling
     	  argument for one or the other arrives.

   * async_send():  support for synchronous return is a requirement.

     Should we allow partial synchronous completion with an async
     initiation for the remainder ?

     Options:
     + return the number of bytes send synchronously and do nothing
       with the remainder (leave it to the application),
     + initiate the remainder for asynchronous completion.

     dme: sync completion is a pain.  perhaps we should get rid of it.

     Vish: if this happens, what should be the behaviour where only
     	   part of the buffer is accepted for async initiation.

     dme: Need a proposal for the removal of synchronous completion
     	  and partial async initiation.  ACTION: dme

     discussion of ENOTSOCK
     discussion of EMSGSIZE (falls out of the above)
     discussion of EAGAIN: should "socket send buffer full" be
     		something else.

     think about error codes for async completions, i.e. can we just
     use errno values, or is there a reason why this is bad.

   * async_sendto(): similar issues to the above.

   * async_sendmsg()/async_recvmsg(): mreg_handle needs to be a
     pointer (one per element of *message) in the ASYNC_MANYPARMS
     case.

   * async_recvmsg(): the data structures indicated by the msghdr
     pointer need to be maintained by the caller across the
     initiation/completion cycle in order that they can be used to
     indicate state and error conditions to the caller.  similarly for
     async_recvfrom(), async_sendfilev().

4. AOB
   None.

--------------------------------

Outstanding Actions:

2002-04-18/02 dme.
	Draft a paper describing why a unified socket/file IO API
	would be useful.

2002-08-06/01 dme.
	Produce a clarification on the results of the C3.1 and C7
	vote.

2002-08-06/02 Jeff.
	Write a sketch of the O2 API.

2002-08-08/01 Jeff.
	Write a proposal to modify the requirements in order to allow
	per-process binding of socket and event queue.

--------------------------------

Next Meeting:
	Tuesday 17th September 2002
	20:00GMT (21:00BST, 04:00p EDT, 03:00p CDT, 01:00p PDT,
		 22:00CEST, 05:00JST+) 

	Toll Free (US & Canada):	  1 877 214 5010
	International:			  +1 (504) 588 9772
	Code:				  769497

--------------------------------

Contact Details:
	email:	icsc-socketwg@opengroup.org
	web:	http://www.opengroup.org/icsc/sockets/protected