commloops -> HEAD, round n+1

Hi,

its time for another round of commloops changes to make their way
into -HEAD.

Please check out the commloops branch on sourceforge for the diff.

In a nutshell:

* async/callback driven read() (comm_read()) and accept() (comm_accept())
  have been implemented
* most stream routines that perform socket reading have been converted
  over to the above routines (notable exceptions are the SSL negotation
  and ftp upload magic - they need a little more thought)
* none of the datagram based code has been converted over (icp, wccp, etc)

The aim of this work is to abstract out a lot of the comm assumptions
from the code - I'd like to support event-driven IO (kqueue, /dev/poll,
rt signals for those in linux land) and then move over to more estoric
(for UNIX!) methods of IO - scatter/gather IO and perhaps even win32
completion IO.

I'd like to commit this code in a few days. Its passed the browsing tests,
but I don't have enough machines here to put it through a polygraph test.
If someone could do it for me (whilst browsing, since polygraph doesn't test
error conditions :) I'd be very grateful.

Thanks!

Adrian
Received on Sat Oct 12 2002 - 21:26:11 MDT