Re: NT port

On Sun, Jan 07, 2001, Henrik Nordstrom wrote:
> Robert Collins wrote:
>
> > I'll check from home what version I've got. I was thinking of creating a
> > diff to 2.3stable4 (which is what his code is based on), and then
> > branching off of 2.3stable4, apply the diff, then cvsmergeinit to HEAD.
> >
> > But I'm happy to leave it to you :]
>
> Done, and attempted to bring it up do date with Squid-2.4. HEAD will
> follow once the Squid-2.4 port is verified.
>
> Yes, there design issues in the port which is questionable, so for the
> HEAD port I'd like to see if there is a more clean cut way to aceive the
> same thing.
>
> Adrian: What is the state of commloops? As you said the NT port might
> benefit from that.

The trouble with commloops is that I need to find a clean way to
destroy deferred reads so the codebase becomes event-IO friendly.
Otherwise we end up with the same problems as I had with
kqueue and devpoll where the deferred reads started to lock the
CPU usage right up.

I just changed how half-closed connections are handled - as long
as there is only ever one CommStateData per fd then it should work
ok.

Everything else in commloops seems to be working just fine. I just
have to tidy up the server copying logic .. :)

Adrian

-- 
Adrian Chadd			"Here's five for the cake, and
<adrian@creative.net.au>	  five to buy a clue."
				    - Ryan, Whatever it Takes