On Mon, 9 Jan 2006, izimmerman@borderware.com wrote:
> ares_fdset? Do you mean ares_fds?
Oops. Yes I meant that.
> That won't do it. ares_getsock would, but it does mean getting the CVS; oh
> well, I guess I have to do that anyway to contribute the DNS packet code,
> but I'll have to convince the superiors it's safe. Or should I wait for the
> next release?
That's your pick. I plan to make a release "soon", but I'm not sure exactly
when that is. There's no real rush to release as I see it.
> You see, the natural algorithm in this situation is _not_ to go through all
> bits in reads[] and test each; that's too wasteful of time. The natural way
> is to know the descriptors you have open, and test those bits only.
I'd even say that the most natural way is to stay away from select() all
together! ;-) (I'm doing work on libcurl for >10000 connections atm, and then
select() just can't cope - which is the main reason I added ares_getsock()...)
-- c-ares -- my preferred DNS asynch resolver libraryReceived on Mon Jan 9 17:00:38 2006