Re: [exprimental patch] wakeup instead of polling for dns

On Sun, Feb 12, 2012 at 08:44:04PM +0100, Johannes Hofmann wrote:
> Hi Jorge,
> 
> On Fri, Feb 10, 2012 at 01:40:10PM -0300, Jorge Arellano Cid wrote:
> > Hi Johannes,
> > 
> > On Mon, Feb 06, 2012 at 09:45:44PM +0100, Johannes Hofmann wrote:
> > > Hi,
> > > 
> > > attached experimental patch replaces the polling for DNS answers
> > > with the wakeup mechanism fltk provides.
> > > What do you think?
> > 
> >   Well dns resolving is delicate, but not sacred ;)
> > 
> >   AFAIS  the main advantage would be a near 0.1 second speedup on
> > average  when  displaying  non dns-cached pages, (0.2 best case).
> > Which is significative.
> > 
> >   Please let me know if there's another advantage I miss.
> 
> Well, I don't like the polling :-)
> There might be an increased power consumption related to the
> 5 times/s timer, but I'm not sure whether it really matters.
> 
> > 
> >   The  cost  is using an FLTK mechanism (awake) which we don't in
> > the rest of the code.
> 
> Yes, plus we need to switch on locking for all fltk just for the dns
> thing. That's the major drawback I think.
> 
> > 
> >   It looks clean. It should be tested on different platforms/OS.
> > 
> >   BTW,  dns.c has code for a blocking server (not threaded). Does
> > this  awake() call work OK when called from the same main thread?
> > Is  this by design or by chance? Maybe it would be safer to avoid
> > calling awake() in this case.
> 
> I didn't test it yet. I still want to try other approaches that
> would avoid switching on the fltk threading stuff with the
> Fl::lock() call.

Ok, attached is another approach that avoids the Fl::lock() and
instead uses a Fl::add_fd() which is already used in src/IO.
Let me know what you think.

Cheers,
Johannes