Re: Message negociation

On Thu, Jan 08, 2004 at 09:43:24AM +0100, Steve Lhomme wrote:
> Toby Hudon wrote:
> 
> >>I was just thinkING about pointers...
> >>
> >>We said that we need messages but that treating them all the time in a 
> >>switch could have an impact on performance.
> >>
> >>So what about having a filter ask for a function to the filter it wants 
> >>to send the message. It only has to do this once and then call this 
> >>function instead of sending a message.
> >>
> >
> >
> >Are we really so starved for cpu power that a couple hundred compares of a 
> >variable to a fixed value will kill us? How slow is it really to do "does 
> >X = #?" for each entry in the switch when you pass it?
> 
> Not really. But that equivalent to the message-oriented system. But more 
> efficient CPU wise. And would probably lead to cleaner code too (on both 
> sides). During the init/connection of 2 filters they request all the 
> callbacks they need once. And then they just call functions the way they 
> always do it...

the vtables with NULLs for not-implemented "methods" will be even more
efficient, cause you request all callbacks at once ;)

-- 
gr{oe|ee}t{en|ings}
artm