[Xevie] My current design efforts

[Jim Gettys]

FYI: IPC between processes, if both are paged in and ready to go, is
very fast these days.

e.g. x11perf reports a 50 microsecond client/server round trip....
(this is on a 1.2ghz Pentium III).
				- Jim

On Mon, 2004-10-04 at 09:53 -0700, Deron Johnson wrote:
> Hi Keith,
> 
> I wanted to update you on my current design efforts in the area of
> the event pathway. I'm currently investigating an approach which
> has the X server sending events to the Display Server to do the pick
> and the Display Server sends the events back to the X server to do the
> grab processing and event propogation. If the event is destined for an
> X window, the event is sent to the event client. But if the event is
> destined for a 3D object, the event is sent back to the Display Server
> for further propogation. This approach avoids the need to export window
> geometry information to the Display Server and it has no race conditions
> because all of the grab processing is done in one place (in the X 
> server). The down side of this approach is that there is extra latency
> for 3D object events (the time it takes to ship the post-grab-processing
> event back to the Display Server). I am currently trying to characterize
> whether this latency is significant.
> 
> Shared memory picking in the X server could provide an optimization for
> this design, but it's just an optimization; it's not central to the
> design. And it's looking like it will probably require a large 
> implementation effort in our scene graph library (Java3D). So I'll 
> probably only go that route if the non-shared-memory-picking approach
> has too great a latency.
> 
> But I still wanted to know why you said that you wanted to support
> systems that don't support shared memory. What type of systems are
> you talking about?
> 
> 
> _______________________________________________
> xevie mailing list
> xevie at freedesktop.org
> http://freedesktop.org/mailman/listinfo/xevie