On Thu, 14 Mar 2002, Michael Matthew Toomim wrote:
> Mark Vojkovich wrote:
>
> > This isn't going to work. Your client gets the notification
> > of the event some time after it happened. If you try to push it
> > a bit beyond where it is to simulate acceleration you'll usually
> > end up pushing it backwards. Most of the time (probably nearly
> > all of the time) more events would have been generated by the
> > time the server even got the request from your client and the
> > cursor would have already moved way beyond that point. I don't
> > see how this sort of thing can be made to work.
>
> So are you saying that there's a significantly large delay (> .01
> seconds) between the time that a client says "set cursor position" and
> the time that the cursor position is set?
There can be. Also, the data the client has is old when it
gets it. These are separate processes communicating through
a unix domain socket. On a single-cpu system these two apps,
client and server, are sharing the cpu with other processes.
The act of the server sending an event to the client and the
client sending a request back requires two context switches.
When the server sends motion events it's potentially sending them
to many clients. It's a race for who wakes up first and gets
a slice of the cpu. Maybe you could have your app run as root
and unnice it to the max, but I think you'll still get a very
jittery cursor.
Mark.
_______________________________________________
Xpert mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/xpert
