On Tue, Mar 10, 2009 at 11:53:58AM +0100, Simon Thum wrote: > Peter Hutterer wrote: > > yeah, of course. sorry. what I meant was that the value x is sent as x * (1 > > << > > 16). A little test program for random values between 0 and 0xFFFF showed an > > average rounding error of 0x000008. > I can't judge whether that is good or bad :) > IMO for the protocol, definition is more important than impl. 1<<16 > should be valid, one can work out rounding later.
that should read "rounding error of 0.000008" of course. 8 millionth of a pixel rounding error should be within reasonable margin of error for most applications. anyway, I finally wrote that code and decided for INT16.CARD16 as wire format. Cheers, Peter > > tbh. I am also contemplating to just send IEEE floats/doubles over the wire > > and get on with life. > I'm not really against it, but that's essentially wasting the exponent > bits. I'd say fixed is a better match, although I'd prefer 24.8 for > on-screen coords. I don't think we have information beyond n.4 in any case. _______________________________________________ xorg mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/xorg
