[cairo] Banker's rounding problems and Nautilus

[Owen Taylor]

On Tue, 2006-12-05 at 15:11 -0500, Behdad Esfahbod wrote:
> On Tue, 2006-12-05 at 14:24 -0500, Owen Taylor wrote:
> > On Tue, 2006-12-05 at 14:14 -0500, Behdad Esfahbod wrote:
> > > On Fri, 2006-12-01 at 20:35 -0500, Carl Worth wrote:
> > > > On Wed, 29 Nov 2006 18:24:14 -0800, "Daniel Amelang" wrote:
> > > > > > Oh, and I've also gotten feedback about a pretty bad text positioning
> > > > > > bug for the text nautilus is drawing on the background window. I think
> > > > > > that's a problem from the new banker's rounding, but I'll bring that
> > > > > > up in a new thread.
> > > > 
> > > > So the bug has already been reported with a nice screenshot here:
> > > > 
> > > > 	Update to cairo-1.3.4 worsen font rendering
> > > > 	https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=217819
> > > > 
> > > > The problem with bankers' rounding being used during text positioning
> > > > is that it's not invariant under translation. So when the text is
> > > > drawn once in black, and then once again with an offset in white, (for
> > > > the drop-shadow effect), it is getting rounded differently.
> > > > 
> > > > Meanwhile, within a single word, you can see "extra space" being
> > > > inserted between letters, (as if one is getting rounded left while the
> > > > next is getting rounded right). I had understood Behdad to say that a
> > > > pango-using application would be using only integer metrics, so I am a
> > > > bit surprised to see this problem.
> > > 
> > > So yeah, my bad previously.  Seems like in None and Slight hinting modes
> > > in gnome-font-properties, you get non-integer metrics.
> > 
> > If that's the case, it's a bug. It needs to be fixed, not worked around
> > somewhere. (In fact it's a bug popped up about a year ago and was
> > supposedly fixed.)
> > 
> > Metrics should be hinted without regard to the gnome-font-properties
> > settings ... a lot of apps can't handle non-integer metrics.
> 
> Yeah, that's what I remember too.  No idea what's going on.  Looking
> into it.  Still, Banker rounding is probably a bad idea.

Yep, agreed. Translational invariance is highly desirable. (Despite what
the C standardizers did for %)
					- Owen