[cairo] Non-rectangular sub-pixels

[Bill Spitzak]

> On 8/11/05, James Cloos <cloos at jhcloos.com> wrote:

>>It seems like cairo (and Xft) will need to add support at least for
>>panels that have the pixels rotated from the conventional layout with
>>either constant or alternating rotations, and perhaps also for panels
>>with chevron-shaped pixels.
>>
>>Obviously, simply scaling width by three and multiplying with the
>>colour matrix as done in Xft is not enough for these panels.  And I
>>can't quite picture what the transform matrix does need to be, but
>>am I correct that the math remains straitforward?  That it is still
>>an affine transform?  At least the sub-pixels are still trapezoids
>>or pairs of trapezoids....

Since the pixel centers are still in a rectangular grid, the transform 
is not different. What would change is the filtering of the transformed 
image used to produce each pixel. Roughly speaking, the pixel should be 
as bright as the amount of the letter that intersects that 
chevron-shaped region, which would be different than the amount that 
intersects a rectangular region. However the algorithim being used is 
only a rough approximation of this intersection anyway, plus real 
sampling theory makes the intersected area a much more blurry and round 
sinc pattern, so I find it hard to believe the difference between these 
is going to be visible, the difference is probably an order of magnitude 
smaller than the errors already in the system.

It would also seem that if the pixel shape was a problem, the screen 
would make vertical lines look bad, or at least different than 
horizontal lines. If this is not true, then I don't believe changing the 
filtering will improve anything. I suspect either the X setup is wrong, 
or the screen is doing very bad conversion of signal to brightness 
levels (for instance throwing away some of the bits of one channel).

Try taking a screen shot from one of the working displays and putting it 
on this one.