note on transparency

[enigma2.git] / doc / SKINS

diff --git a/doc/SKINS b/doc/SKINS
index 2db4c6f99cc9f9793285c9846191b3f30ccab283..6164cb1002d10d6f3070bc10975f056960c02485 100644 (file)
--- a/doc/SKINS
+++ b/doc/SKINS
@@ -70,3 +70,35 @@ if you're having less than 16 colors in your png file, pngquant generates a
 16color (4bit) PNG, which is not really what we want. Either add more colors 
 (evil workaround), fix enigma (good idea!) or patch pngquant to
 always create 8bit pngs.
 16color (4bit) PNG, which is not really what we want. Either add more colors 
 (evil workaround), fix enigma (good idea!) or patch pngquant to
 always create 8bit pngs.

+
+transparency:
+=============
+
+First, do not mix up transparency with alpha. I agree, both terms refer to
+the same thing, but in our case, it's more complicated. As long as we don't
+run on accelerated graphic, we need to differentiate between these two.
+
+Alpha, in our context, is the ratio between the video and the OSD. Alpha = 0
+means that OSD is completely visible, and no video, whereas Alpha = 1 (or
+0xFF, if you express it so) means that there is only video.
+
+Transparency, on the other hand, is OSD-only. When you have a background,
+say a color, and you want to blit a picture on it, you can mix the
+background with the picture. (*) This allows pixmap with a non-rectangular
+shape.
+
+Because of speed reasons, we don't support real alpha blending when not
+running with acceleration. Instead, we use alpha testing. This means that a
+special color is "transparent", whereas all other colors are not. This
+colors *HAS TO BE AT INDEX 0*. As you can't count on automatic palette
+merging (it's disabled when the target is 32bit, so you you're not
+restricted to 8bit in total, only 8bit per pixmap), you have to tell your
+gfx program to put the transparent color at index 0. This is what's normally
+done, but in case transparency doesn't work correctly, you might want to
+check this.
+
+
+(*) Note, that, technically, if we start with a alpha=1 background (i.e.
+only video), and alphablend on this, we generalized the problem enough that
+we can handle both cases the same. But currently, we have to differentiate
+between them, as we can't always alphablend.