[Gimp-developer] Background color property for GIMP images

[saulgoode at flashingtwelve.brickfilms.com]

Perhaps I am misunderstanding this proposal, but the ramifications  
seem to be more confusing than the present method. And while I realize  
that GIMP does not make any guarantees about retaining the colors of  
transparent pixels, its current behavior is quite useful for editing  
files destined for applications which employ the alpha channel in  
unconventional ways. It also offers a few other atypical benefits, but  
mainly it is consistent and easy to comprehend what is happening.

Some questions:

Currently, erasing on a layer having an alpha channel only affects the  
alpha channel, the color values remain the same. If a special case  
were to be created such that transparent erasures (alpha < 1/256)  
result in changes to color values, what then is to happen when color  
depths are increased such that the minimum non-zero alpha becomes  
1/65556 (16-bit per color), or even lower (32-bit or floating point)?  
Erasures of an 8-bit PNG image  using these higher color depths will  
reveal not the original image, but instead some unassociated color and  
this might cause some problematic "fringing".

If erasing is to be changed such that color channels are painted, is  
this to be offered as a tool option which can be disabled? And if  
erasures are done with an tool opacity less than 100%, would the  
option be provided to decide whether the color channels are painted at  
100% background or instead blend with the underlying color at the  
tool's opacity level? or would using the tool's transparency level  
make more sense?

If the eraser's color is to blend with the existing color channels,  
should blend modes be enabled for the eraser tool?

If a PNG is loaded as a layer, should the "image background color" be  
updated to the PNG file's background color? or should it remain what  
it was originally? If a JPEG is loaded as a layer, should the "image  
background color" be set to white? Maybe every layer should be  
assigned its own background color?

Should gradients be using the "image background color" or the "second  
color in a color slot history"?

I don't mean to stomp on the idea of an image background color  
completely but I do think some of the deeper consequences need to be  
addressed in detail, and the options being presented to, and removed  
from, the user weighed.