@Simon (View Post)
Actually, it looks like the r, g, and b values just decrease by 1 every time no matter what, so it's not a rounding problem. If for example r is 1, or it's 255, when you sample it you'll get 0 and 254, so adding 1 to each will fix it.

The problem is the ">> 8", used instead of "/ 255" when blending colours in render_parts (and other places, but render_parts is the function causing this particular problem). ">> 8" is equivalent to dividing by 256.

The maximum alpha value (cola, deca, firea) is 255.
255 >> 8 = 255 / 256 which is less than 1, rounding error occurs, and the resulting RGB values are slightly too small.

Changing ">> 8" to "/ 255" does fix the issue, but I haven't tested what the impact on performance is.

Even, if there is rounding error, I found out that it's always the same, so you can just add 1 to each color value as long as its not 0, and that will mostly fix it except for when r, g, or b is 1. There's no need to change all those if this way is easier.