|
Okay, I've been reading the source code, particularly the variations, and am kind of upset.
Let me note that I'm okay with variations having parameters. It's basically kind of like an infinite number of variations in one. I'm also kind of okay with variations that use the transformation matrix. It's kind of a poor man's parameter, but I can deal with it. However, I'm not okay with variations having state, and I am definitely not okay with variations using the weight value in horrid ways or ignoring it completely. The two biggest offenders are the radial blur and secant variations. Why does radial blur need to save random numbers? Not only does this make it not thread-safe, but it also makes the result less random. Also, it ignores the proper use of weight, which is to produce a *weighted average*. Secant is an egregious offender. As the weight of that transform approaches zero, the amount that it adds to the y coordinate approaches 1. This should not happen. >_< Arch, rays, blade, and twintarian are also terrible offenders. They should not be taking the sine of the weight. But at least they go to zero as the weight does. Ideally, a variation should not know its weight at all. So why have these transformations been added? I'm thinking of changing those transformations so that they regard weight, and have separate parameters for what weight is being used for in them currently. It might break backwards compatibility, but it makes interpolation work in a sane way again. And the state for radial blur should be removed, in favor of simply using four random numbers. Unless there is some reason for rndG to be within .5 of the previous rndG.
|
Messages