As much as I would have to slow down the process, I'd like to suggest another change to radial-gradient. Right now, if you have a position change so that the gradient is not centered, the gradient gets clipped when you have farthest-side or *-corner, but for closest side the gradient just gets smaller as you get closer to a side, and disappears altogether (except for the last stop) if you align the center with an edge.
What I think we should consider is that closest-side should not make the gradient smaller. The way this would work is that 'closest-side' would not consider the side(s) that you've offset the gradient towards when determining which is closer, and the gradient would get clipped instead.
I think this would be more useful design-wise, especially when aligning the center with a side or corner, and is probably more in the spirit of what closest-side was intended for. If you wanted the old behavior you could still achieve it through the gradient sizing, e.g. 'radial-gradient(-12.5% -25%, 75% 50%, white, black)' (I think). But I think the clipping behavior would be more useful more often.
Below (or attached image, if you don't see it embedded), I show a sketch of this idea (for 'circle' shape), with current spec behavior on the left and proposed on the right.
