```Andrew Fedoniouk wrote:
> Zack Weinberg wrote:
>> Andrew Fedoniouk <news@terrainformatica.com> wrote:
>>>> I don't think example 9 is rendered right -- I would expect the
>>>> color gradient to extend as far as "case 3" specifies.  But when
>>>> the inner corner is not sharp (cases 1-8), cases 1-3 are
>>>> indistinguishable, because the inner and outer curves start and end
>>>> on the same lines.  So you don't need to worry about that.
>>> but this : "color gradient to extend as far as "case 3" specifies"
>>> will break current spec. that says that transition is limited by
>>> quarter-ellipses of rounded-corners (as far as I understand wording
>>> there).
>>
>> To reiterate: the spec speaks only of the INNER curve.  Your cases 1
>> and 2 are bounding the transition based on the OUTER curve.  That's
>> clearly wrong.
>
> And now I have lost any idea what all this means. Anyway...
>
> Here is what I've got so far:
>
> There are three possible cases of the area where gradient transition may
> happen. They are presented on this figure:
>
>
> Here is a screenshot where the renderer (latest Sciter build) uses
> variant (B) of transition areas:
>
> http://www.terrainformatica.com/w3/round-corners-sciter-b.png
>
> Here is a screenshot where the renderer uses variant (A):
>
> http://www.terrainformatica.com/w3/round-corners-sciter.png
>
> Variant (C) (gradient area is limited by rectangle-intersection of two
> borders) will not work as such rectangle can be outside of the curve
> completely - no area to do transition at all.

Thank you for the great illustrations Andrew, I understand better
what the argument is here.

The intention of the spec is to limit transitions to area B.
This is because the conjunction of different styles may need
the full area of B. If you're just joining two solid borders
and want to do a color transition, you wouldn't want to use
the full area of B, you'd want to draw lines through the border
from the ends of the inner curve to the ends of the outer curve.
In the case where the inner curve is a sharp corner, you draw
angled lines through the border from the point to the ends of
the curve on the outer edge. Such a transition is still contained
within the region dictated by the spec, but by fanning out from
the inner corner, it looks much more reasonable. If you can make
an illustration of that, I'll be happy to put it in the spec as
an example. It's a little beyond my (very limited) graphical
abilities.

Zack, I've added a modified version of your image to the spec: