are the corners on a rounded rectangle circular?

I think rects must have equal roundness always. If you want
it unequal, just convert to curves. The scaling display must
therefore be fixed.

This may be difficult to fix because currently, what you see
during scaling is made by applying a transform matrix to the
object (temporarily, without writing to repr). Therefore it
is scaled as a whole. For the scaling display to be correct,
seltrans must actually write transform into the repr and
then regenerate object from that repr on each mouse move,
which would be slow and require a serious change in how
seltrans works. I think this will need to be done
eventually, but probably not now.

It actually is worse than just being messed up while
scaling. create a rect, whose width is smaller than it's
height. You are able to move the "roundness"-node until the
curves meet on the long side. This leaves you with an elipse
and not a rect with rounded corners. What should happen, is
that you can't move the "roundness"-node any further than
until the curves meet on the _shorter_ side not the _left_
side, no matter if the rectangle is landscape or portrait.

dcberg: I see no reason at all to introduce this limitation.
I may very well need "rects" that are almost like ellipses,
with one of the radii longer than the perpendicular side.
Why remove some capability that may be useful to someone, if
it fits well into the overall picture and does not break
anything? The current rects are a strict superset of
"normal" rounded rects and that is good.

Superceded by the more general bug 969406.