[Bf-committers] Bug Fix Sequencer Transform effect

[Peter Schlaile]

Hi,

> This fix just moves the problem to another case I'm guessing.
> What should be done is find the cases where testing the input
> buffer isn't sufficent and come up with a fix for those cases.

uhm, that case is simple: in sequencer, the decision of using float vs
no-float is based on _output_ not on input. (And yes, I don't want to
calculate always both of them or switch everything to float for speed
reasons...)

The problem will arise, if someone feeds an imbuf into that function, that
has rectf _and_ rect set. (That can happen, if we had to convert an
input imbuf on the fly from rect -> rectf.)

Since we only want to calculate _one_ output, the decision, which rect
should get calculated should be made on output.

Probably adding additional sanity checks for input won't hurt also,
though.

(Since I was pretty sure, that this was a real solution I already applied
the patch yesterday. If those sanity checks are indeed needed, I can add
them, too. But having the wrong rect-value set in output would have
resulted in a black screen with the current code, so someone would have
noticed... )

Sincerely,
Peter