I'm testing Gimp 2.5.4 and it's amazing. I know it's late for a feature request, but I think it's worth to discuss about the current behavior of the new feature present in the transform tools: the ability to set the transparency of the layer or selection being transformed.
I think it would make more sense to apply that transparency to the original layer instead of the transformed instance. In most cases, the original occludes the background, and what the user usually wants is to see the transformed object on the final context. This situation becomes particularly annoying when scaling down images. The larger original doesn't let the user see the real background, where the transformed image will end up. So this new transparency function would be much more helpful (in my oppinion) for the original, not for the "target".
Somebody suggested in the IRC channel that the original should simply dissapear, because the user wants to see the result of the transformation, not the original.
I think that there are lots of situations where having a reference of the original, un-transformed image would be very useful, but not that useful if having it doesn't let me see the background. So, what do you think about applying the transparency to the original insteado of applying it to the transformed image?

Gez.

P.s.: I'm having a problem with the windows position when compiz is active (it works fine with metacity). The position of the windows is remembered, but they open a tad lower than the original position. It causes that the lower part of the windows appear behind the gnome's bottom panel (I'm using Ubuntu 8.04.1, 64 bits).

(hiding the original layer during the operation is possible, but because of the simplicity of the preview rendering, the preview may look much different that you'd expect.)

Probably this should be discussed a little bit more. There's a particular situation where having an opaque original makes very hard to use a transform tool: when you paste a layer that is bigger than the image area and you have to scale it down to a desired size. You simply can't see the background because the opaque original is in front, so you can't apply the transformation with precision. The same applies when you have to rotate an element to match an angle of something that is behind the layer that you want to transform. I can think of a couple more of examples where this situation makes very hard to work.
There's a workaround, that is lowering the original layer opacity then transforming, but it's not very handy (you have to do that and then remember to raise the opacity after you apply each transformation). This makes working with transformations quite slow and tedious. If it's possible to directly hide the original, I'd prefer that option until the GEGL porting of the display code is ready, even knowing that the transformation proxy isn't very accurate. I'd like to know how other users feel about this.

Gez.