Can you suggest new API that would implement this? Thanks.

CC @nigeltao

I would suggest changing all the functions with a similar signature:

Scale(dst Image, dr image.Rectangle, src image.Image, sr image.Rectangle, op Op, opts *Options)

Into something like:

Scale(dst Image, dr image.Rectangle, src image.Image, sr image.Rectangle[Number], op Op, opts *Options)

Or introduce an image.RectangleFloat to be used as type for the source rectangle of the image.
The idea is that during transformations, images have an intermediate 2d function representation. Although the output of a transformation is always an image (a set of sampled values in a 2d array), forcing the user to always sample on a grid that lies on integral numbers of the source domain seems to me restrictive.

@lromor Thanks, but we don't want to change any existing function, as that would break existing working programs that call that function. We can add new functions, but we can't change existing ones.

@ianlancetaylor , sorry for the fuss. I don't want to defend this proposal anymore. My conclusion is that the Transformer interface and relevant implementations are more than enough for my use case. One could make a case about the Scaler as well, but I'm not a fan of adding another function.