On Fri, Feb 21, 2003 at 04:55:31PM -0800, Daniel Rogers wrote: > As for a scheme, would it be worthwhile to simply set a property (like > "in-place") of the ImageData object to true or false, and then insist > that every Op know how to deal with objects where the destination may > overlap with the source? I dont think you would put it in ImageData, cause each op has a different ImageData object for each image input and output it has. The Image value the ImageData object holds is what is being passed along. But I think it might be an Op (or subclass of Op) property. So that in the case where the op can operate in place (not blurs for example) when the op goes to create its output image buffer (in validate_outputs) it doesnt actually create an image buffer, but just makes sure that the appropriate image from the input ImageData is used in the output ImageData as well. > For binary ops, you could insist that only one Image be operated on in > place, or assume the left one is in-place if they both try to be. > However, if the code is overlap safe, you could take care of the > overlapping regions in the instantiation init code, simply setting > destination to the correct input. > Comp ops are good examples of ops that can happen in-place. A series of images comped onto a background could all get computed in the background image buffer. Calvin
