Recomputing optimization for image compare

[pierrelabreche]

When adding or moving an image compare transformation, for which the output image is set to the input image, downstream branch recomputations should not be triggered.

[jsachs]

Not all of the compare options pass the input image through unchanged, even if you are comparing it to itself.

[pierrelabreche]

You are right, what I wrote is not what I meant.
I should have written:

When adding or moving an image compare transformation, whenever the output image is set to the input image, the downstream branch should not be recomputed.

[jsachs]

This is an example of a much larger set of optimizations that could theoretically be performed by checking if the output image is always identical to the input image. For example, a brightness curve that does not change the input image. While I do in most cases detect these cases and simply copy the input to the output, I have so far chosen not to implement these optimizations as the code that handles propagation of changes is already quite complicated and every time I make changes to it the program becomes unstable for a few weeks or months as special cases emerge that I did not test for.