Features in Office
Committing "progress" for today, filters that use the FFT convolution worker like blur still do not work.
My copybook math says that convoluting channels premultiplied by alpha and then dividing them by the convoluted alpha channel will yield correct alpha-aware convolution.
I only managed to demonstrate that (to myself) with values ranging from 0 to 1.
I suspect my current failure is due to multiplying values between 0 and something >1. But I tried dividing by 255 and testing in 8 bit images and it still doesn't work.
Here I be frustrated.