Is there something inherently different about how Disk and Gaussian styles of deconvolving filters handle boundary pixels? With Disk (or Polygon), I get black rings on the edges. It’s especially noticeable if you look at a solid alpha input vs output. If I switch to Gaussian (or Airy Disk), the ringing is gone.
I’m getting better results out of Disk than anything else, but the edges are all wonky, so I’m having to crop things around, and even then it’s not quite right at the edges, though the middle looks great.
I’m not sure if this is a problem I had earlier, where I messed up the boundary handling in Frequency Blur. Does the boundary change when you switch between duplicate and black edge handling? If not, then the last release I did had the bad edge handling.
If they are not the same, then it is probably a more fundamental issue. (or a different bug)
Is this happening in images with a solid alpha? With a solid alpha input the output image should also have a solid alpha with duplicate edge handling.
If it is with particle renders then I assume the problem is the duplicate edge handling. This will stretch out those particles as soon as they touch the edge of the image. The effect is less visible with the gaussian because the weights fall off further from the centre of the blur. The disc blur is weighted constantly, so the duplicate pixels are probably causing the problem. If this sounds like the issue I can probably come up with a more clever edge handling methods.
If you have solid alpha, you do end up with darker values in the alpha. Seems odd. But if your alpha is black, you get the same problem in the color, so the alpha seems to show the problem, but not cause it.
Just feeding a default BG to the FqBl shows funniness in the alpha. Duplicate on Gaussian looks great, but nothing looks good with Disk.