I'm investigating converting existing plug-ins from 2.8 to use GEGL.

It appears that one would implement the calculations in GEGL, leave the interface and parameter input code pretty much as is, and then pass the parameters to the GEGL code from GIMP plug-in code (making two plug-ins, actually). Am I missing something?

Handling previews effectively seems like a task better left to GEGL or a generalized GIMP callback than being done in each plug-in.

Algorithm question, using despeckle.c as a starting point:

The "median" algorithm uses luminance of the pixels around the point of interest, creates a histogram of 256 levels of luminance, throws those below and above a user specified threshold out, finds the median level of luminance, and then chooses randomly from the pixels that have that level of luminance. This would seem to choose a random color with an appropriate luminance from the region. Seems to me this makes the algorithm useful for gray scale images that are not black and white. Only

The literature is full of other despeckle algorithms. The code itself mentions 4, including "adaptive" but appears to implement none of them. Handling the following all seem worthy at some point 1) black/white images, 2) maintaining detail and contrast while smoothing larger areas of more closely matching pixels, 3) separating luminance or value from color.

Questions:

Is renaming for example despeckle to despleckle-median easy to do in the future or is it somehow easier during the initial conversion process? My own preference would be to use a new more specific name in GEGL and keep the old despeckle name for now in GIMP, though adding a method parameter could also work for the GEGL code.

If I do take on a conversion, is it acceptable if I include comments explaining the algorithm, or will they just be deleted?

If I include comments about when the algorithm seems useful, will those just be deleted?