Distance Transform: Slightly incorrect handling of image edges
Within gegl:distance-transform, there is a small issue stemming from how the filter handles the left and right image edges, where the output can end up with a lower distance (i.e. darker) than it should have. This can be most clearly seen when running the op on a blank white square, producing an erroneously asymmetrical output.
For example, the euclidean output from a 32x32 white square:
I'm pretty sure this is caused during binary_dt_2nd_pass(), where the left and right extremities of each dest row is maxed to 1 before the second pass of the algorithm is run. Since these values still represent a vertical distance from the nearest below-threshold pixel as set in binary_dt_1st_pass(), adjacent pixels are then essentially calculated as if there were a black pixel diagonally adjacent to them, rather than two pixels away horizontally (representing the abyss) as is the desired effect, and so on.