I am putting this enhancement request here so I can point freedomcoders to it, to get some funding to program it.

Right now, only one GEGL sampler is Jacobian-adaptive, which means that only one sampler properly reduces as well as enlarge, and has the internal machinery to deal with general affine transformations and non-affine warps. This sampler is called "LoHalo", and no-one has never seen halo come out of its use, however if is pretty jaggy when enlarging, and it does not suppress moire quite enough when reducing, as well as being on the soft side in this situation.

I propose to remedy this situation as follows:

0) Rename the LoHalo sampler NoHalo. In practical terms, the current LoHalo truly is halo free. My naming it LoHalo was being a perfectionist: In theory, it can let a minuscule amount of halo through.

Recycling the same skeleton, add three samplers:

1) The new LoHalo, which will be sigmoidized EWA (Elliptical Weighted Averaging) with the Robidoux bicubic for upsampling, blended with non-sigmoidized EWA Robidoux for downsampling.
You can obtain this filter with recent ImageMagick with this command when enlarging:
The upsampling component:
convert {input} -colorspace RGB +sigmoidal-contrast 6.5 -distort Resize {size specification, for example, 200%} -sigmoidal-contrast 6.5 -colorspace sRGB {output}
The downsampling component:
convert {input} -colorspace RGB +sigmoidal-contrast 6.5 -distort Resize {size specification, for example, 50%} -sigmoidal-contrast 6.5 -colorspace sRGB {output}

I'll do this one first because I have all the formulas figured out.

2) The new LoJaggy, which will be EWA LanczosSharp (the new version, discussed here: http://www.imagemagick.org/Usage/filter/nicolas/, which is almost identical to the one named LanczosSharp in ImageMagick, also used by the MadVR video renderer) all around (meaning upsampling and downsampling and everything in between).

With ImageMagick:
convert {input} -colorspace RGB -filter Lanczos -define filter:blur=.9891028367558475 -distort Resize {size specification} -colorspace sRGB {output}

3) The new LoBlur, which will be standard tensor Lanczos 3 for upsampling, blended with EWA Lanczos Radius 3 for downsampling.

With ImageMagick:
The upsampling component:
convert {input} -colorspace RGB -filter Lanczos -resize {size spec.} -colorspace sRGB {output}
The downsampling component (with very very recent ImageMagick; otherwise see http://www.imagemagick.org/Usage/filter/nicolas/):
conver {input} -colorspace RGB -filter LanczosRadius -distort Resize {size spec.} -colorspace sRGB {output}

Part of the fun of doing this is actually using recent research work on fast polynomial approximations of complex and expensive functions like the kernel of Lanczos 3 and Jinc-windowed Jinc 3, which hopefully makes the use of Look Up Tables pointless. 2) and 3) require that I run fairly careful computations with a modified version of the Boost C++ Minimax code to get cheap and accurate approximations of the key functions. ImageMagick contains outdated (and not quite suitable for GEGL) versions of such approximations. I'll make sure that other projects that could benefit from this work will get the new formulas as well.