FillSingleCellPits creates problematic (NODATA + 1) values on raster edges

[geotom]

Problem

We observed that the FillSingleCellPits method algorithm is flawed, when the kernel window works along raster edges.
The result is an elevation raster, which has line artefacts on the raster edge (where a flattened lake creates same values).

Some visual explanations

Flattened lake with same elevation (370m)

Line artefacts caused by FillSingleCellPits

Code

The problematic part in the code is happening in line 222 (fill_pits.rs)

for n in 0..8 {
    zn = input.get_value(row + dy[n], col + dx[n]);
    if zn < min_zn {
        min_zn = zn;
    }
    if zn != nodata && zn < z {
        flag = false;
        break;
    }
}

Solution

A new min value for the observed raster window should only be set, when the pixel value is not NODATA (in this case outside of the raster bounds)

line 222: if zn != nodata && zn < min_zn {