bug: get_structured_faceflows outputting wrong flows about flopy

[vcantarella]

Looking at the source code of the function I think the problem is in this snippet:

    # fill flow terms
    vmult = [-1.0, -1.0, -1.0]
    flows = [frf, fff, flf]
    for n in range(grb.nodes):
        i0, i1 = ia[n] + 1, ia[n + 1]
        for j in range(i0, i1):
            jcol = ja[j]
            if jcol > n:
                if jcol == n + 1:
                    ipos = 0
                elif jcol == n + grb.ncol: # <======HERE
                    ipos = 1
                else:
                    ipos = 2
                flows[ipos][n] = vmult[ipos] * flowja[j]
    return frf.reshape(shape), fff.reshape(shape), flf.reshape(shape)

It seems like it already assumes that jcol == n+grb.ncol means it is in the next row, but in this case it should jump straight to the next layer.

[langevin-usgs]

Thanks for picking this up @w-bonelli. I looked at this a bit and it should be easy to fix, but it is trickier than it seems. Curious to see what you come up with.

[jdhughes-usgs]

@w-bonelli we have run into edge case issues with function before. It seems we should have better testing for this. Probably need cases (1, nrow, 1), (1, 1, ncol), (1, nrow, ncol), (2, nrow, 1), (2, 1, ncol), (2, nrow, ncol), (3, nrow, 1), (3, 1, ncol), and (3, nrow, ncol) to cover all bases.

[wpbonelli]

Changes were needed to the way we determine which face each flow belongs to, while iterating through the intercell flow matrix to build right/front/lower face arrays.

In the given model, flows with m - n == ncol go to lower face as noted by @vcantarella. Previously the logic was

• 0 (right) if m - n == 1
• 1 (front) if m - n == ncol
• 2 (lower) otherwise

With #1968 it becomes

• 0 (right) if m - n == 1
• 2 (lower) if m - n == ncpl
• 1 (front) otherwise

But this is not generic, it only works if the grid is extended (has multiple cells) in all 3 dimensions. Special treatment is added for "degenerate" cases among those listed above by @jdhughes-usgs