Anisotropic mesh refinement with Fortran

[tpg2114]

For various reasons, I am using AMReX built and linked in 3D, but I'd like to run simulations in "2D" or "1D" where the unused dimensions have blocking_factor cells in that direction and no more. I can do this if I don't adapt the grid, but with adaptation I'd like to specify a refinement ratio of {2,2,1} or {2,1,1}, respectively. It looks like this is supported in the C++ side of the code, but the FAmrCore aborts if the refinement ratios are not all equal on a given level.

Is there a technical or design constraint that prevents anisotropic refinement? I can see that the way to access and store the refinement ratios in Fortran would need to change because right now it is a 1D array of integers and to support different values, it would need to be either a 2D array (ndim,nlevel) or a derived type array-of-array setup. But I'm not sure if there is a more fundamental limitation that I am overlooking.

[WeiqunZhang]

There are no fundamental reasons for the limitation other than we have not tried. If you want to give this a try and it requires changes in AMReX, please feel free to submit PRs.