Reproducibility issue for ROMS-Tools Grid when RiverForcing or Nesting is used

[NoraLoose]

I want to surface a potential reproducibility issue in C-Star due to the way ROMS / ROMS-Tools handles river forcing and nesting.

A grid is created and saved via step 0:

grid = roms_tools.Grid(**kwargs)
grid.save(path)  # saves grid variables to NetCDF
grid.to_yaml(yaml_path)  # saves kwargs to YAML

In two instances (both of which are optional), ROMS-Tools modifies the grid file after the fact (because a modified grid file is what ROMS expects):

1. When a RiverForcing class is created, a new variable river_flux gets added to the grid file, see here.
2. When a Nesting class is created via

nesting = roms_tools.Nesting(parent_grid=parent_grid, child_grid=child_grid)
nesting.save(path, child_grid_path)  # saves nesting information and updated child grid to NetCDF

the child grid topography and mask get modified along the boundaries so that they are compatible with the parent grid, see here.

In other words, there is a difference in the grid files depending on whether (optional) steps 1 and 2 are executed or not. Step 1 does not modify the existing grid variables (so it's not too bad), but step 2 modifies the existing grid variables.

Just reporting this here because this needs further discussion.