[BUG] Default boundary condition is periodic even when specified otherwise

[dhruvbalwada]

The issue: Even when xgcm has been told that a coordinate is not periodic, it still chooses that as the default behavior if no explicit boundary information is passed - instead of throwing an error or warning.

import xarray as xr 
import xgcm
import numpy as np

dx = 0.5
L = 20
x_left = np.arange(0,L-dx, dx)
x = np.arange(dx, L, dx)

slope = 2.
y = slope*x

da_y = xr.DataArray(y, dims=['xc'], coords={'xc':x})
ds_y = xr.Dataset(coords={'xc':x, 'xg':x_left})
ds_y['y'] = da_y

grid = xgcm.Grid(ds_y, 
                 coords={"X": {"center": "xc", "left": "xg"}},
                periodic=None)

dy_default = grid.diff(ds_y.y, 'X')
dy = grid.diff(ds_y.y, 'X', boundary='extend')

In this example we told grid that X is not a periodic direction. When we print grid, it says that X is not periodic. However, the dy_default (where we did not pass a boundary condition) still assumes X is periodic without alerting the user.

[dhruvbalwada]

Might be related to #509

[hdrake]

Confirming I'm running into this same issue.

[jbusecke]

Is this fixed by using the boundary= input syntax? In that case deprecation of periodic might be the solution here.

[hdrake]

Is this fixed by using the boundary= input syntax? In that case deprecation of periodic might be the solution here.

Yep–see #606 (comment).