Problem when slicing a NEMO 2d grid objects defined with dask chunks.

[lesommer]

a slice of a grid object defined with chunks appears to be disfunctionnal.

eg.

from oocgcm.oceanmodels.nemo import grids
_grd = grids.nemo_2d_grid(...,chunks=xr_chunks)
grd = _grd[500:1500,500:1500]

this works :

gssh = grd.horizontal_gradient(buoyancy)

but this raises an exception

ghb = grd.norm_of_vectorfield(grd.horizontal_gradient(buoyancy))

[lesommer]

I have investigated this issue over the last week.

The simplest version of the issue with grid objects can be illustrated when trying to access any array that involved arithmetic operations on dataarrays. one of them is :

grd['cell_area_at_t_location']

It appears that this behavior is related to the default use of lock=False when opening NEMO files.

A short illustration can be found in this notebook

[lesommer]

I have added a unit test for testing the validity of grd['cell_area_at_t_location'] when grd is a sliced grid defined with chunks : see this commit

I have changed the default option for creating a xarray.Dataarray commit. We no longer use lock=False as default.

The user can still specify this options as an optional argument.

[lesommer]

The test runs successfully on my laptop but it fails on Travis ci.

[lesommer]

• test passes with netcdf4 backend: see
• test fails with scipy backend : see