issue_comments

Working as intended according to the maintainer. Apologies for not closing this sooner.

In order to maintain a list of currently relevant issues, we mark issues as stale after a period of inactivity If this issue remains relevant, please comment here; otherwise it will be marked as closed automatically

Thanks. I was mistaken.

The "infinite tree" was causing some of the tests back at #1147 to do weird things (pytest seems not to handle infinite recursion responsibly). But it makes sense

@MaximilianR Let me help clarify:

• Currently, DataArray does not have a .variables attributes.
• The state of a DataArray is stored in the form of variables in ._variable (an xarray.Variable) and ._coords (of type OrderedDict[Any, xarray.Variable]).
• From a user facing API, you can get access to these variables via .variable and .coords.variables (I don't know if the later is clearly documented, but if not it should be).

So the "infinite tree of DataArrrays" only applies if you are accessing them directly via .coords, and under the covers data is represented in a pretty sane way.

Tangentially connected this - should a DataArray always have a .variables attribute? When would / wouldn't it?

Currently, it may or may not. And its coords may or may not be DataArrays, which makes the tree of objects seemingly complicated (maybe it needs to be?). But it means we're special-casing some of the checks in https://github.com/pydata/xarray/pull/1147

python (Pdb++) a <xarray.DataArray 'xy' (x: 3)> array([-0.04729734, 0.08287289, 0.13965946]) (Pdb++) a.variables *** AttributeError: 'DataArray' object has no attribute 'variables' (Pdb++) a.variable <xarray.Variable (x: 3)> array([-0.04729734, 0.08287289, 0.13965946])

sometime coordinates really only describe the data variable(s)

Yes. What I meant with my yesterday's late night answer is that keeping the coordinates around could be most useful when using irregular coordinates...

So this definitely was intentional, according to the principle that coordinates are shared between everything in a Dataset/DataArray.

I agree that doesn't always make sense -- sometime coordinates really only describe the data variable(s), so keeping them around on coordinates merely because they contain the appropriate dimensions is inappropriate.

I think it's fine. Coordinates sometimes need to be defined by paired coords too:

python import xarray as xr import numpy as np lon, lat = np.meshgrid(np.linspace(-20, 20, 5), np.linspace(0, 30, 4)) da = xr.DataArray(np.arange(20).reshape(4, 5), dims=['y', 'x'], coords = {'lat': (('y', 'x'), lat), 'lon': (('y', 'x'), lon)}) da.isel(x=2)['lat'] <xarray.DataArray 'lat' (y: 4)> array([ 0., 10., 20., 30.]) Coordinates: lon (y) float64 0.0 0.0 0.0 0.0 lat (y) float64 0.0 10.0 20.0 30.0 * y (y) int64 0 1 2 3 x int64 2