Current best way to build dataset or arrays with varying coords

[EtienneCmb]

Hi,

I'm working with relatively large files, all loaded as DataArray. All of them have the same dimension names, however, the coordinates might be different. I want to select sub-data across all the files e.g using .sel. My first idea was to use a Dataset composed of all of my files. Here's a minimal example :

da_1 = xr.DataArray([1, 2, 3], dims='x', coords=([0, 1, 2],))
da_2 = xr.DataArray([3, 4, 5], dims='x', coords=([2, 3, 4],))

xr.Dataset({'x1': da_1, 'x2': da_2})

Missing values are filled with nan, I suppose that it's also going to increase quite drastically the size in memory when considering much larger files?

In perfect world, I would like to have a container such as the Dataset, which doesn't create internal copies (or extend it with nan) and still allow me to sub-select data across all arrays. What would be the best strategy using xarray?

Thank you in advance,

[keewis]

I'm not quite sure but my guess would be to use sparse or to use different dimension names (see also #3984). Selecting would be a bit more difficult for the latter, though: you would have to translate x to x1 and x2 (assuming the new dimension names are called that), with identical indexers.