issue_comments: 345394914

I'm rather a numpy-xarray user than a dask-xarray user (since most often my data fits in memory), but I wouldn't mind at all having to install dask as a requirement!

Potentially chunks=False in open_dataset could indicate that you're OK loading everything into memory with NumPy. We would then have to choose between making the default use Dask or NumPy.

Maybe like other users who are used to lazy loading, I'm a bit more concerned by this. I find it so handy to be able to load a medium-sized file instantly, quickly inspect its content, and then work with only a small subset of the variables / data, all of this without worrying about chunks.

Assuming that numpy-loading is the default, new xarray users coming from netcdf4-python and who don't know much about dask might find xarray a very inefficient tool when trying to first import a medium-sized netcdf file.

If choosing chunks=False as default, I can also imagine often forgetting to set it to True when loading a file that is a bit too big to load in memory... That would be annoying.

By saying "making the default use Dask", do you mean that data from a file will be "loaded" as dask arrays by default? If this is the case, new xarray users which are probably not familiar with dask (at least less likely than they are familiar with numpy) will have to learn 1-2 concepts from dask before using xarray. This might not be a big deal, though.

In summary, I'm also really not opposed to use dask to replace all the current lazy-loading machinery, but ideally it should be as transparent as possible with respect to the current "user experience".