Merge with join left or right does not what you'd expect as a Pandas user

Hello everyone,

First of all, thank you for this great library and the work that all contributers put into it!

I try to do, what some call, a table join in xarray. But xr.merge with parameter join='left' or join='right' does not work as expected, at least not as Pandas' users would.

import numpy as np
import xarray as xr
y = xr.DataArray(np.arange(5, 10), [('x', np.arange(0,5))], name='y')  # create y(x)
z = xr.DataArray(np.arange(15,20), [('y', np.arange(5, 10))], name='z')  # create z(y)
expected_result = xr.DataArray(np.arange(15,20), [('x', np.arange(0,5))], name='z')  # This is what I want: z(x)

Try to do left join, to get z(x):

result = xr.merge([y, z], join='left')  # this fails with MergeError

MergeError: unable to determine if these variables should be coordinates or not in the merged result: {'y'}

Try workaroud according to error message with xr.Dataset and and set_coords:

y_ds = xr.Dataset({'y': y}).set_coords('y')
result = xr.merge([y_ds, z], join='left')  # does not error, but gives z(y)

The workaround does not error but gives z(y) with unreferenced additional x coordinate rather than z(x), thus not a true left join.

I suppose that this is a use case with quite heavy use, so a fix would be very appreciated.

Is there a true left join workaround (at best without falling back to Pandas) that I have overseen in the documentation?

Environment:

Reproduce the reported xr.merge examples with join='left' and join='right', comparing the result with the stated Pandas-style expectation. Clarify the intended behavior when a data variable is also a dimension name, then establish tests showing the expected aligned result and ensuring the current MergeError and incorrect z(y) workaround are addressed.