When I try to use some different coordinates as the index of a dim, I notice the new API on v0.9 provided by DataArray.set_index: ```python

import numpy as np import xarray as xr arr = xr.DataArray(np.r_[:4].reshape(4,4),dims=('t','x')) arr['x_str'] = "x",["a","b","c","d"] arr["x_num"] = "x",[1,2,3,4]; arr <xarray.DataArray (t: 4, x: 4)> array([[ 0, 1, 2, 3], [ 4, 5, 6, 7], [ 8, 9, 10, 11], [12, 13, 14, 15]]) Coordinates: x_str (x) <U1 'a' 'b' 'c' 'd' x_num (x) int64 1 2 3 4 Dimensions without coordinates: t, x arr.set_index(x='x_str'); arr <xarray.DataArray (t: 4, x: 4)> array([[ 0, 1, 2, 3], [ 4, 5, 6, 7], [ 8, 9, 10, 11], [12, 13, 14, 15]]) Coordinates: x_num (x) int64 1 2 3 4 * x (x) object 'a' 'b' 'c' 'd' Dimensions without coordinates: t

... some code with "arr[{'x':[... some str index ...]}]" ...

That's really convenient, what a nice API. But when I want to switch to another coordinate, I found that I can not recovery my arr to the version before I using :python arr=arr.reset_index('x'); arr # why the croodinate to used as index now lose its name "x_str"? <xarray.DataArray (t: 4, x: 4)> array([[ 0, 1, 2, 3], [ 4, 5, 6, 7], [ 8, 9, 10, 11], [12, 13, 14, 15]]) Coordinates: x_num (x) int64 1 2 3 4 x_ (x) object 'a' 'b' 'c' 'd' Dimensions without coordinates: t, x arr=arr.set_index(x="x_num");arr # anyway, continue going to the code use coordinate "x_num" as index

... some code with "arr[{'x':[... some int index ...]}]" ...

arr=arr.reset_index('x');arr # now I need "x_str" coordinate as index, here we go <xarray.DataArray (t: 4, x: 4)> array([[ 0, 1, 2, 3], [ 4, 5, 6, 7], [ 8, 9, 10, 11], [12, 13, 14, 15]]) Coordinates: x_ (x) int64 1 2 3 4 Dimensions without coordinates: t, x

NOOP!!! the "x_int" coordinate COVER the "x_str" coordinate, I can't access the later any more :(

To solve this problem, I get following ways: - Add new function to do this. - may be most directly way. - add new function to implement and new API to design. - stop the DataArray.set_index to change coordinate name - The benefit is that without new function, its the most convenient one. - But it may change the calling-promise of DataArray.set_index, so it may be useless. - Here are some example to show how it can be used: python arr = arr.set_index({'x':'x_str'})

... some code with "arr[{'x':[... some str index ...]}]" ...

arr = arr.set_index({'x':'x_int'})

... some code with "arr[{'x':[... some int index ...]}]" ...

arr = arr.set_index({'x':'x_str'}) ... - store the coordinate name in DataArray when calling "DataArray.set_index" and recovery them when calling "DataArray.reset_index" - Just a bit more complex than previews one. - But it need to add redundant data inner DataArray, useless too. - Example: python arr = arr.set_index({'x':'x_str'})

... some code with "arr[{'x':[... some str index ...]}]" ...

arr = arr.reset_index('x').set_index({'x':'x_int'})

... some code with "arr[{'x':[... some int index ...]}]" ...

arr = arr.reset_index('x').set_index({'x':'x_str'}) ... - let DataArray.reset_index support Mapping as names parameters, while use the keys as dims to reset indices and the value as the names of coordinates converted from those indices. - More complex. - Maybe the one cause least change, so I prefer it. - Example: python arr = arr.set_index({'x':'x_str'})

... some code with "arr[{'x':[... some str index ...]}]" ...

arr = arr.reset_index({'x':'x_str'}).set_index({'x':'x_int'})

... some code with "arr[{'x':[... some int index ...]}]" ...

arr = arr.reset_index({'x':'x_int'}).set_index({'x':'x_str'}) ... ```