Is there a way to reorder Dataset dimensions to match the DataArrays?

[ahuang11]

import xarray as xr
ds = xr.tutorial.open_dataset('air_temperature')
ds['airx2'] = ds['air'] * 2
ds = ds.transpose('time', 'lon', 'lat')
print(ds.dims)

yields
Frozen(SortedKeysDict({'lat': 25, 'time': 2920, 'lon': 53}))

Since it's labeled SortedKeysDict, I expected the same ordering as shown, but when I call
list(ds.dims)

I get:
['lat', 'lon', 'time']

which is different from the provided, transposed order
('time', 'lon', 'lat')

If I subset a DataArray, it's correct:
print(ds['air'].dims)
('time', 'lon', 'lat')

[keewis]

sorry for the late reply, @ahuang11.

To me the the name SortedKeysDict implies that it is sorted, not ordered (the difference being that – similar to a OrderedDict – "ordered" follows the order of insertion while "sorted" is sorted somehow, in this case alphabetically).

The difficulty with Dataset dimensions is that the order has no actual meaning: consider a Dataset with three variables, one with the dimensions ("x", "y"), the second ("x", "t") and the other ("z", "t"). Which order should we use for those?

In order to completely avoid that somewhat complex decision, the Dataset dimensions are always sorted alphabetically, while for DataArray (or the semi-internal Variable) the order does have an actual meaning so those objects keep that. I know that for single variable Dataset objects that is confusing, but for multi-variable datasets this is a lot easier to maintain (and to understand).