issue_comments: 122484583

I thought we were writing variables in order of appearance in datasets, but it looks like we weren't doing that. #480 should fix that issue, at least when using netCDF4-python (the default backend).

Once that patch is merged, you can reorder variables in datasets by indexing with a list of variable names, e.g.,

``` In [6]: ds = xray.Dataset(OrderedDict([('x', 10), ('a', -1), ('z', np.nan)]), {'c': 0})

In [7]: ds Out[7]: <xray.Dataset> Dimensions: () Coordinates: c int64 0 Data variables: x int64 10 a int64 -1 z float64 nan

In [9]: ds.to_netcdf('tmp.nc')

In [10]: ! ncdump -h tmp.nc netcdf tmp { variables: int64 x ; int64 a ; double z ; int64 c ;

// global attributes: :coordinates = "c" ; }

In [11]: ds[['z', 'c', 'x', 'a']].to_netcdf('tmp.nc')

In [12]: ! ncdump -h tmp.nc netcdf tmp { variables: double z ; int64 c ; int64 x ; int64 a ;

// global attributes: :coordinates = "c" ; } ```