xr.Dataset.map drops attrs of DataArray

[mathause]

MCVE Code Sample

import xarray as xr
import numpy as np

ds = xr.DataArray([1, 2], attrs=dict(tst="DataArray")).to_dataset(name="data")
ds.attrs["tst"] = "Dataset"
ds.map(np.mean, keep_attrs=True).data

returns

<xarray.DataArray 'data' ()>
array(1.5)

Expected Output

<xarray.DataArray 'data' ()>
array(1.5)
Attributes:
    tst:      DataArray

Problem Description

Applying xr.Dataset.map(..., keep_attrs=True) does not retain the attributes of the DataArrays. Should it?

In constrast ds.mean(keep_attrs=True) retains DataArray-level attrs.

EDIT: corrected example

Output of xr.show_versions()

INSTALLED VERSIONS

commit: bd4f048
python: 3.7.3 | packaged by conda-forge | (default, Jul 1 2019, 21:52:21)
[GCC 7.3.0]
python-bits: 64
OS: Linux
OS-release: 4.12.14-lp151.28.32-default
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_GB.UTF-8
LOCALE: en_US.UTF-8
libhdf5: 1.10.5
libnetcdf: 4.6.2

xarray: 0.14.0+117.gbd4f048b.dirty
pandas: 0.25.2
numpy: 1.17.3
scipy: 1.3.1
netCDF4: 1.5.1.2
pydap: installed
h5netcdf: 0.7.4
h5py: 2.10.0
Nio: 1.5.5
zarr: 2.3.2
cftime: 1.0.4.2
nc_time_axis: 1.2.0
PseudoNetCDF: installed
rasterio: 1.1.0
cfgrib: 0.9.7.2
iris: 2.2.0
bottleneck: 1.2.1
dask: 2.6.0
distributed: 2.6.0
matplotlib: 3.1.1
cartopy: 0.17.0
seaborn: 0.9.0
numbagg: installed
setuptools: 41.6.0.post20191029
pip: 19.3.1
conda: None
pytest: 5.2.2
IPython: 7.9.0
sphinx: None

[dcherian]

Looking at the code, keep_attrs controls whether the dataset attrs are kept, not the invidual variable attrs.

xarray/xarray/core/dataset.py

Lines 4233 to 4240 in 577d3a7

	variables = {
	k: maybe_wrap_array(v, func(v, *args, **kwargs))
	for k, v in self.data_vars.items()
	}
	if keep_attrs is None:
	keep_attrs = _get_keep_attrs(default=False)
	attrs = self.attrs if keep_attrs else None
	return type(self)(variables, attrs=attrs)

This may be fixable by threading keep_attrs down through maybe_wrap_array and then ultimately to Variable.__array_wrap__ when you can choose to keep attrs if keep_attrs is True.

[mathause]

Yes, it works for the Dataset-level attrs and the docstring also specifies that:

keep_attrs : bool, optional
    If True, the dataset's attributes (`attrs`) will be copied from
    the original object to the new one. If False, the new object will
    be returned without attributes.

So the question is more if this is the expected behaviour?

[mathause]

Sorry, read before you edited - I can try to add a PR if you agree to change this behavior.

[dcherian]

I think that it's sensible that keep_attrs controls whether individual variable attrs are kept after mapping a function on each variable.

This is a behaviour change though. I'm not sure how to deal with that.

cc @max-sixty

[max-sixty]

(forgive me, am on vacation)

I actually didn't know about the multiple levels of attrs. Given that does exist, I agree with @dcherian re keep_attrs impacting individual variable attrs too.

While it is a behavior change, I think it's fine: the current behavior is somewhere between a bug and an undocumented artefact. Unless others see a more material change here?

[TomNicholas]

When I wrote the (set_options) keep_attrs switch in then I intended it to function as a global switch, so ensuring that individual DataArray attrs are also propagated fits with that intention. I just apparently never thought of this case.