Attributes are dropped after clip even if keep_attrs is True

[mraspaud]

MCVE Code Sample

import xarray as xr
import numpy as np

arr = xr.DataArray(np.ones((5, 5)), attrs={'units': 'K'})
xr.set_options(keep_attrs=True)
arr
# <xarray.DataArray (dim_0: 5, dim_1: 5)>
# array([[1., 1., 1., 1., 1.],
#        [1., 1., 1., 1., 1.],
#        [1., 1., 1., 1., 1.],
#        [1., 1., 1., 1., 1.],
#        [1., 1., 1., 1., 1.]])
# Dimensions without coordinates: dim_0, dim_1
# Attributes:
#     units:    K

arr.clip(0, 1)
# <xarray.DataArray (dim_0: 5, dim_1: 5)>
# array([[1., 1., 1., 1., 1.],
#        [1., 1., 1., 1., 1.],
#        [1., 1., 1., 1., 1.],
#        [1., 1., 1., 1., 1.],
#        [1., 1., 1., 1., 1.]])
# Dimensions without coordinates: dim_0, dim_1

Expected Output

I would expect the attributes to be kept

Problem Description

keep_attrs set to True doesn't seem to be respected with the DataArray.clip method.

Output of xr.show_versions()

INSTALLED VERSIONS ------------------ commit: None 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: 3.10.0-1062.1.1.el7.x86_64 machine: x86_64 processor: x86_64 byteorder: little LC_ALL: None LANG: en_GB.UTF-8 LOCALE: en_GB.UTF-8 libhdf5: 1.10.5 libnetcdf: 4.6.2

xarray: 0.14.0
pandas: 0.25.1
numpy: 1.17.0
scipy: 1.3.0
netCDF4: 1.5.1.2
pydap: None
h5netcdf: 0.7.4
h5py: 2.10.0
Nio: None
zarr: 2.3.2
cftime: 1.0.3.4
nc_time_axis: None
PseudoNetCDF: None
rasterio: 1.0.28
cfgrib: None
iris: None
bottleneck: None
dask: 2.6.0
distributed: 2.6.0
matplotlib: 3.1.1
cartopy: 0.17.0
seaborn: None
numbagg: None
setuptools: 41.4.0
pip: 19.3
conda: None
pytest: 5.0.1
IPython: 7.8.0
sphinx: 2.2.0

[dcherian]

Thanks @mraspaud

clip and other injected functions need to be updated to obey keep_attrs (also see #3348)

xarray/xarray/core/ops.py

Lines 321 to 349 in 72be873

	def inject_all_ops_and_reduce_methods(cls, priority=50, array_only=True):
	# prioritize our operations over those of numpy.ndarray (priority=1)
	# and numpy.matrix (priority=10)
	cls.__array_priority__ = priority
	# patch in standard special operations
	for name in UNARY_OPS:
	setattr(cls, op_str(name), cls._unary_op(get_op(name)))
	inject_binary_ops(cls, inplace=True)
	# patch in numpy/pandas methods
	for name in NUMPY_UNARY_METHODS:
	setattr(cls, name, cls._unary_op(_method_wrapper(name)))
	for name in PANDAS_UNARY_FUNCTIONS:
	f = _func_slash_method_wrapper(getattr(duck_array_ops, name), name=name)
	setattr(cls, name, cls._unary_op(f))
	f = _func_slash_method_wrapper(duck_array_ops.around, name="round")
	setattr(cls, "round", cls._unary_op(f))
	if array_only:
	# these methods don't return arrays of the same shape as the input, so
	# don't try to patch these in for Dataset objects
	for name in NUMPY_SAME_METHODS:
	setattr(cls, name, _values_method_wrapper(name))
	inject_reduce_methods(cls)
	inject_cum_methods(cls)

Are you up for sending in a PR?

[mraspaud]

Sure! What do you reckon the default should be ? False for backwards compatibility ?

[mraspaud]

ooh, maybe I answered too fast. I hadn't really looked at the code yet... but let's see:
if I understand correctly, in the function you cited, the functions get wrapped and include in the current cls. So I would need to fix the wrapper itself (_func_slash_method_wrapper) so that the attrs get copied and applied to the resulting array (if keep_attrs is True that is). However, do we have a guarantee that the wrapped functions are actually returning something that can have a .attrs ?

[dcherian]

I don't think so. You'll have to check which functions are used to inject methods on groupby, resample and rolling objects (at least)

[mraspaud]

Ok, then I probably won't have the time to dig into this for now, sorry. If someone else with better knowledge of the code can work on this it would probably be for the best.