RecursionError when defining __slots__ on subclass using Containment

[BorjaEst]

What happened:
Inheritance of xarray.Dataset request the addition of __slots__:

/home/borja/miniconda3/envs/o3as/lib/python3.6/abc.py:133: FutureWarning: xarray subclass Model should explicitly define __slots__
  cls = super().__new__(mcls, name, bases, namespace, **kwargs)
/home/borja/Projects/o3skim/o3skim/sources.py:77: FutureWarning: Setting attribute 'dataset' on a 'Model' object. Explicitly define __slots__ to suppress this warning for legitimate custom attributes and raise an error when attempting variables assignments.
  self.dataset = ds

however when added, produces RecursionError.

What you expected to happen:
It should be possible to use Containment & Delegation on xarray classes.

Minimal Complete Verifiable Example:

import xarray

class MyDSet(xarray.Dataset):
    __slots__ = ['dataset',]
    def __init__(self, ds):
        self.dataset = ds

    def __getattr__(self, k):
        return getattr(self.dataset, k)

    def __setattr__(self, k, v):
        setattr(self.dataset, k, v)

Returns:

>>> reload(mytest)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/borja/miniconda3/envs/o3as/lib/python3.6/importlib/__init__.py", line 166, in reload
    _bootstrap._exec(spec, module)
  File "<frozen importlib._bootstrap>", line 618, in _exec
  File "<frozen importlib._bootstrap_external>", line 678, in exec_module
  File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed
  File "/home/borja/Projects/o3skim/o3skim/mytest.py", line 3, in <module>
    class BCorrect(xarray.Dataset):
  File "/home/borja/miniconda3/envs/o3as/lib/python3.6/typing.py", line 978, in __new__
    self = super().__new__(cls, name, bases, namespace, _root=True)
  File "/home/borja/miniconda3/envs/o3as/lib/python3.6/typing.py", line 137, in __new__
    return super().__new__(cls, name, bases, namespace)
  File "/home/borja/miniconda3/envs/o3as/lib/python3.6/abc.py", line 133, in __new__
    cls = super().__new__(mcls, name, bases, namespace, **kwargs)
  File "/home/borja/miniconda3/envs/o3as/lib/python3.6/site-packages/xarray/core/common.py", line 199, in __init_subclass__
    if not hasattr(object.__new__(cls), "__dict__"):
  File "/home/borja/Projects/o3skim/o3skim/mytest.py", line 9, in __getattr__
    return getattr(self.dataset, k)
  File "/home/borja/Projects/o3skim/o3skim/mytest.py", line 9, in __getattr__
    return getattr(self.dataset, k)
  File "/home/borja/Projects/o3skim/o3skim/mytest.py", line 9, in __getattr__
    return getattr(self.dataset, k)
  [Previous line repeated 325 more times]
RecursionError: maximum recursion depth exceeded

Anything else we need to know?:
I would recommend to add the following test or similar to ensure it is working:

class X(xarray.Dataset):
    __slots__ = ['dataset']
    def __init__(self, ds):
        self.dataset = ds

ds = xarray.Dataset()
x = X(ds)

Environment:

Output of xr.show_versions()

INSTALLED VERSIONS

commit: None
python: 3.6.12 |Anaconda, Inc.| (default, Sep 8 2020, 23:10:56)
[GCC 7.3.0]
python-bits: 64
OS: Linux
OS-release: 5.4.0-56-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: en_US.UTF-8
libhdf5: 1.10.5
libnetcdf: 4.6.3

xarray: 0.16.2
pandas: 1.1.4
numpy: 1.19.4
scipy: 1.5.4
netCDF4: 1.5.4
pydap: None
h5netcdf: None
h5py: None
Nio: None
zarr: None
cftime: 1.3.0
nc_time_axis: None
PseudoNetCDF: None
rasterio: None
cfgrib: None
iris: None
bottleneck: None
dask: 2.30.0
distributed: None
matplotlib: None
cartopy: None
seaborn: None
numbagg: None
pint: None
setuptools: 51.0.0.post20201207
pip: 20.3.1
conda: None
pytest: 6.1.2
IPython: None
sphinx: 3.2.1

[keewis]

inheriting from a Dataset or DataArray is not something we support right now, although that might change in the future.

The reason your examples fail with a RecursionError is a different one, though, you also get the same error if you remove all references to xarray and pass a string to your class. As pointed out in the SO question you referenced, you need to change your __init__:

class MyClass:
    __slots__ = ("dataset",)

    def __init__(self, ds):
        self.__class__.dataset.__set__(self, ds)

    def __getattr__(self, k):
        return getattr(self.dataset, k)

    def __setattr__(self, k, v):
        setattr(self.dataset, k, v)

ds = xr.Dataset()
x = MyClass(ds)

but I've also seen the use of __new__ instead of __init__ to avoid this somewhat ugly trick.

Be aware that the result of __getattr__ as it is right now will remove the custom class, so if you would like to avoid that you will have to add some sort of check:

    def __getattr__(self, k):
        result = getattr(self.dataset, k)
        if isinstance(result, (xr.Dataset, xr.DataArray)):
            return type(self)(result)
        else:
            return result

[keewis]

I assume this answered your question. If you disagree, feel free to reopen.

[max-sixty]

Though we don't support it officially, I've found subclassing a Dataset can be useful when adding methods. Most operations on the object will not persist the subclass, though.

[BorjaEst]

Sorry for my long time to reply, christmas...
@keewis thank you a lot for your anwer, I am glad to hear subclass might be supported one day. At the end, those are the python tools the programmers we work with.

For the moment, I will try to implement Extending xarray as you pointed using "xr.register_dataset_accessor". Create a new class that cannot be identified as 'isinstance(object, xarray.Dataset)' might be an issue in my case.