Initialize custom backends in open_dataset

[Illviljan]

The backend classes are initialized in the build_engine function:

xarray/xarray/backends/plugins.py

Line 93 in 8b95da8

return {name: backend() forname, backendinbackend_entrypoints.items()}

This wasn't the case for custom backends:

xarray/xarray/backends/plugins.py

Line 161 in 8b95da8

backend=engine

This PR initializes the engine, fixes the incorrect signature in the test (#5033) and reverts the doc changes done in #5532.

• Tests added
• Passes pre-commit run --all-files
• User visible changes (including notable bug fixes) are documented in whats-new.rst

[github-actions]

Unit Test Results

6 files 6 suites 56m 48s ⏱️
16 226 tests 14 491 ✔️ 1 735 💤 0 ❌
90 552 runs 82 368 ✔️ 8 184 💤 0 ❌

Results for commit a4d12c1.

♻️ This comment has been updated with latest results.

[Illviljan]

I noticed this issue because my local mypy version was complaining about:
xrtools.py:750: error: Signature of "open_dataset" incompatible with supertype "BackendEntrypoint"

Anyone have any ideas why the CI version doesn't complain about this?

[Illviljan]

I noticed this issue because my local mypy version was complaining about:
xrtools.py:750: error: Signature of "open_dataset" incompatible with supertype "BackendEntrypoint"

Anyone have any ideas why the CI version doesn't complain about this?

The original test function did not have any typing defined. For mypy to do any signature checking either of the inputs or outputs needs to be typed.

This will throw the supertype error, uncomment self to get it passing:

importnumpyasnpimportxarrayasxrfromxarray.backendsimportBackendEntrypointexpected=xr.Dataset(
dict(a=2*np.arange(5)), coords=dict(x=("x", np.arange(5), dict(units="s")))
)
classCustomBackend(BackendEntrypoint):
defopen_dataset(
# self,filename_or_obj,
drop_variables=None,
**kwargs,
) ->xr.Dataset:
returnexpected.copy(deep=True)

[keewis]

This looks ready to me, and I think we should merge soon.