DataFrame._data deprecation in pandas

[j-bennet]

Fixes test failure in upstream:

FAILED dask/dataframe/tseries/tests/test_resample.py::test_series_resample[frame-ohlc-5-h-left-right] - FutureWarning: DataFrame._data is deprecated and will be removed in a future version. Use public APIs instead.

Xref pandas-dev/pandas#52003.

• Tests added / passed
• Passes pre-commit run --all-files

[jrbourbeau]

Thanks @j-bennet! I pushed a small commit to just use ._mgr unconditionally since that appears to exist for all supported pandas versions (including dev pandas).

@phofl the deprecation warning says to use a public API instead of ._data. ._mgr works, but is still private. Is there a public API you'd recommend? Alternatively, is there a different approach we could take in this function? This function is meant to produce a unique, deterministic hash for a given pd.DataFrame

[phofl]

_mgr should be fine for now.

You could iterate over all columns of the DataFrame to collect the arrays, but this is obviously slower then accessing them directly. So not really a good idea if performance is relevant here. I guess hash_pandas_object is not what you are looking for?

[jrbourbeau]

_mgr should be fine for now.

Sounds good 👍

I guess hash_pandas_object is not what you are looking for?

Ah, good point. If there was a utility in pandas that we could offload this to, and there weren't any serious performance regressions, we probably would. I'll open a separate issue for this

[jrbourbeau]

xref #10083