BUG (string dtype): convert dictionary input to materialized string array in ArrowStringArray constructor

[jorisvandenbossche]

#54074 allowed to pass dictionary encoded string data in the ArrowStringArray constructor, because such values can be passed when reading partitioned datasets (it was fixing #53951).

However, when you actually have a column with string dtype but backed by a dictionary encoded pyarrow array, our StringArray implemention is not set up for that, so once you call some string specific functionality, you can run into errors. Example:

In [2]: arr = pd.core.arrays.ArrowStringArray(pa.array(["a", "b", None, "a"], pa.large_string()).dictionary_encode()) In [3]: arr Out[3]: <ArrowStringArray> ['a', 'b', <NA>, 'a'] Length: 4, dtype: string In [4]: pd.Series(arr) Out[4]: 0 a 1 b 2 <NA> 3 a dtype: string In [5]: pd.Series(arr).str.len() ... ArrowNotImplementedError: Function 'utf8_length' has no kernel matching input types (dictionary<values=large_string, indices=int32, ordered=0>) 

Maybe we could at some point properly support storing dict encoded string values under the hood. But until then, I think it is better to materialize the dictionary encoded input to a plain pyarrow string array, which is what this PR is doing.

• Tests added and passed if fixing a bug or adding a new feature
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

[mroeschke]

Thanks @jorisvandenbossche