BUG: .str.contains `na` validation

[jbrockmendel]

import pandas as pd
import pyarrow as pa

ser = pd.Series(['a', 'b', None], dtype=pd.StringDtype(storage="pyarrow"))
ser2 = ser.astype(pd.ArrowDtype(pa.string()))

ser.str.contains("foo", na="bar")  # <- casts "bar" to True
ser2.str.contains("foo", na="bar")  # <- raises

There's a small difference in the _str_contains methods in ArrowExtensionArray vs ArrowStringArray. The latter uses bool(na) when filling null entries, the former uses na directly.

I prefer the no-casting behavior, but mainly think we should be consistent.

update Looks like pandas/tests/strings/test_find_replace.py::test_contains_nan specifically tests na="foo"

[mroeschke]

+1 for the no casting behavior

[jorisvandenbossche]

Regardless of which option we choose, it is different from current object dtype (which just puts "bar" in the result as object dtype), so it's another case we should document in #59328

[jbrockmendel]

how about we deprecate (in 2.3) the no-validation behavior across the board?

[jorisvandenbossche]

Yes, that sounds as a good idea. That makes it less of a plain breaking change for 3.0