BUG: maybe_convert_numeric fails with uint64 (original) (raw)

master at aba7d2:

from pandas import lib, np arr = np.array([2**63], dtype=object) lib.maybe_convert_numeric(arr, set()) ... OverflowError: Python int too large to convert to C long

At first, I thought the patch would be similar to #14916, but there are several complications here:

• Handling an array with negative and uint64 values: in maybe_convert_objects, we would just return object to avoid data truncation. However, this function currently only returns numeric arrays. And if we do decide to return non-numeric, should we then raise if coerce_numeric=True then?
• uint64 has no convenient na_value, so what happens if a uint64 is in na_values?