Fix evaluate_condition for non-bool result

[albertvillanova]

Fix evaluate_condition for non-bool result, as discussed in:

• Test evaluate_condition #634

I think there are 2 bugs in evaluate_condition:

• The call to .all() method in:

  smolagents/src/smolagents/local_python_executor.py

  Line 759 in b20da6a

  return result if isinstance(result, (bool, pd.Series)) else result.all()

  • This should be called explicitly (when pertinent) by the calling code
  • A library may return a non-bool result that does not implement the .all method
• The early return in:

  smolagents/src/smolagents/local_python_executor.py

  Lines 750 to 751 in cfe599c

  	if not isinstance(current_result, bool):
  	return current_result

  • Introduced by:
    • Add support for non-bool comparison operators. #612
  • Gives a wrong result for: pd.Series([1, 2, 3]) == pd.Series([2, 2, 2]) == pd.Series([3, 3, 3])
    • It was returning: pd.Series([False, True, False]
    • Instead of raising a ValueError("The truth value of a Series is ambiguous. Use a.empty, a.bool(), a.item(), a.any() or a.all().")

[albertvillanova]

Moving the evaluate_ast inside the loop, we avoid evaluating it for all condition.comparators.

• Now, once a current_result is False, we return False

[kc9zyz]

Thanks for cleaning this up @albertvillanova! I hadn't considered chained conditions in my PR, and I think your simplifications improve readability!

[albertvillanova]

Thank YOU, @kc9zyz. It was your PR which addressed the issue of non-bool results! Thanks again for your contribution.