BUG: pd.Timedelta np.int, np.float. fixes #8757

[wholmgren]

closes #8757

[jreback]

maybe add another loop which tests each of the kwargs (e.g. days,minutes,seconds,milliseconds,microseconds,nanoseconds)

[wholmgren]

Done.

[jreback]

perfect. Can you put an example which triggers this last (e.g. Timedelta(days='foo'))

[wholmgren]

hm, actually, no. np.issubdtype chokes on anything that's not a dtype, so I don't think that the additional error text is reachable. Two potential options:

1. Remove my changes to the error reporting and let the numpy error reach the user.
2. Move the kwargs = dict() line inside the try block and test for numpy's TypeError: data type "foo" not understood in the except block.

I'd go for the first option.

[jreback]

better then to use com.is_integer/com.is_float (where you have np.issubdtype now, as these will always return properly) (so maybe just simpler to coerce directly)

e.g.

def convert(v):
    if com.is_integer(v):
        return int(v)
    elif com.is_float(v):
        return float(v)
   raise ValueError("invalid type {0}".format(type(v)))

[wholmgren]

Ah, I didn't know about these functions. It seems that I've run into a circular import problem, though. I added import pandas.core.common as com to the top of tslib and I get this when importing pandas:

  File "/home/will/git_repos/pandas/pandas/__init__.py", line 7, in <module>
    from . import hashtable, tslib, lib
  File "pandas/tslib.pyx", line 41, in init pandas.tslib (pandas/tslib.c:77611)
    import pandas.core.common as com
  File "/home/will/git_repos/pandas/pandas/core/common.py", line 18, in <module>
    import pandas.algos as algos
  File "pandas/algos.pyx", line 64, in init pandas.algos (pandas/algos.c:181716)
    from pandas import lib
  File "pandas/lib.pyx", line 50, in init pandas.lib (pandas/lib.c:78075)
    from tslib import NaT, Timestamp, Timedelta
ImportError: cannot import name NaT

Could we import is_float, is_int in core.api?

[jreback]

oh, forgot, these you are in cython already, use use is_intege_object/is_float_object they are already defined

[wholmgren]

Great. One more question: Did you really want a ValueError for the convert function? I thought that TypeError would be more appropriate.

[jreback]

or sorry, TypeError is right. Was a typo :)

[wholmgren]

Thanks for the pointers. Let me know if you want any other changes.

[jreback]

perfect thanks!