ENH: implement non-nano Timedelta scalar #46688
Conversation
|
@WillAyd thoughts on troubleshooting non-Mac build failures at import-time So far i've determined that yelling "but it isn't undefined" at the screen doesn't help. |
|
I think you need to modify setup.py so that timedeltas includes the header file np_datetime.h during compilation. |
|
Err sorry undefined symbol is a runtime error, so not a matter of include but rather a linkage error. Try adding |
|
OK, adding np_datetime.c in setup.py for timedeltas's "sources" tentatively helps |
jreback
added
Timedelta
| cdef: | ||
| str abbrev = npy_unit_to_abbrev(self._reso) | ||
| # TODO: way to create a np.timedelta64 obj with the reso directly | ||
| # instead of having to get the abbrev? |
There was a problem hiding this comment.
@seberg is there a C-API way to create a timedelta64 object?
There was a problem hiding this comment.
I guess PyArray_Scalar assuming you got the correct dtype available. (That function should only be used for NumPy dtypes IMO, but that isn't a problem)
There was a problem hiding this comment.
assuming you got the correct dtype available
what we have on hand is the correct NPY_DATETIMEUNIT. I guess we need to create the dtype from the unit (we have a function to go the other direction, so i guess this shouldn't be too hard to figure out). If I figure this out, I'll probably try to upstream it into numpy's __init__.pxd
There was a problem hiding this comment.
It would probably be OK to add a function that works with the unit directly for the C-API, also. But it doesn't exist yet. It seems PyArray_Scalar is commented out from __init__.pxd, I am not sure if there is a reason for that.
There was a problem hiding this comment.
could the reason by that PyArray_Scalar's first arg is void * which might not play so well with cython? i've figured out how to create the dtype object from the unit (copied create_datetime_dtype_with_unit over from multiarray/datetime.c) but so far having no luck in calling PyArray_Scalar
There was a problem hiding this comment.
Maybe? I don't really see a good reason, void * seems perfectly fine, if cython doesn't like it, just use char * (which is likely better anyway?)
But, I also don't think it is API that should be used a lot, so that may just be the reason also, that it is pretty unused.
| if len(state) == 1: | ||
| # older pickle, only supported nanosecond | ||
| value = state[0] | ||
| reso = NPY_FR_ns |
There was a problem hiding this comment.
sufficient testing on this?
doc/source/whatsnew/vX.X.X.rstfile if fixing a bug or adding a new feature.Not user-exposed. This is just enough to allow us to have non-nano TimedeltaArray.
asvs for tslibs.timedelta unchanged