Show microseconds in str() output, and suggestied change to repr() output

[davidhassell]

Hello,

Currently, fractions of seconds are not shown in the str() representation of a cftime date-time object:

In [5]: import cftime                                                                                                                                                             

In [6]: cftime.__version__                                                                                                                                                        
Out[6]: '1.0.3.4'

In [7]: dt = cftime.DatetimeGregorian(2000, 2, 13, 12, 34, 23, 123456)                                                                                                            

In [8]: repr(dt)                                                                                                                                                                  
Out[8]: 'cftime.DatetimeGregorian(2000, 2, 13, 12, 34, 23, 123456, 6, 44)'

In [9]: str(dt)                                                                                                                                                                   
Out[9]: '2000-02-13 12:34:23'

I would like to suggest that they are shown, in a ISOI 8601-compliant fashion (https://en.wikipedia.org/wiki/ISO_8601#General_principles), so that the str() output would look like:

>>> str(dt)
'2000-02-13 12:34:23.123456'

I am happy to provide a pull request if this idea is accepted.

On a related note, might it be possible to alter the repr() output to improve its readaibility? I fully appreciate that this is subjective, and the developers have the last word, but I'll describe how it could perhaps be changed in case you agree!

We currently have:

In [8]: repr(dt)                                                                                                                                                                  
Out[8]: 'cftime.DatetimeGregorian(2000, 2, 13, 12, 34, 23, 123456, 6, 44)'

I would like to suggest:

 >>> rep(dt)
'cftime.DatetimeGregorian(2000-02-13 12:34:23.123456)'

My use case for this is that I frequently inspect numpy arrays of cftime objects, and the current repr() output delivered by numpy is hard for a (this!) human to parse quickly. We currently have:

In [10]: dt2 = cftime.DatetimeGregorian(2000, 2, 14, 12, 34, 23, 123456)                                                                                                          

In [11]: dt3 = cftime.DatetimeGregorian(2000, 3, 13, 12, 34, 23)                                                                                                          

In [12]: import numpy                                                                                                                                                             

In [13]: numpy.array([dt, dt2, dt3])                                                                                                                                              
Out[13]: 
array([cftime.DatetimeGregorian(2000, 2, 13, 12, 34, 23, 123456, 6, 44),
       cftime.DatetimeGregorian(2000, 2, 14, 12, 34, 23, 123456, 0, 45),
       cftime.DatetimeGregorian(2000, 3, 13, 12, 34, 23, 0, 0, 73)],
      dtype=object)

... but I think the following is easier to read:

>>> numpy.array([dt, dt2, dt3])
array([cftime.DatetimeGregorian(2000-02-13 12:34:23.123456),
       cftime.DatetimeGregorian(2000-02-14 12:34:23.123456),
       cftime.DatetimeGregorian(2000-03-13 12:34:23)],
      dtype=object)

I think that this output is still true to that required by repr - i.e you can still recreate the original object from this information.

Again, I would happy to include this in a pull request.

Many thanks for considering this,
David

[jswhit]

A pull request for both of these would be welcome, thanks. One comment on changing the repr() output for Datetime objects, the proposed change would be inconsistent with python datetime behaviour (which is similar to the current cftime behavior). I don't really mind - but someone might.

[davidhassell]

Thanks. I'll put together a pull request, and see if any further comments or objections arise.

[jswhit]

thinking about it some more, it seems like

 >>> repr(dt)
'cftime.DatetimeGregorian(2000-02-13 12:34:23.123456)'

implies that the DatetimeGregorian constructor accepts an ISO 8601 string. I think I'd prefer we leave the repr() behavior as it is, but keep your improvement to str().

[davidhassell]

Fair enough.

I'm thinking of writing a pull request that would allow the cftime.datetime object to accept ISO 8601-like strings for initialization (in addition to the existing integer arguments). Would that be OK?

[jswhit]

Sure, that would be fine.

[jswhit]

PR merged, thansk!