Python difference in datetime now against datetime now

I have got into an issue or might quite possibly feature in turn! Not sure, wondering!! In python's datetime library, to get difference in time, as in below snippet.

>>> import datetime
>>> datetime.datetime.now() - datetime.datetime.now()
datetime.timedelta(-1, 86399, 999958)
>>> tnow = datetime.datetime.now()
>>> datetime.datetime.now() - tnow
datetime.timedelta(0, 4, 327859)

I would like to understand why datetime.datetime.now() - datetime.datetime.now() is producing output as -1 days, 86399 seconds whereas assigning current time to some variable and computing difference gives desired output 0 days, 4 seconds.

The results seems to be bit confusing, it would be helpful if someone could decode whats going behind

Note: I'm using Python 2.7

Solution

From the documentation for timedelta objects:

Note that normalization of negative values may be surprising at first. For example:
>>> from datetime import timedelta
>>> d = timedelta(microseconds=-1)
>>> (d.days, d.seconds, d.microseconds)
(-1, 86399, 999999)

This occurs in both Python 2.7 and Python 3.

The reason for the observed result is simple:

a , b = datetime.datetime.now(), datetime.datetime.now()
# here datetime.now() in a will be <= b.
# That is because they will be executed separately at different CPU clock cycle.

a - b
# datetime.timedelta(-1, 86399, 999973)

b - a
# datetime.timedelta(0, 0, 27)

To get the proper time difference:

(tnow - datetime.datetime.now()).total_seconds()
# output: -1.751166