Summing nanoseconds time delta to timestamp

[deponovo]

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the master branch of pandas.

Reproducible Example

import pandas as pd
res1 = (pd.Timestamp(1001) == pd.Timedelta(seconds=1e-9) + pd.Timestamp(1000))  # False, but should be True
res2 = (pd.Timestamp(1001) == pd.Timestamp(1) + pd.Timedelta(seconds=1000e-9))  # True
res1 == res2  # False, should be True
# this sum works properly `pd.Timestamp(1) + pd.Timedelta(nanoseconds=1)`

Issue Description

Adding Timedeltas that have nanoseconds defined as a float do not work properly for timedelta values in the [1-999] range.
This report is a result of this SO question

Expected Behavior

all([pd.Timestamp(1000 + x) == (pd.Timedelta(seconds=x*1e-9) + pd.Timestamp(1000)) for x in range(1000)]) should be True

Or any equivalent not using the explicit default nanoseconds unit or the nanoseconds keyword as in: pd.Timedelta(1) or pd.Timedelta(nanoseconds=1)

Installed Versions

INSTALLED VERSIONS

commit : c7f7443
python : 3.8.5.final.0
python-bits : 64
OS : Windows
OS-release : 10
Version : 10.0.18362
machine : AMD64
processor : Intel64 Family 6 Model 142 Stepping 12, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : de_DE.cp1252
pandas : 1.3.1
numpy : 1.20.3
pytz : 2021.1
dateutil : 2.8.1
pip : 21.1.2
setuptools : 49.6.0.post20210108
Cython : None
pytest : 6.2.4
hypothesis : None
sphinx : 4.0.2
blosc : None
feather : None
xlsxwriter : None
lxml.etree : 4.6.3
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : 3.0.1
IPython : 7.24.1
pandas_datareader: None
bs4 : None
bottleneck : 1.3.2
fsspec : None
fastparquet : None
gcsfs : None
matplotlib : 3.4.1
numexpr : 2.7.3
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyxlsb : None
s3fs : None
scipy : 1.6.3
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
numba : None

[FObersteiner]

To clarify a bit more, you can create a pandas Timedelta with nanoseconds specified;

td = pd.Timedelta(nanoseconds=1)
print(td.seconds, td.microseconds, td.nanoseconds)
# 0 0 1

You can also specify a certain number of microseconds by providing the microseconds as fractional seconds:

td = pd.Timedelta(seconds=1e-6)
print(td.seconds, td.microseconds, td.nanoseconds)
# 0 1 0

However what does not work is specifying nanoseconds like this:

td = pd.Timedelta(seconds=1e-9)
print(td.seconds, td.microseconds, td.nanoseconds)
# 0 0 0

So the point here to me is: everything that pandas Timedelta inherited from Python's timedelta works as expected. But when it comes to the additional feature, the nanoseconds attribute, a correct conversion from some inputs seems to be missing.

---

>>> sys.version
'3.9.4 (tags/v3.9.4:1f2e308, Apr  6 2021, 13:40:21) [MSC v.1928 64 bit (AMD64)]'
>>> pd.__version__
'1.3.3'

[jbrockmendel]

res1 = (pd.Timestamp(1001) == pd.Timedelta(seconds=1-9) + pd.Timestamp(1000)) # False, but should be True

First off I'm guessing you want 1e-9 instead of 1-9. Fixing that doesn't actually make a difference bc the Timedelta constructor casts 1e-9 to integer, in this case 0.

Could take a PR either a) raising if a non-integer is passed to Timedelta (depends on how much that check affects perf) or b) documenting this requirement

[jreback]

this is a duplicate issue
pls have a search of the tracker

[deponovo]

@jbrockmendel fixed the typo.

Timedelta constructor casts 1e-9 to integer, in this case 0.

This is not consistent with:
td = pd.Timedelta(seconds=1e-6)
as mentioned by @MrFuppes and also exemplified in my original message:

res2 = (pd.Timestamp(1001) == pd.Timestamp(1) + pd.Timedelta(seconds=1000e-9)) # True

[FObersteiner]

this is a duplicate issue pls have a search of the tracker

@jreback would not surprise me - but I couldn't find an exact dupe by searching the issues (also not knowing where exactly the described behavior originates from). If you have one in mind, please share.

potentially related:

• BUG: total_seconds() method returns zero for timedeltas smaller then 1 microsecond #40946 (comment)

[deponovo]

Take

[deponovo]

@mroeschke @jbrockmendel @jreback would it be ok if I solve this issue together with #40946 in one PR?