BUG: Wrong Result when subtracting BusinessHour offset

[DavidKleindienst]

Pandas version checks

• 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 main branch of pandas.

Reproducible Example

import pandas as pd
from datetime import datetime

offset = pd.offsets.BusinessHour()
dt = datetime(2020, 1, 1, 10, 00)

dt - offset
# Expected Timestamp('2020-01-01 09:00:00') but got Timestamp('2019-12-31 17:00:00')

Issue Description

Subtracting BusinessHour offsets from datetime returns wrong results if the datetime is not an opening time.
Rather than returning the hour before, it returns the opening time of the next day.

This seems to be caused by the following code in offsets.pyx BusinessHour._apply:

if (
    bhour_remain > bhour
    or bhour_remain == bhour
    and nanosecond != 0
):

According to the comments nanosecond != 0 serves to detect edge cases but incorrectly also detects this typical case.
Removing the nanosecond != 0 from the if condition fixes this issue, but causes several other tests (presumably those checking for edge cases) to fail.

Expected Behavior

datetime(2020, 1, 1, 10, 00) - BusinessHour() should return Timestamp('2020-01-01 09:00:00') and not Timestamp('2019-12-31 17:00:00')

Installed Versions

INSTALLED VERSIONS

commit : 8dab54d
python : 3.10.8.final.0
python-bits : 64
OS : Windows
OS-release : 10
Version : 10.0.19045
machine : AMD64
processor : Intel64 Family 6 Model 167 Stepping 1, GenuineIntel
byteorder : little
LC_ALL : None
LANG : None
LOCALE : German_Austria.1252

pandas : 1.5.2
numpy : 1.23.5
pytz : 2022.7
dateutil : 2.8.2
setuptools : 65.5.0
pip : 22.3.1
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : None
IPython : None
pandas_datareader: None
bs4 : None
bottleneck : 1.3.5
brotli : None
fastparquet : None
fsspec : None
gcsfs : None
matplotlib : None
numba : None
numexpr : 2.8.4
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : None
snappy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
zstandard : None
tzdata : None

[dicristina]

I could not reproduce the issue with the example given. I got Timestamp('2019-12-31 17:00:00') in 1.5.2 and 1.5.0, as well as in master.

INSTALLED VERSIONS

commit : 8dab54d
python : 3.10.6.final.0
python-bits : 64
OS : Linux
OS-release : 5.15.0-25-generic
Version : #25-Ubuntu SMP Wed Mar 30 15:54:22 UTC 2022
machine : x86_64
processor : x86_64
byteorder : little
LC_ALL : None
LANG : en_US.UTF-8
LOCALE : en_US.UTF-8

pandas : 1.5.2
numpy : 1.24.1
pytz : 2022.7
dateutil : 2.8.2
setuptools : 59.6.0
pip : 22.0.2
Cython : None
pytest : None
hypothesis : None
sphinx : None
blosc : None
feather : None
xlsxwriter : None
lxml.etree : None
html5lib : None
pymysql : None
psycopg2 : None
jinja2 : None
IPython : 8.7.0
pandas_datareader: None
bs4 : None
bottleneck : None
brotli : None
fastparquet : None
fsspec : None
gcsfs : None
matplotlib : None
numba : None
numexpr : None
odfpy : None
openpyxl : None
pandas_gbq : None
pyarrow : None
pyreadstat : None
pyxlsb : None
s3fs : None
scipy : None
snappy : None
sqlalchemy : None
tables : None
tabulate : None
xarray : None
xlrd : None
xlwt : None
zstandard : None
tzdata : None

[DavidKleindienst]

Oh, probably I copied result from a branch I was working on, sorry for that! I've changed the result in the post above to Timestamp('2019-12-31 17:00:00'), thanks for pointing that mistake out.
Regardless of that, Timestamp('2019-12-31 17:00:00') is still the wrong result, the correct result would be Timestamp('2020-01-01 9:00:00')

[dicristina]

With respect to arithmetic with BusinessHour offsets, the end time of a day, the start time of the following day and the times in between those two times are equivalent. According to the user guide:

Different from other offsets, BusinessHour.rollforward may output different results from apply by definition.

This is because one day’s business hour end is equal to next day’s business hour start. For example, under the default business hours (9:00 - 17:00), there is no gap (0 minutes) between 2014-08-01 17:00 and 2014-08-04 09:00.

Consider the following examples:

t1 = pd.Timestamp("2023-01-03T10:00") - pd.offsets.BusinessHour() # End time of 2023-01-02
t2 = pd.Timestamp("2023-01-02T16:00") + pd.offsets.BusinessHour() # Start time of 2023-01-03
t1, t2

# We can see that they were equivalent for our arithmetic
o = 2 * pd.offsets.BusinessHour()
t1 + o, t2 + o

This is counterintuitive because t1 and t2 are very different in everyday (calendar) usage. Here is another example:

timestamps = pd.DatetimeIndex(["2023-01-02T17:00", "2023-01-02T19:00", "2023-01-02T20:47:23", "2023-01-03T09:00"])
for t in timestamps:
     print(f"Add 1H to {t}:", t + pd.offsets.BusinessHour())

[DavidKleindienst]

Oh, that makes sense. Thank you so much for the thorough explanation!
Everything working as intended, so I'm closing this.