Haha, sure My point was more on the semantic.
I agree a week is 604800 seconds but what about month (and year to a greater extent)?

From what I see in the code, it looks like in this particular case the end date would be adjusted to 48 hours (i.e. start: Jan1@midnight, end: Jan3@midnight) but the 'return size' will be 2, so you'll get two output rows: Jan1@midnight, and Jan2@midnight.
Understood! Thanks for looking into the code!
I think all this deserves to be documented.