humanize overflows before datetime does

What did you do?

I tried to humanize a very long time.

What did you expect to happen?

I expect the time to be converted into years/millenia/eon if it is within the bounds of datetime i.e. <= 999 999 999 days, and if the number is higher than that or uncountable with the compute platform (64/32 bits) then something like “longer than countable”.

At the very least, this type of error should be handled like others in _date_and_delta: the timedelta is returned unaltered.

What actually happened?

I received an overflow error from an operation within humanize.

What versions are you using?

• OS: Ubuntu x64
• Python: 3.9.6
• Humanize: 3.13.1

Please include code that reproduces the issue.

The best reproductions are self-contained scripts with minimal dependencies.

import datetime as dt
import humanize
d = dt.timedelta(days=999999999)
print(d)
h = humanize.naturaldelta(d)
print(h)
# ... e/time.py", line 131, in _date_and_delta
#    date = now - value
# OverflowError: date value out of range

Once we discuss how to best handle this error, I’d be happy to open a PR.

[carterbox]

To summarize: the original issue is that an OverflowError is raised by naturaldelta() at deltas much smaller than what datetime can handle. Other errors in naturaldelta() cause a timedelta to be returned instead of a string, but this is undocumented. We want to retain the current behavior (to raise errors sometimes, return string sometimes, and return timedelta sometimes), but we want to document when each of these outcomes occurs.

Can I still refactor naturaldelta() so that the OverflowError is raised by datetime instead of by humanize? I believe this is compatible with the approach above given the unrelated deprecation of naturaldelta(..., when).

[hugovk]

✅