ODS supports timedelta ([hh]:mm:ss). times_1904.ods is broken, see #55045. Maybe fix the file and do something like?

if h > 23:
    return pd.Timedelta(...)
else:
    return cast(Scalar, datetime.time(...))

Right, timedelta seem to make more sense than timestamp - at least for ODF time-value, because durations can be larger than 24h and they can be negative, see duration.ods:

It's my first PR here, so I thought I'd better keep it tight and clean - hoping we could leave timedelta for a follow-up discussion and new PR. For sure unit tests would need adjustments (they specifically require datetime.time timestamps) and I don't know how other spreadsheet formats correspond.

Am I correct in saying that the current behavior of pandas will read the top value in the screenshot above as having 50 hours, where as this change will now be 50 - 48 = 2 hours?

@rhshadrach Which commit/version of pandas yields 50 hours for you ?
I thought you'd get an error in _odfreader.py:217 similar to:

>>> pd.Timestamp('50:15:00') 
Traceback (most recent call last):
  File "parsing.pyx", line 681, in pandas._libs.tslibs.parsing.dateutil_parse
ValueError: hour must be in 0..23

Which commit/version of pandas yields 50 hours for you ?

I haven't run anything.

I thought you'd get an error in _odfreader.py:217 similar to:

I think you're saying that both main and this PR will raise on duration.ods, is that right?

Nope, I only said that the current implementation does not support time value equal or larger than 24 hours in ODF files.

Could you please run the files you'd like to check and share any results that raise your concerns ?

I only said that the current implementation

Does current implementation mean main or this PR?