BUG: Timedelta components no longer rounded with high precision integers #31380
Conversation
| @@ -859,14 +859,6 @@ cdef class _Timedelta(timedelta): | |||
| """ | |||
| return self.to_timedelta64() | |||
|
|
|||
| def total_seconds(self): | |||
There was a problem hiding this comment.
No longer need this implementation since we can rely on timedelta.total_seconds
There was a problem hiding this comment.
what change for the "no longer" part?
There was a problem hiding this comment.
Since we are now dispatching the correct microseconds to datetime.timedelta here: https://github.com/pandas-dev/pandas/pull/31380/files#diff-d23555c3fb6b3ffc9f2088e09f9ec5f9R1245
we can use datetime.timedelta.total_seconds as mentioned in this #31155 (comment) since we are no longer rounding.
There was a problem hiding this comment.
sounds good, thanks for explaining
There was a problem hiding this comment.
FWIW, I did consider suggesting the removal of total_seconds here, but when I benchmarked it, I found that the implementation where you truncate ._value was about 2x as fast as the native timedelta.total_seconds(), if that matters at all.
| @@ -109,7 +109,7 @@ Datetimelike | |||
| Timedelta | |||
| ^^^^^^^^^ | |||
|
|
|||
| - | |||
| - Bug in constructing a :class:`Timedelta` with a high precision integer that would round the :class:`Timedelta` components (:issue:`31354`) | |||
There was a problem hiding this comment.
BTW, I'm not sure if it was already covered by the earlier PR, but it seems like the change in precision to total_seconds() from "dubiously nanosecond-precision" to "microsecond-precision" should probably show up in the release notes. It seems like this is the only entry under Timedelta, so maybe it was done silently? Or was the other change already released?
black pandasgit diff upstream/master -u -- "*.py" | flake8 --diff