Week.onOffset looks fishy

[jbrockmendel]

A Week offset with weekday=None behaves very much like a Tick offset. The main difference is how they handle onOffset. Assuming normalize is False, the Week object will always return week.onOffset(other) --> False while the Day object will always return day.onOffset(other) --> True.

Is this intentional?

[gfyoung]

@jbrockmendel : Would you mind posting a small code sample for this just for reference?

[jbrockmendel]

In [2]: week = pd.offsets.Week()
In [3]: day7 = pd.offsets.Day(n=7)
In [5]: ts = pd.Timestamp.now()
In [6]: ts
Out[6]: Timestamp('2017-11-27 13:03:35.592386')
In [7]: ts + day7
Out[7]: Timestamp('2017-12-04 13:03:35.592386')
In [8]: ts + week
Out[8]: Timestamp('2017-12-04 13:03:35.592386')
In [9]: week.onOffset(ts)
Out[9]: False
In [10]: day7.onOffset(ts)
Out[10]: True
In [12]: day7.rollforward(ts)
Out[12]: Timestamp('2017-11-27 13:03:35.592386')
In [13]: week.rollforward(ts)
Out[13]: Timestamp('2017-12-04 13:03:35.592386')

It isn't that the Week.onOffset behavior is wrong per se, just that I don't see why it behaves this way. In every other way it behaves like a Tick offset.

[gfyoung]

@jbrockmendel : Indeed, it's difficult for me to say that it's wrong when there is ZERO documentation on that function. 😄 (in fact, if you would like to add documentation to that method, be my guest!)

That being said, I do see what you mean re: consistency. @jreback @jorisvandenbossche thoughts?

[jbrockmendel]

@jreback is it the case that offset.onOffset(dt) should always be the same as (the base class implementation) (dt + offset) - offset == dt? If so, this is a bug and should be resolved before #18738.

[jbrockmendel]

@gfyoung mind changing the tag from Timedelta to Frequencies?

[gfyoung]

@jbrockmendel : done