BUG: offsets normalize and __eq__

[jbrockmendel]

Two otherwise-identical DateOffsetobjects with different normalize attributes evaluate to equal. But they do not behave equivalently:

In [1]: from pandas.tseries.offsets import *
In [2]: import pandas as pd
In [3]: ts = pd.Timestamp.now()
In [4]: off1 = DateOffset(days=1)
In [5]: off2 = DateOffset(days=1, normalize=True)

In [6]: off1 == off2
Out[6]: True

In [7]: ts + off1
Out[7]: Timestamp('2017-09-27 21:13:54.502976')

In [8]: ts + off2
Out[8]: Timestamp('2017-09-27 00:00:00')

These objects should not be considered equal. If there's consensus, I'll make a PR.

[jreback]

this is probably a bug. we want to compare all params (though that is not entirely cheap).

[jbrockmendel]

Great. I'll add it to the tslibs TODO list. Don't worry about the cheapness; even without the cythonizing, there are some pretty significant speedups in the pipeline.

[jbrockmendel]

Does fixing the normalize comparison count as a breaking change that needs some special treatment, or can I just go ahead and fix it?

[jbrockmendel]

Closed by #21404