Add frequency inference to DatetimeIndex set ops to preserve frequency where possible

[wesm]

No description provided.

[adamklein]

Do you agree that only a regular conforming datetimeindex (ie with no gaps) should have a frequency associated with it? An aside, it gets a bit muddy to think about "frequency" with an interval index, which uses the same term to refer to duration of time within each interval, and not to the regular recurrence of observations. I might prefer another term altogether ("interval" comes to mind) for intervalindex to disambiguate.

[wesm]

I'm not sure. freq was what was used in scikits.timeseries, but given the dual distinctions... Will have to think about it and see what seems most reasonable