Sum of all NA and empty should be 0

[TomAugspurger]

And the prod of those should be 1.

xref: #9422, #17630, https://mail.python.org/pipermail/pandas-dev/2017-November/000657.html

We need to implement that and design and implement the alternative (either a new method, or a keyword-argument to sum and prod).

[jbrockmendel]

I've been following this conversation a bit and remain confused on one point. Options 1 and 2 both appear to rely on the premise that Series([nan, nan]).sum() should equal Series([]).sum().
Without having an intuition for why this is a desirable behavior, the merits of those two options hard to evaluate. Is it just consistency with [other software]? LMK if this question belongs elsewhere.

[TomAugspurger]

Here's fine. I think the shortest answer is that it's nice to have s.sum(skipna=True) equal s.dropna().sum(skipna=True/False).

[jorisvandenbossche]

I would keep the more general discussion on the mailing list, and here more the details of the changes we then want to make

[jreback]

this is not happening for 0.21.1

[jorisvandenbossche]

this is not happening for 0.21.1

I think the question is rather: do we want to release this quickly, or only in a few months? (and then once answered, we can still discuss whether we call that 0.21.1, 0.21.2 or 0.22.0). But let's keep that discussion in #18244 (comment)

[ryanbressler]

Just to add a data point I tried to update to .21 and a large number of our unit tests failed so I came looking to see if this had been reported as a bug yet. A lot of our code assumes pd.Series([]).sum() == 0 for simple things like dollar weighted sums.

We won't be upgrading until this is fixed and the fact a breaking change like this was made has lead us to seriously consider prohibiting pandas use within critical sections of our code base.