sum for OneElement

[jishnub]

This computes the sum over a OneElement in O(1). Along with the new FillMatrix multiplication method, this also means that we don't need to specialize ::AbstractFillMatrix * ::OneElementVector anymore.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 99.91%. Comparing base (9868632) to head (647c554).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #375   +/-   ##
=======================================
  Coverage   99.90%   99.91%           
=======================================
  Files           8        8           
  Lines        1110     1112    +2     
=======================================
+ Hits         1109     1111    +2     
  Misses          1        1           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[dlfivefifty]

We also want to support dims keyword

but should we be overloading mapreduce directly?

[jishnub]

In principle, we should be specializing mapreduce, but that's more effort. We can do that in the future and remove this sum specialization. As of now, sum is more useful than other functions.

[dlfivefifty]

I do think we should support dims. Ie sum(::OneElementMatrix; dims=1) would return a OneElementMatrix

Or at the very least we need tests. Actually I suspect this has broken dims

[dlfivefifty]

It might be necessary to overload Base._sum to ensure type stability

[jishnub]

I don't think this has broken dims, as sum(A::OneElement; dims) would fall back to the method for AbstractArrays. We may specialize the method with dims, though.
Currently

julia> A = OneElement(3, 4);

julia> sum(A, dims=1) == sum(Array(A), dims=1)
true

julia> sum(A, dims=2) == sum(Array(A), dims=2)
true

julia> sum(A, dims=1:2) == sum(Array(A), dims=1:2)
true

Is there anything in particular that you think is broken? Also, what type-stability were you referring to? Between the method without dims and the one with it? As such, these are different methods, and each is type-stable by itself, although they return different types.