#2157 Difference in total() and reduceSum() behaviour

[Ivar Christopher]

I have (abridged):

PoyFluxFromEq = rank-2 dataset with units.
NetPoyFluxFromEq = fltarr(PoyFluxFromEq.length())
for I in range(PoyFluxFromEq.length()):

   NetPoyFluxFromEq[I]= total(PoyFluxFromEq[I, some_rank-1_index])

or

   NetPoyFluxFromEq[I]= reduceSum(PoyFluxFromEq[I, some_rank-1_index], 0)

The version using reduceSum() transfers the units of PoyFluxFromEq to NetPoyFluxFromEq. The version using total() does not.

My understanding was that total() and reduceSum() should behave identically now.

[Jeremy Faden]

Small problem, it turns out, and that's that the total function returns a double, not a rank0 dataset with a unit. There's a "total" function which takes a second argument, the index to total over, which also handles the units and is probably the source of the confusion.

This script demos:

randomSeed(5334)
ds= dataset( randn(20), units='Kg' )
print total(ds)
print total(ds,0)
print reduceSum(ds,0)
ds= dataset( randn(20,30), units='Kg' )
while ( ds.rank()>0 ): ds= total(ds,0)
print ds

So we could make the total function return a rank 0 dataset instead of a double, but I'd want to do that over a major release (v2019b or v2020a)...

[Jeremy Faden]

For v2019a_5, I should sweep through all the filters and see what still uses double. Once rank 0 datasets were introduced, this sort of code should have been removed, but I suspect it wasn't, and this is the last remaining one.