mapreduce with multiple input containers

[andyferris]

It occurs to me that it would be great if operations like mapreduce(f, op, a, b) actually worked where a and b are two containers of the same length, like it does for map.

For me, there are two issues here.

• There is no varargs form for mapreduce that applies zip to the inputs and the appropriate form.
• Perhaps this is just my taste, but I honestly find the placement of the optional v0 term is a little annoying in mapreduce(f, op, v0, a). If it is optional, maybe it should be an optional argument or a keyword argument. And it can't be an optional (positional) argument when the last term is a varargs list of the input containers.

My suggestion would be a signature something like mapreduce(f, op, a...; v0 = nothing) (we might need something other than nothing, since it seems to mean something these days). Moving v0 would be necessary to have the v0-less mapreduce algorithm with two or more input containers (since it can't magically infer whether the 3rd argument is meant to be v0 or a container to reduce over).

xref: #27677 (comment) cc: @StefanKarpinski

[StefanKarpinski]

Agree that v0 should be a keyword; v0 is not the best name, however, perhaps initial or first?

[andyferris]

Changing v0 in mapreduce would probably mean we should change v0 in reduce...

initial seems OK, I think. R uses init. Python uses initializer. C++ uses init. Haskell documenation refers to "initial value". Rust used init. OTOH, LINQ uses seed but not sure I love it.

We could save 3 letters and go with init.

[StefanKarpinski]

init seems good to me and fairly widely used it seems. seed is terrible.

[fredrikekre]

This should be implementable now since init is a kwarg, right? Would be nice to have.