`NullityIndex` -- indexing non-empty entries of a `NullableArray`

nullityindex.jl

immutable NullityIndex
    non_nulls::Vector{Int}
end

function NullityIndex(X::NullableArray)
    inds = Vector{Int}()
    sizehint!(inds, length(X.isnull))
    for i in eachindex(X.isnull)
        !X.isnull[i] && push!(inds, i)
    end
    return NullityIndex(inds)
end

# test that 'NullityIndex'es cover all and only the non-null indices of a
# `NullableArray` -- returns `true` if success.
function test(X::NullableArray)
    ni = NullityIndex(X)
    res = true
    for i in ni.non_nulls
        res &= !X.isnull[i]
    end
    for i in setdiff(eachindex(X.isnull), ni.non_nulls)
        res &= X.isnull[i]
    end
    return res
end

function f(tup::Tuple{NullableArray, NullityIndex})
    X, ni = tup
    res = Nullable(0.0)
    for i in ni.non_nulls
        @inbounds res += X[i]
    end
    return res
end

function g(X::NullableArray)
    res = Nullable(0.0)
    for i in eachindex(X)
        @inbounds !X.isnull[i] && (res += X[i])
    end
    return res
end

function profile(N::Int)
    X = NullableArray(rand(N), rand(Bool, N))
    ni = NullityIndex(X)
    print("NullityIndex(X) (~half missing entries, N = $N)")
    @time NullityIndex(X)
    a = f((X, ni))
    b = g(X)
    println("f(X) = ", a, " g(X) = ", b)
    println()
    print("f (uses NullityIndex): "); @time f((X, ni))
    print("g (normal looping):    "); @time g(X)
    nothing
end

Yes, this is better:

julia> profile(10_000_000)
NullityIndex(X) (~half missing entries, N = 10000000) 106.550 milliseconds (3 allocations: 78125 KB, 10.81% gc time)
f(X) = Nullable(2.4977616855296516e6) g(X) = Nullable(2.4977616855296516e6)

f (uses NullityIndex):   11.968 milliseconds (1 allocation: 32 bytes)
g (normal looping):      53.224 milliseconds

💯