How to work with type unstable functions?

General Usage
,
2

Well, after some considerations, it looks like there is a third option. If it is known beforehand, that values are of little interest, and they should be processed to get some result, we can just do it in place.

function solve3(f, poly)
    a, b, c = poly
    if a == 0
        if b == 0
            return f(NoRoots()) # It should be infinity but 4 types is more than enough already
        else
            return f(SingleLinearRoot(-c/b))
        end
    end

    D = b^2 - 4*a*c
    if D < 0
        return f(NoRoots())
    elseif D == 0
        return f(SingleRoot(-b/(2*a)))
    else
        sD = sqrt(D)
        return f(TwoRoots((-b + sD)/(2*a), (-b - sD)/(2*a)))
    end
end

This small change uses multiple dispatch system without abusing it (I guess?). And it is even faster than flag system, probably because Julia can inline functions and optimize resulting code.

julia> @btime sum(x -> solve3(cnt, x), $polynomials)
  1.554 μs (0 allocations: 0 bytes)

I can’t see any problems with this approach. Am I wrong or it is an idiomatic approach to this issue?

5 Likes