Hi, I'm trying to implement the max-plus semiring algebra in Julia 1.0.3 like the follow code:

struct MP{T} <: Number val::T end
Base.:+(a::MP, b::MP) = MP(max(a.val, b.val))
Base.:*(a::MP, b::MP) = MP(a.val + b.val)
Base.show(io::IO, k::MP) = show(io, k.val)
Base.convert(::MP{T}, x) where T = MP(T(x))
Base.zero(::MP{T}) where T = MP(typemin(T))
Base.one(::MP{T}) where T = MP(zero(T))
Base.zero(::Type{MP{T}}) where T = MP(typemin(T))
Base.one(::Type{MP{T}}) where T = MP(zero(T))
mparray(A::Array) = map(MP, A)

The following code gives the expected behavior :

julia> Base.zero(MP{Float64})
-Inf

julia> Base.one(MP{Float64})
0.0

But the following code does not gives the expected behavior :

julia> Matrix{MP{Float64}}(I, 2, 2)
2×2 Array{MP{Float64},2}:
  1.0  -Inf
 -Inf   1.0

What I was expected:

julia> Matrix{MP{Float64}}(I, 2, 2)
2×2 Array{MP{Float64},2}:
  0.0  -Inf
 -Inf   0.0

I think Matrix{MP{Float64}}(I, 2, 2) creates a matrix initialized by Base.zero(MP{Float64}) calling MP(typemin(Float64)) giving -Inf but the I does not create the diagonal initialized by one(MP{Float64}) which would called MP(zero(Float64)) giving 0.0.

I followed this link JuliaLang/julia#30298 so fortunatly I can fix by writting code like this:

julia> Matrix{MP{Float64}}(0I, 2, 2)
2×2 Array{MP{Float64},2}:
  0.0  -Inf
 -Inf   0.0

Is it because I use Bool that I have this behavior ? And how to replace code 0I by one(MP{Float64})I ? I really think that Julia shall implement I calling one() and without giving Type::T calls one(Bool).

I'm starting learning julia! Thx!