Type inference on an array of arrays

New to Julia
1

Hello, could you please help me to understand the following code. I think the types line up really nicely, but Julia 1.2 thinks otherwise.

d = 2
a1 = [0.0]
a2 = [0.0]
b1 = zeros(d, 1)
b2 = zeros(d, 1)
c1 = [0.0]
c2 = [0.0]

function test1(a::AbstractVector{AbstractVector{T}},
               b::AbstractVector{AbstractMatrix{T}},
               c::AbstractVector{AbstractVector{T}}
               ) where T <: AbstractFloat
    return "Done!"
end

This seems fine:

methods(test1)

1 method for generic function “test1”:

[1] test1(a::AbstractArray{AbstractArray{T,1},1}, b::AbstractArray{AbstractArray{T,2},1}, c::AbstractArray{AbstractArray{T,1},1}) where T<:AbstractFloat in Main at typeerror.jl:13

But this fails:

test1([a1, a2], [b1, b2], [c1, c2])

ERROR: MethodError: no method matching test1(::Array{Array{Float64,1},1}, ::Array{Array{Float64,2},1}, ::Array{Array{Float64,1},1})

Stacktrace:

[1] top-level scope at REPL[3]:1

2

Have you read through this part in the manual Types · The Julia Language? Especially the point about

This last point is very important: even though Float64 <: Real we DO NOT have Point{Float64} <: Point{Real} .

1 Like
3

OK, I was not paying attention. Thank you!

TL;DR:

function test1(a::AbstractVector{TV},
               b::AbstractVector{TM},
               c::AbstractVector{TV}
               ) where {TV <: AbstractVector{T}, TM <: AbstractMatrix{T}} where T <: AbstractFloat
    return "Done!"
end