CODEWITHSUNDEEP
typesjuliatype-inference
mohamedmoussa
mohamedmoussa

Reputation: 559

Unexpected type instability in Julia function

I think I am getting a type instability in my code, but I'm not sure why. Consider the following code.

module test
    mutable struct Data
        arr::Array{Int64, 1}
        Data(n) = new(zeros(n))
    end

    function main()
        data = Data(5)
        data.arr
    end

    println(main())
    @code_warntype main()
end

The output of @code_warntype main() starts with the following three lines

Variables:
  #self#::test.#main
  data::Any
        ^^^

Maybe I am not interpreting the output of @code_warntype correctly, but it seems that I have a type-instability for the data variable. Can someone shed some light on what is going on here? I'm using 0.6.2 if that helps.

Upvotes: 1

Views: 182

Answers (1)

Mark Birtwistle
Mark Birtwistle

Reputation: 372

This is not really an answer, but it may shed some light on the problem. If I rewrite your code as a 2-dimensional array, all the type instabilities seem to disappear and the native code is minimised:

module test
    mutable struct Data
        arr::Array{Int64,2}
        Data(n::Int64, m::Int64=1) = new(zeros(Int64, n, m))
    end

    function main()
        data = Data(5)
        data.arr
    end

    println(main())
    @code_warntype main()
end

The complete output in this case is as follows:

[0; 0; 0; 0; 0]
Variables:
  #self# <optimized out>
  data::test.Data

Body:
  begin 
      data::test.Data = $(Expr(:invoke, MethodInstance for test.Data(::Int64, ::Int64), :(test.Data), 5, 1)) # line 37:
      return (Core.getfield)(data::test.Data, :arr)::Array{Int64,2}
  end::Array{Int64,2}

All references to Any now seem to have miraculously disappeared from the output. I have no idea why!

Upvotes: 1

Related Questions