Type piracy of complex conversion of UniformScaling

[andreasvarga]

I am trying to define in my package MatrixEquations a simple conversion to complex type of an arbirary UniformScaling. For my first trial with

Base.complex(A::UniformScaling) = complex(A.λ)*I

Aqua complains of possible type piracy. The same occurs for two other definitions

Base.complex(A::UniformScaling{T}) where {T<:Real} = Complex{T}(A.λ)*I

Base.complex(A::UniformScaling) = Complex{eltype(A)}(A.λ)*I

I wonder what is the cause of this, because all three definitions work well for my purposes.

[hyrodium]

If you would like to avoid type piracies, your package needs to own function or argument type. The definition Base.complex(A::UniformScaling) = complex(A.λ)*I looks correct, so that should be defined in LinearAlgebra.jl.

The following articles would be useful to learn type piracies.

• https://docs.julialang.org/en/v1/manual/style-guide/#Avoid-type-piracy
• https://juliatesting.github.io/Aqua.jl/stable/piracies/
• https://discourse.julialang.org/t/how-bad-is-type-piracy-actually/37913

[lgoettgens]

Thanks for the explanation @hyrodium!

[andreasvarga]

Thanks for the explanation. I fixed the type piracy issue by defining my own function tocomplex as follows:

tocomplex(A::LinearAlgebra.UniformScaling) = complex(A.λ)*I
tocomplex(A::AbstractArray) = complex(A)
tocomplex(A::Number) = complex(A)