Incorrect estimation of the numeric type of the distance between Real elements: result is Int64, not Float64 or Real

[aborzunov]

Steps to reproduce:

julia> result_type(SqEuclidean(), Matrix{Real}(undef, 1, 1), Vector{Real}(undef, 1))
Int64

Julia: Version 1.11.1 (2024-10-16)
Distances: [b4f34e82] Distances v0.10.12

Issue originally was found during use of Clustering.jl: JuliaStats/Clustering.jl#285

[dkarrasch]

In the OP, why is the data of generic type Real? This is not an issue with Distances.jl, this behaviour is deeply rooted in julia itself:

julia> typeof(one(Real))
Int64

This is because one(::Type{T}) where {T<:Number} = convert(T,1). If you have floating point data anyway, it's better to use the most specific eltype possible. In your case, the elements of your arrays (in the OP) are all (but one) of type Float64. If, for some reason, you need to be more flexible, you could even use AbstractFloat:

julia> typeof(one(AbstractFloat))
Float64

or otherwise

convert(Array{Float64}, fs)