Release 0.10.1 broke pairwise haversine

[briochemc]

The latest release (v0.10.1) broke a simple pairwise call with the Haversine distance that used to work with (v0.10.0):

pairwise(Haversine(1), rand(3,2), dims=1)

I'm not sure what changed, but I guess something went wrong, and the example above could be added in the test suite to ensure this does not happen again?

Below are complete MWEs:

• using v0.10.0:

  (@v1.5) pkg> activate --temp
   Activating new environment at `/var/folders/75/bdw1v8j10w9dbmvjmc13fkj80000gn/T/jl_1MvJRQ/Project.toml`
  
  (jl_1MvJRQ) pkg> add Distances@v0.10.0
     Updating registry at `~/.julia/registries/General`
     Updating git-repo `https://github.com/JuliaRegistries/General.git`
    Resolving package versions...
  Updating `/private/var/folders/75/bdw1v8j10w9dbmvjmc13fkj80000gn/T/jl_1MvJRQ/Project.toml`
    [b4f34e82] + Distances v0.10.0
  Updating `/private/var/folders/75/bdw1v8j10w9dbmvjmc13fkj80000gn/T/jl_1MvJRQ/Manifest.toml`
    [b4f34e82] + Distances v0.10.0
    [8f399da3] + Libdl
    [37e2e46d] + LinearAlgebra
    [9a3f8284] + Random
    [9e88b42a] + Serialization
    [2f01184e] + SparseArrays
    [10745b16] + Statistics
  
  julia> using Distances
  
  julia> pairwise(Haversine(1), rand(3,2), dims=1)
  3×3 Array{Float64,2}:
   0.0         0.00649202  0.00718727
   0.00649202  0.0         0.0129741
   0.00718727  0.0129741   0.0

• and using v0.10.1:

  (@v1.5) pkg> activate --temp
   Activating new environment at `/var/folders/75/bdw1v8j10w9dbmvjmc13fkj80000gn/T/jl_kXGxKf/Project.toml`
  
  (jl_kXGxKf) pkg> add Distances@v0.10.1
     Updating registry at `~/.julia/registries/General`
     Updating git-repo `https://github.com/JuliaRegistries/General.git`
    Resolving package versions...
  Updating `/private/var/folders/75/bdw1v8j10w9dbmvjmc13fkj80000gn/T/jl_kXGxKf/Project.toml`
    [b4f34e82] + Distances v0.10.1
  Updating `/private/var/folders/75/bdw1v8j10w9dbmvjmc13fkj80000gn/T/jl_kXGxKf/Manifest.toml`
    [b4f34e82] + Distances v0.10.1
    [8f399da3] + Libdl
    [37e2e46d] + LinearAlgebra
    [9a3f8284] + Random
    [9e88b42a] + Serialization
    [2f01184e] + SparseArrays
    [10745b16] + Statistics
  
  julia> using Distances
  
  julia> pairwise(Haversine(1), rand(3,2), dims=1)
  ERROR: ArgumentError: expected both inputs to have length 2 in Haversine{Int64}(1) distance
  Stacktrace:
  [1] haversine_error(dist::Haversine{Int64})
    @ Distances ~/.julia/packages/Distances/glUtY/src/haversine.jl:34
  
  [2] Haversine(x::Float64, y::Float64)
    @ Distances ~/.julia/packages/Distances/glUtY/src/haversine.jl:17
  
  [3] result_type(f::Haversine{Int64}, a::Type{T} where T, b::Type{T} where T)
    @ Distances ~/.julia/packages/Distances/glUtY/src/generic.jl:36
  
  [4] result_type(dist::Haversine{Int64}, a::Array{Float64,2}, b::Array{Float64,2})
    @ Distances ~/.julia/packages/Distances/glUtY/src/generic.jl:35
  
  [5] #pairwise#5(dims::Int64, ::typeof(pairwise), metric::Haversine{Int64}, a::Array{Float64,2})
    @ Distances ~/.julia/packages/Distances/glUtY/src/generic.jl:328
  
  [6] top-level scope
    @ REPL[7]:1

[briochemc]

Not sure, but it looks like the issue is with the new generic return_type function, which I think assumes that any distance can work in 1D space, i.e., take scalars like dist(a::Number, b::Number), while haversine always expects 2-element arrays that live in a 2D space?

[dkarrasch]

😭 I'm very sorry! Apparently, we don't have any pairwise tests for Haversine, so this didn't come up in the process. This is closely related to #209 in that this used to be handled by the Float64 fallback.

[briochemc]

No worries! I had a quick look at the code and just tried to decipher the problem to help you speed up the recovery process! :)