Failed type assertion in `pairwise`

[yha]

using HypothesisTests, StatsBase
function test(x)
	pairwise(eachindex(x); symmetric=true) do j,i
		pval = pvalue(SignedRankTest(x[i], x[j]))
		pval, rand()
	end
end

x = [rand(1000) for _=1:10]

test(x)

ERROR: TypeError: in typeassert, expected Matrix{<:Tuple{Union{Nothing, Float64, Int64}, Float64}}, got a value of type Matrix{Tuple{Real, Float64}}
Stacktrace:
 [1] _pairwise(#unused#::Val{:none}, f::var"#29#30"{Vector{Vector{Float64}}}, x::Base.OneTo{Int64}, y::Base.OneTo{Int64}, symmetric::Bool)
   @ StatsBase C:\Users\sternlab\.julia\packages\StatsBase\pJqvO\src\pairwise.jl:155
 [2] pairwise(f::Function, x::Base.OneTo{Int64}, y::Base.OneTo{Int64}; symmetric::Bool, skipmissing::Symbol)
   @ StatsBase C:\Users\sternlab\.julia\packages\StatsBase\pJqvO\src\pairwise.jl:288
 [3] test(x::Vector{Vector{Float64}})
   @ Main .\Untitled-9:3
 [4] top-level scope
   @ Untitled-9:11

Package versions:

  [09f84164] HypothesisTests v0.10.6
  [2913bbd2] StatsBase v0.33.16

(this doesn't happen if the pairwise call is in global scope rather than a function)

[nalimilan]

Good catch. This is a very tricky case indeed. #772 fixes this. The fact that inference gives such an imprecise type in the first place could also be improved in HypothesisTests, see JuliaStats/HypothesisTests.jl#263.