testvalue stackoverflows when called on LKJCholesky

[sethaxen]

Currently this happens.

julia> testvalue(LKJCholesky(10, 2.0))
ERROR: StackOverflowError:
Stacktrace:
 [1] testvalue(μ::Lebesgue{()}) (repeats 79984 times)
   @ MeasureTheory ~/.julia/packages/MeasureTheory/cKsEt/src/utils.jl:38

This causes any Soss models that use LKJCholesky to fail.

[sethaxen]

It seems this is due to at least two separate issues. First, testvalue is not implemented for several primitive measures, but when the default implementation hits a primitive measure, it stackoverflows. Second, we have this:

julia> d = Lebesgue(ℝ)^5
For(Lebesgue, Fill((), 5))

julia> marginals(d)
5-element mappedarray(Lebesgue, ::FillArrays.Fill{Tuple{}, 1, Tuple{Base.OneTo{Int64}}}) with eltype Lebesgue:
 Lebesgue(())
 Lebesgue(())
 Lebesgue(())
 Lebesgue(())
 Lebesgue(())

I'm guessing each marginal should be a Lebesgue(ℝ)? Because the default implementation of testvalue maps over the marginals, it won't know what to do when it hits a Lebesgue(()).

[cscherrer]

Thanks @sethaxen ! Yes, this is a bug.

[cscherrer]

In the testvalues branch,

julia> using MeasureTheory

julia> testvalue(LKJCholesky(10, 2.0))
LinearAlgebra.Cholesky{Float64, LinearAlgebra.UpperTriangular{Float64, Matrix{Float64}}}
U factor:
10×10 LinearAlgebra.UpperTriangular{Float64, Matrix{Float64}}:
 1.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0
  ⋅   1.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0
  ⋅    ⋅   1.0  0.0  0.0  0.0  0.0  0.0  0.0  0.0
  ⋅    ⋅    ⋅   1.0  0.0  0.0  0.0  0.0  0.0  0.0
  ⋅    ⋅    ⋅    ⋅   1.0  0.0  0.0  0.0  0.0  0.0
  ⋅    ⋅    ⋅    ⋅    ⋅   1.0  0.0  0.0  0.0  0.0
  ⋅    ⋅    ⋅    ⋅    ⋅    ⋅   1.0  0.0  0.0  0.0
  ⋅    ⋅    ⋅    ⋅    ⋅    ⋅    ⋅   1.0  0.0  0.0
  ⋅    ⋅    ⋅    ⋅    ⋅    ⋅    ⋅    ⋅   1.0  0.0
  ⋅    ⋅    ⋅    ⋅    ⋅    ⋅    ⋅    ⋅    ⋅   1.0

I've linked that PR so merging will close this