fit/fit_mle for DiagNormal/IsoNormal fails: "suffstats is not implemented"

[rsenne]

Fitting a diagonal multivariate normal (DiagNormal/IsoNormal) via fit/fit_mle errors because suffstats is not defined for that type. Both the weighted and unweighted forms fail, and full-covariance MvNormal works fine.

MWE

using Distributions, LinearAlgebra

d = MvNormal(zeros(3), Diagonal(ones(3)))   # a DiagNormal
# d = MvNormal(zeros(3), I) # an IsoNormal
typeof(d) 

fit(typeof(d), randn(3, 50))                 # ERROR

Error

ERROR: suffstats is not implemented for (IsoNormal{Float64, PDMats.ScalMat{Float64}, Vector{Float64}}, Matrix{Float64}).
Stacktrace:
 [1] error(s::String)
   @ Base .\error.jl:44
 [2] suffstats(dt::Type{IsoNormal{Float64, PDMats.ScalMat{Float64}, Vector{Float64}}}, xs::Matrix{Float64})
   @ Distributions C:\Users\senne\.julia\packages\Distributions\XAng5\src\genericfit.jl:5
 [3] fit_mle(dt::Type{IsoNormal{Float64, PDMats.ScalMat{Float64}, Vector{Float64}}}, x::Matrix{Float64})
   @ Distributions C:\Users\senne\.julia\packages\Distributions\XAng5\src\genericfit.jl:30
 [4] fit(dt::Type{IsoNormal{Float64, PDMats.ScalMat{Float64}, Vector{Float64}}}, x::Matrix{Float64})
   @ Distributions C:\Users\senne\.julia\packages\Distributions\XAng5\src\genericfit.jl:46
 [5] top-level scope
   @ REPL[12]:1

Version Info

Julia: 1.12.6
Distributions 0.25.129

[rsenne]

For additional info I've confirmed this was introduced in v0.25.129. All preceding versions pass

[rsenne]

Okay I figured out the issue. In the latest version the following were introduced:

const IsoNormal{T}  = MvNormal{T,<:ScalMat{T},<:AbstractVector{T}}
const DiagNormal{T} = MvNormal{T,<:PDiagMat{T},<:AbstractVector{T}}
const FullNormal{T} = MvNormal{T,<:PDMat{T},<:AbstractVector{T}}

Because these are now UnionAlls rather than concrete DataTypes, the invariant D::Type{IsoNormal} / Type{DiagNormal} / Type{FullNormal} signatures on suffstats/fit_mle only match the bare alias object and miss every instantiated or concrete member.
I think the fix is simple--widening the type on the respective functions e.g., D::Type{<:IsoNormal} . I have confirmed empirically--will start a PR and add tests against this regression as well.