Fix eachvariate with zero variates

Project.toml

@@ -1,7 +1,7 @@
 name = "Distributions"
 uuid = "31c24e10-a181-5473-b8eb-7969acd0382f"
 authors = ["JuliaStats"]
-version = "0.25.104"
+version = "0.25.105"
 
 [deps]
 ChainRulesCore = "d360d2e6-b24c-11e9-a2a3-2a2ae2dbcce4"

src/eachvariate.jl

@@ -7,7 +7,7 @@ end
 
 function EachVariate{V}(x::AbstractArray{<:Real,M}) where {V,M}
     ax = ntuple(i -> axes(x, i + V), Val(M - V))
-    T = typeof(view(x, ntuple(i -> i <= V ? Colon() : firstindex(x, i), Val(M))...))
+    T = Base.promote_op(view, typeof(x), ntuple(i -> i <= V ? Colon : eltype(axes(x, i)), Val(M))...)
     return EachVariate{V,typeof(x),typeof(ax),T,M-V}(x, ax)
 end
 

test/eachvariate.jl

@@ -1,17 +1,44 @@
+using Distributions
 using ChainRulesTestUtils
 using ChainRulesTestUtils: FiniteDifferences
 
+using Random
+using Test
+
 # Without this, `to_vec` will also include the `axes` field of `EachVariate`.
 function FiniteDifferences.to_vec(xs::Distributions.EachVariate{V}) where {V}
     vals, vals_from_vec = FiniteDifferences.to_vec(xs.parent)
     return vals, x -> Distributions.EachVariate{V}(vals_from_vec(x))
 end
 
+# MWE in #1817
+struct FooEachvariate <: Sampleable{Multivariate, Continuous} end
+Base.length(::FooEachvariate) = 3
+Base.eltype(::FooEachvariate) = Float64
+function Distributions._rand!(rng::AbstractRNG, ::FooEachvariate, x::AbstractVector{<:Real})
+    return rand!(rng, x)
+end
+
 @testset "eachvariate.jl" begin
     @testset "ChainRules" begin
         xs = randn(2, 3, 4, 5)
         test_rrule(Distributions.EachVariate{1}, xs)
         test_rrule(Distributions.EachVariate{2}, xs)
         test_rrule(Distributions.EachVariate{3}, xs)
     end
+
+    @testset "No variates (#1817)" begin
+        @test size(Distributions.eachvariate(rand(0), Univariate)) == (0,)
+        @test size(Distributions.eachvariate(rand(3, 0, 1), Univariate)) == (3, 0, 1)
+        @test size(Distributions.eachvariate(rand(3, 2, 0), Univariate)) == (3, 2, 0)
+
+        @test size(Distributions.eachvariate(rand(4, 0), Multivariate)) == (0,)
+        @test size(Distributions.eachvariate(rand(4, 0, 2), Multivariate)) == (0, 2)
+        @test size(Distributions.eachvariate(rand(4, 5, 0), Multivariate)) == (5, 0)
+        @test size(Distributions.eachvariate(rand(4, 5, 0, 2), Multivariate)) == (5, 0, 2)
+
+        draws = @inferred(rand(FooEachvariate(), 0))
+        @test draws isa Matrix{Float64}
+        @test size(draws) == (3, 0)
+    end
 end