Fix #182

src/InfiniteArrays.jl

@@ -49,7 +49,7 @@ import LinearAlgebra: AdjOrTrans, HermOrSym, diag, norm, norm1, norm2, normp
 
 import LazyArrays: AbstractPaddedLayout
 
-import Random: default_rng
+import Random: default_rng, seed!
 
 export ∞, ℵ₀, Hcat, Vcat, Zeros, Ones, Fill, Eye, BroadcastArray, cache
 import Base: unitrange, oneto

src/infrand.jl

@@ -1,10 +1,24 @@
+_dist_type(dist) = typeof(dist)
+_dist_type(dist::Type{T}) where {T} = Type{T} # avoid DataType for typeof(Float64), since DataType means rand(seq.rng, seq.dist) is inferred as Any
+function _reset_seed!(rng)
+    new_seed = rand(rng, UInt32)
+    return seed!(rng, new_seed)
+end
+
 """
     InfRandVector([rng=default_rng()], [dist=Float64])
 
 Represents a random infinite sequence. The random number generator can be specified 
 by the first argument `rng`, which defaults to `Random.default_rng()`, and the distribution 
 to generate from can be specified using the `dist` argument, which defaults to `Float64`.
 
+!!! note "Mutation of rng"
+
+    When an InfRandVector is constructed, the `rng` (or, if 
+    no `rng` is provided, the global `rng`) gets reseeded
+    using `Random.seed!(rng, rand(rng, UInt32))`.
+
+# Examples
 ```julia-repl
 julia> using InfiniteArrays
 
@@ -41,14 +55,19 @@ mutable struct InfRandVector{T,D,RNG} <: AbstractCachedVector{T}
     const data::Vector{T}
     datasize::Int
 end
+
 function InfRandVector(rng=default_rng(), dist=Float64)
     T = typeof(rand(copy(rng), dist))
     _rng = copy(rng)
-    return InfRandVector{T,typeof(dist),typeof(_rng)}(_rng, dist, T[], 0)
+    _reset_seed!(rng)
+    return InfRandVector{T,_dist_type(dist),typeof(_rng)}(_rng, dist, T[], 0)
 end
+
 Base.size(::InfRandVector) = (ℵ₀,)
 Base.axes(::InfRandVector) = (1:ℵ₀,)
 Base.length(::InfRandVector) = ℵ₀
+
+@inline _single_rand(seq::InfRandVector) = rand(seq.rng, seq.dist)
 function resizedata!(seq::InfRandVector, inds)
     newlen = maximum(inds)
     curlen = length(seq.data)
@@ -57,7 +76,7 @@ function resizedata!(seq::InfRandVector, inds)
     # rand!(seq.rng, view(seq.data, curlen+1:newlen), seq.dist)
     # ^ rand() is not actually sequential.. rand(Random.seed!(123), 1000) ≠ (rng = Random.seed!(123); [rand(rng) for _ in 1:1000])
     for i in (curlen+1):newlen
-        seq.data[i] = rand(seq.rng, seq.dist)
+        seq.data[i] = _single_rand(seq)
     end
     seq.datasize = newlen
     return seq

test/test_infrand.jl

@@ -13,13 +13,15 @@
         @test size(seq) == (ℵ₀,)
         @test length(seq) == ℵ₀
         @test axes(seq) == (1:ℵ₀,)
+        @inferred InfiniteArrays._single_rand(seq)
     end
 
     @testset "Providing an RNG and a distribution" begin
         rng = MersenneTwister(123)
         seq = InfiniteArrays.InfRandVector(rng, Float16)
         rng2 = MersenneTwister(123)
         @test seq[1:10000] == [rand(rng2, Float16) for _ in 1:10000]
+        @inferred InfiniteArrays._single_rand(seq)
     end 
 
     @testset "Distributions.jl" begin 
@@ -28,5 +30,18 @@
         seq = InfiniteArrays.InfRandVector(rng, dist)
         rng2 = Xoshiro(5)
         @test seq[1:100] == [0.3 + 1.7randn(rng2) for _ in 1:100]
+
+        @test InfiniteArrays._dist_type(dist) == Normal{Float64}
+        @test InfiniteArrays._dist_type(Float64) == Type{Float64}
+        @inferred InfiniteArrays._single_rand(seq)
+    end
+
+    @testset "Issue #182" begin
+        kp = InfiniteArrays.InfRandVector()[1:1000]
+        kp2 = InfiniteArrays.InfRandVector()[1:1000]
+        kp3 = InfiniteArrays.InfRandVector()[1:1000]
+        @test kp ≠ kp2 
+        @test kp2 ≠ kp3 
+        @test kp ≠ kp3
     end
 end