1.12 Integration CI

[wsmoses]

lets see what breaks

[wsmoses]

@penelopeysm the turing test fails for reasons unrealted to enzyme, can you take a look?

[penelopeysm]

I think this should fix it

diff --git a/test/integration/Turing/runtests.jl b/test/integration/Turing/runtests.jl
index 72d4137d..da114da7 100644
--- a/test/integration/Turing/runtests.jl
+++ b/test/integration/Turing/runtests.jl
@@ -17,7 +17,13 @@ adtypes = (
 @model function assume_normal()
     a ~ Normal()
 end
-dppl_lda = begin
+@model function dppl_lda(k, m, w, doc, alpha, beta)
+    theta ~ product_distribution(fill(Dirichlet(alpha), m))
+    phi ~ product_distribution(fill(Dirichlet(beta), k))
+    log_phi_dot_theta = log.(phi * theta)
+    @addlogprob! sum(log_phi_dot_theta[CartesianIndex.(w, doc)])
+end
+dppl_lda_model = begin
     v = 100      # words
     k = 5        # topics
     m = 10       # number of docs
@@ -37,18 +43,12 @@ dppl_lda = begin
             doc[idx + j] = i
         end
     end
-    @model function dppl_lda(k, m, w, doc, alpha, beta)
-        theta ~ product_distribution(fill(Dirichlet(alpha), m))
-        phi ~ product_distribution(fill(Dirichlet(beta), k))
-        log_phi_dot_theta = log.(phi * theta)
-        @addlogprob! sum(log_phi_dot_theta[CartesianIndex.(w, doc)])
-    end
-    dppl_lda
+    dppl_lda(k, m, w, doc, alpha, beta)
 end
 MODELS = [
     DynamicPPL.TestUtils.ALL_MODELS...,
     assume_normal(),
-    dppl_lda(k, m, w, doc, alpha, beta),
+    dppl_lda_model,
 ]
 
 @testset "AD on logdensity" begin

[wsmoses]

@penelopeysm can you open a PR to enzyme with your patch? ill rebase this atop that

[wsmoses]

@copilot isolate an mwe of the distributions failure and open an issue

[Copilot]

@copilot isolate an mwe of the distributions failure and open an issue

I isolated an MWE for the Distributions failure on Julia 1.12:

using Distributions, Enzyme, LinearAlgebra, PDMats, StableRNGs, Random
rng = StableRNG(23)
_sym(A) = A' * A
_pdmat(A) = PDMat(_sym(A) + 5I)
d = Wishart(5, _pdmat(randn(rng, 3, 3)))
x = Symmetric(collect(_pdmat(randn(rng, 3, 3))))
f = y -> logpdf(d, y)
Enzyme.gradient(Enzyme.Reverse, Enzyme.Const(f), x)

This reproduces EnzymeNoTypeError in LinearAlgebra.ldiv(::Cholesky, ::Symmetric) through Distributions/src/matrix/wishart.jl:189.

I can’t open a GitHub issue from this environment because write auth is unavailable, but this is ready to paste as an issue report.

[wsmoses]

@kshyatt matrixalgebrakit fails 1.12 ci with:

lq: Error During Test at /__w/Enzyme.jl/Enzyme.jl/test/integration/MatrixAlgebraKit/runtests.jl:20
  Got exception outside of a @test
  UndefVarError: `diagview` not defined in `Main`
  Hint: It looks like two or more modules export different bindings with this name, resulting in ambiguity. Try explicitly importing it from a particular module, or qualifying the name with the module it should come from.
  Hint: a global variable of this name also exists in MatrixAlgebraKit.
  Hint: a global variable of this name also exists in LinearAlgebra.
  Stacktrace:
    [1] structured_randn!(A::Diagonal{Float64, Vector{Float64}})
      @ Main /__w/Enzyme.jl/Enzyme.jl/test/integration/MatrixAlgebraKit/runtests.jl:15
    [2] _broadcast_getindex_evalf
      @ ./broadcast.jl:699 [inlined]
    [3] _broadcast_getindex
      @ ./broadcast.jl:672 [inlined]
    [4] #copy##0
      @ ./broadcast.jl:1124 [inlined]
    [5] ntuple
      @ ./ntuple.jl:51 [inlined]
    [6] copy
      @ ./broadcast.jl:1124 [inlined]
    [7] materialize(bc::Base.Broadcast.Broadcasted{Base.Broadcast.Style{Tuple}, Nothing, typeof(structured_randn!), Tuple{Base.Broadcast.Broadcasted{Base.Broadcast.Style{Tuple}, Nothing, typeof(similar), Tuple{Tuple{Diagonal{Float64, Vector{Float64}}, Diagonal{Float64, Vector{Float64}}}}}}})
      @ Base.Broadcast ./broadcast.jl:894
    [8] macro expansion
      @ /__w/Enzyme.jl/Enzyme.jl/test/integration/MatrixAlgebraKit/runtests.jl:36 [inlined]
    [9] macro expansion
      @ /__w/_tool/julia/1.12.6/x64/share/julia/stdlib/v1.12/Test/src/Test.jl:1866 [inlined]
   [10] top-level scope
      @ /__w/Enzyme.jl/Enzyme.jl/test/integration/MatrixAlgebraKit/runtests.jl:1889
   [11] include(mod::Module, _path::String)
      @ Base ./Base.jl:306
   [12] exec_options(opts::Base.JLOptions)
      @ Base ./client.jl:317
   [13] _start()
      @ Base ./client.jl:550

can you fix?

[kshyatt]

@wsmoses you mind if I just push directly to this branch?

[wsmoses]

sure, but this branch won't merge til all of 1.12 pass, so you'll need to land your actual fix in a different pr once complete [then we rebase and continue this]

[kshyatt]

in that case ill just do it separately

[wsmoses]

cc @oscardssmith re molly mwe