diff --git a/src/compiler.jl b/src/compiler.jl
index bf7e62cc4a..5bec0dd255 100644
--- a/src/compiler.jl
+++ b/src/compiler.jl
@@ -6740,19 +6740,31 @@ function _thunk(job, postopt::Bool = true)::Tuple{LLVM.Module, Vector{Any}, Stri
         mstr = if job.config.params.ABI <: InlineABI
             ""
         else
+            fixup_callconv!(mod, JIT.get_tm())
+            for f in functions(mod)
+                for i in 1:length(parameters(f))
+                    for a in collect(parameter_attributes(f, i))
+                       if kind(a) == "enzyme_sret"
+                           API.EnzymeDumpValueRef(f)
+                       end
+                       @assert kind(a) != "enzyme_sret"
+                       @assert kind(a) != "enzyme_sret_v"
+                    end
+                end
+            end
             string(mod)
         end
         if job.config.params.ABI <: FFIABI || job.config.params.ABI <: NonGenABI
             if DumpPrePostOpt[]
                 API.EnzymeDumpModuleRef(mod.ref)
             end
-            post_optimize!(mod, JIT.get_tm())
+            post_optimize!(mod, JIT.get_tm(); callconv=false)
             if DumpPostOpt[]
                 API.EnzymeDumpModuleRef(mod.ref)
             end
         else
             propagate_returned!(mod)
-	    Compiler.JIT.prepare!(mod)
+            Compiler.JIT.prepare!(mod)
         end
         mstr
     else
diff --git a/src/compiler/optimize.jl b/src/compiler/optimize.jl
index 713e8440ad..cc2903f807 100644
--- a/src/compiler/optimize.jl
+++ b/src/compiler/optimize.jl
@@ -377,7 +377,7 @@ end
 const DumpPreCallConv = Ref(false)
 const DumpPostCallConv = Ref(false)
 
-function post_optimize!(mod::LLVM.Module, tm::LLVM.TargetMachine, machine::Bool = true)
+function fixup_callconv!(mod::LLVM.Module, tm::LLVM.TargetMachine)
     addr13NoAlias(mod)
     
     removeDeadArgs!(mod, tm, #=post_gc_fixup=#false)
@@ -430,6 +430,13 @@ function post_optimize!(mod::LLVM.Module, tm::LLVM.TargetMachine, machine::Bool
             ),
         )
     end
+    return
+end
+
+function post_optimize!(mod::LLVM.Module, tm::LLVM.TargetMachine, machine::Bool = true; callconv::Bool = true)
+    if callconv
+        fixup_callconv!(mod, tm)
+    end
     @dispose pb = NewPMPassBuilder() begin
         registerEnzymeAndPassPipeline!(pb)
         register!(pb, ReinsertGCMarkerPass())
diff --git a/src/utils.jl b/src/utils.jl
index c0ca94de7e..be75f0ef12 100644
--- a/src/utils.jl
+++ b/src/utils.jl
@@ -562,11 +562,11 @@ function sret_ty(fn::LLVM.Function, idx::Int)::LLVM.LLVMType
         end
 
         if ekind == "enzyme_sret"
-	    ety = parse(UInt, LLVM.value(attr))
-	    ety = Base.reinterpret(LLVM.API.LLVMTypeRef, ety)
-	    ety = LLVM.LLVMType(ety)
+            ety = parse(UInt, LLVM.value(attr))
+            ety = Base.reinterpret(LLVM.API.LLVMTypeRef, ety)
+            ety = LLVM.LLVMType(ety)
             if !LLVM.is_opaque(vt)
-		@assert ety == eltype(vt)
+                @assert ety == eltype(vt) "Mismatched sret type $(string(fn))\nidx=$idx\nety ($(string(ety))) != eltype(vt) (vt = $(string(vt)))"
             end
         
             return ety
diff --git a/test/ext/staticarrays.jl b/test/ext/staticarrays.jl
index c2c55a9aa4..8d8a501935 100644
--- a/test/ext/staticarrays.jl
+++ b/test/ext/staticarrays.jl
@@ -89,4 +89,21 @@ end
     @test res ≈ [1.0, 0.0]
     res = Enzyme.gradient(Enzyme.Forward, unstable_fun, inp)[1]
     @test res ≈ [1.0, 0.0]
-end
\ No newline at end of file
+end
+
+function inner_forhess(x)
+    return tanh.(x)
+end
+
+function for_hess(x)
+    return sum(inner_forhess(x))
+end
+
+grad_forhess(x) = autodiff(Reverse, for_hess, Active, Active(x))[1][1]
+hess(x) = jacobian(Forward, grad_forhess, x)[1]
+
+@testset "StaticArrays hessian" begin
+    x = @SVector zeros(10)
+    res = [-2.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 -2.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 -2.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 -2.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 -2.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 -2.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 -2.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -2.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -2.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -2.0]
+    @test jacobian(Forward, grad_forhess, x)[1] ≈ res
+end
diff --git a/test/hessian_mwe.jl b/test/hessian_mwe.jl
new file mode 100644
index 0000000000..8d3016711b
--- /dev/null
+++ b/test/hessian_mwe.jl
@@ -0,0 +1,56 @@
+#!/usr/bin/env julia
+# Minimal Working Example (MWE) for testing enzyme_sret type handling
+# This file isolates the core functionality being tested to help diagnose timeouts
+
+using Enzyme
+using Test
+using LinearAlgebra
+
+println("Testing Forward-over-Reverse (Hessian) computation...")
+println("This test exercises sret type handling in nested AD")
+
+@testset "Hessian MWE" begin
+    # Function: f(x) = x[1]^2 + x[1]*x[2]
+    # Gradient: ∇f = [2*x[1] + x[2], x[1]]
+    # Hessian: H = [[2, 1], [1, 0]]
+    
+    function origf(x::Array{Float64}, y::Array{Float64})
+        y[1] = x[1] * x[1] + x[2] * x[1]
+        return nothing
+    end
+
+    function grad(x, dx, y, dy)
+        Enzyme.autodiff(Reverse, Const(origf), Duplicated(x, dx), DuplicatedNoNeed(y, dy))
+        nothing
+    end
+
+    x = [2.0, 2.0]
+    y = Vector{Float64}(undef, 1)
+    dx = [0.0, 0.0]
+    dy = [1.0]
+
+    grad(x, dx, y, dy)
+
+    vx = ([1.0, 0.0], [0.0, 1.0])
+    hess = ([0.0, 0.0], [0.0, 0.0])
+    dx2 = [0.0, 0.0]
+    dy = [1.0]
+
+    Enzyme.autodiff(
+        Enzyme.Forward, grad,
+        Enzyme.BatchDuplicated(x, vx),
+        Enzyme.BatchDuplicated(dx2, hess),
+        Const(y),
+        Const(dy)
+    )
+
+    @test dx ≈ dx2
+    @test hess[1][1] ≈ 2.0
+    @test hess[1][2] ≈ 1.0
+    @test hess[2][1] ≈ 1.0
+    @test hess[2][2] ≈ 0.0
+end
+
+println("\nTest completed successfully!")
+println("If this test hangs or times out, the issue is with Forward-over-Reverse AD")
+println("with BatchDuplicated arguments, which requires proper enzyme_sret handling.")