Zygote broken on 1.12

[marcobonici]

Describe the bug 🐞

When differentiating an IntegralProblem, everything works fine on 1.12 and 1.11, but Zygote gives segmentation fault for 1.12.

Expected behavior

I expect the differentiation to work as in the previous versions of Julia.

Minimal Reproducible Example 👇

using Integrals, ForwardDiff, FiniteDiff, Zygote, Cuba
f(x, p) = sum(sin.(x .* p))
domain = (ones(2), 3ones(2)) # (lb, ub)
p = ones(2)

function testf(p)
    prob = IntegralProblem(f, domain, p)
    sin(solve(prob, CubaCuhre(), reltol = 1e-6, abstol = 1e-6)[1])
end
testf(p)
dp1 = Zygote.gradient(testf, p)
dp2 = FiniteDiff.finite_difference_gradient(testf, p)
dp3 = ForwardDiff.gradient(testf, p)
dp1[1] ≈ dp2 ≈ dp3

Error & Stacktrace ⚠️

[80341] signal 11 (1): Segmentation fault
in expression starting at REPL[8]:1
emit_unboxed_coercion at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/intrinsics.cpp:394 [inlined]
emit_unbox at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/intrinsics.cpp:458
emit_condition at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/codegen.cpp:6066
emit_condition at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/codegen.cpp:6079 [inlined]
emit_function at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/codegen.cpp:9271
jl_emit_code at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/codegen.cpp:9727
jl_emit_codeinst at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/codegen.cpp:9798
jl_emit_codeinst_to_jit_impl at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/jitlayers.cpp:770
jl_add_codeinst_to_jit at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/gf.c:3372
add_codeinsts_to_jit! at ./../usr/share/julia/Compiler/src/typeinfer.jl:1436
typeinf_ext_toplevel at ./../usr/share/julia/Compiler/src/typeinfer.jl:1443 [inlined]
typeinf_ext_toplevel at ./../usr/share/julia/Compiler/src/typeinfer.jl:1451
jfptr_typeinf_ext_toplevel_117555.1 at /home/marcobonici/.julia/juliaup/julia-1.12.0+0.x64.linux.gnu/lib/julia/sys.so (unknown line)
jl_apply at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/julia.h:2387 [inlined]
jl_type_infer at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/gf.c:462
jl_compile_method_internal at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/gf.c:3512
_jl_invoke at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/gf.c:4002 [inlined]
ijl_apply_generic at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/gf.c:4210
jl_apply at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/julia.h:2387 [inlined]
jl_f__apply_iterate at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/builtins.c:868
jl_apply at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/julia.h:2387 [inlined]
jl_f__apply_iterate at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/builtins.c:868
_apply at ./boot.jl:1019
adjoint at /home/marcobonici/.julia/packages/Zygote/55SqB/src/lib/lib.jl:211 [inlined]
_pullback at /home/marcobonici/.julia/packages/ZygoteRules/CkVIK/src/adjoint.jl:67 [inlined]
IntegralProblem at /home/marcobonici/.julia/packages/SciMLBase/b4Q81/src/problems/integral_problems.jl:79 [inlined]
_pullback at /home/marcobonici/.julia/packages/Zygote/55SqB/src/compiler/interface2.jl:0
unknown function (ip: 0x7fbb2fccb3b2) at (unknown file)
jl_apply at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/julia.h:2387 [inlined]
jl_f__apply_iterate at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/builtins.c:868
jl_apply at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/julia.h:2387 [inlined]
jl_f__apply_iterate at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/builtins.c:868
_apply at ./boot.jl:1019
_pullback at /home/marcobonici/.julia/packages/Zygote/55SqB/src/compiler/interface2.jl:81
unknown function (ip: 0x7fbb2fccaf02) at (unknown file)
jl_apply at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/julia.h:2387 [inlined]
jl_f__apply_iterate at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/builtins.c:868
jl_apply at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/julia.h:2387 [inlined]
jl_f__apply_iterate at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/builtins.c:868
_apply at ./boot.jl:1019
adjoint at /home/marcobonici/.julia/packages/Zygote/55SqB/src/lib/lib.jl:211 [inlined]
_pullback at /home/marcobonici/.julia/packages/ZygoteRules/CkVIK/src/adjoint.jl:67 [inlined]
IntegralProblem at /home/marcobonici/.julia/packages/SciMLBase/b4Q81/src/problems/integral_problems.jl:76 [inlined]
_pullback at /home/marcobonici/.julia/packages/Zygote/55SqB/src/compiler/interface2.jl:0
unknown function (ip: 0x7fbb2fca1722) at (unknown file)
testf at ./REPL[6]:2 [inlined]
_pullback at /home/marcobonici/.julia/packages/Zygote/55SqB/src/compiler/interface2.jl:0
pullback at /home/marcobonici/.julia/packages/Zygote/55SqB/src/compiler/interface.jl:96
pullback at /home/marcobonici/.julia/packages/Zygote/55SqB/src/compiler/interface.jl:94 [inlined]
gradient at /home/marcobonici/.julia/packages/Zygote/55SqB/src/compiler/interface.jl:153
jfptr_gradient_19838 at /home/marcobonici/.julia/compiled/v1.12/Zygote/4kbLI_9sbRe.so (unknown line)
jl_apply at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/julia.h:2387 [inlined]
do_call at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/interpreter.c:123
eval_value at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/interpreter.c:243
eval_stmt_value at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/interpreter.c:194 [inlined]
eval_body at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/interpreter.c:707
jl_interpret_toplevel_thunk at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/interpreter.c:898
jl_toplevel_eval_flex at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/toplevel.c:1035
__repl_entry_eval_expanded_with_loc at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/usr/share/julia/stdlib/v1.12/REPL/src/REPL.jl:301
jl_apply at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/julia.h:2387 [inlined]
jl_f_invokelatest at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/builtins.c:881
toplevel_eval_with_hooks at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/usr/share/julia/stdlib/v1.12/REPL/src/REPL.jl:308
toplevel_eval_with_hooks at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/usr/share/julia/stdlib/v1.12/REPL/src/REPL.jl:312
toplevel_eval_with_hooks at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/usr/share/julia/stdlib/v1.12/REPL/src/REPL.jl:305 [inlined]
eval_user_input at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/usr/share/julia/stdlib/v1.12/REPL/src/REPL.jl:330
repl_backend_loop at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/usr/share/julia/stdlib/v1.12/REPL/src/REPL.jl:452
#start_repl_backend#41 at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/usr/share/julia/stdlib/v1.12/REPL/src/REPL.jl:427
start_repl_backend at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/usr/share/julia/stdlib/v1.12/REPL/src/REPL.jl:424 [inlined]
#run_repl#50 at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/usr/share/julia/stdlib/v1.12/REPL/src/REPL.jl:653
run_repl at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/usr/share/julia/stdlib/v1.12/REPL/src/REPL.jl:639
jfptr_run_repl_18494.1 at /home/marcobonici/.julia/juliaup/julia-1.12.0+0.x64.linux.gnu/share/julia/compiled/v1.12/REPL/u0gqU_g0Uq3.so (unknown line)
run_std_repl at ./client.jl:478
jfptr_run_std_repl_69973.1 at /home/marcobonici/.julia/juliaup/julia-1.12.0+0.x64.linux.gnu/lib/julia/sys.so (unknown line)
jl_apply at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/julia.h:2387 [inlined]
jl_f_invokelatest at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/builtins.c:881
run_main_repl at ./client.jl:499
repl_main at ./client.jl:586 [inlined]
_start at ./client.jl:561
jfptr__start_85044.1 at /home/marcobonici/.julia/juliaup/julia-1.12.0+0.x64.linux.gnu/lib/julia/sys.so (unknown line)
jl_apply at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/julia.h:2387 [inlined]
true_main at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/jlapi.c:971
jl_repl_entrypoint at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/src/jlapi.c:1139
main at /cache/build/builder-amdci4-0/julialang/julia-release-1-dot-12/cli/loader_exe.c:58
unknown function (ip: 0x7fbb4582a1c9) at /lib/x86_64-linux-gnu/libc.so.6
__libc_start_main at /lib/x86_64-linux-gnu/libc.so.6 (unknown line)
unknown function (ip: 0x4010b8) at /workspace/srcdir/glibc-2.17/csu/../sysdeps/x86_64/start.S
Allocations: 27724868 (Pool: 27724361; Big: 507); GC: 20
Segmentation fault (core dumped)

Environment (please complete the following information):

• Output of using Pkg; Pkg.status()

• Output of using Pkg; Pkg.status(; mode = PKGMODE_MANIFEST)

[d360d2e6] ChainRulesCore v1.26.0
  [8a292aeb] Cuba v2.3.0
  [6a86dc24] FiniteDiff v2.29.0
  [f6369f11] ForwardDiff v1.2.2
  [de52edbc] Integrals v4.7.0
  [0bca4576] SciMLBase v2.121.1
  [e88e6eb3] Zygote v0.7.10
  [8dfed614] Test

• Output of versioninfo()

versioninfo()
Julia Version 1.12.0
Commit b907bd0600f (2025-10-07 15:42 UTC)
Build Info:
  Official https://julialang.org release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 20 × 13th Gen Intel(R) Core(TM) i7-13700H
  WORD_SIZE: 64
  LLVM: libLLVM-18.1.7 (ORCJIT, alderlake)
  GC: Built with stock GC
Threads: 1 default, 1 interactive, 1 GC (on 20 virtual cores)
Environment:
  LD_PRELOAD = /usr/lib/x86_64-linux-gnu/libgomp.so.1
  LD_LIBRARY_PATH = /usr/local/cuda/lib64

[marcobonici]

Disclosure: tried to Claude my way out, he proposed a solution, by writing the following rrules. It works, the problem went away. The gradients also look correct, but I did not check for performance. If you ok @ChrisRackauckas , I will fork and push a PR (and check line by line what Claude wrote).

function ChainRulesCore.rrule(::Type{IntegralProblem}, f, domain, p=SciMLBase.NullParameters(); kwargs...)
    # Forward pass: create the IntegralProblem normally
    prob = IntegralProblem(f, domain, p; kwargs...)

    # Pullback function for reverse-mode AD
    function IntegralProblem_pullback(Δ)
        # Δ is the sensitivity from upstream (typically a Tangent for IntegralProblem)

        if Δ isa Tangent
            # Extract sensitivities from the Tangent
            # We use hasproperty to safely check for field existence
            ∂f = NoTangent()  # Function derivatives not supported
            ∂domain = hasproperty(Δ, :domain) ? Δ.domain : NoTangent()
            ∂p = hasproperty(Δ, :p) ? Δ.p : NoTangent()
        elseif Δ isa ChainRulesCore.ZeroTangent || Δ === nothing
            # No gradient to propagate
            ∂f = NoTangent()
            ∂domain = NoTangent()
            ∂p = NoTangent()
        else
            # Fallback for other types
            ∂f = NoTangent()
            ∂domain = NoTangent()
            ∂p = NoTangent()
        end

        # Return gradients for: Type, f, domain, p
        return (NoTangent(), ∂f, ∂domain, ∂p)
    end

    return prob, IntegralProblem_pullback
end

"""
    ChainRulesCore.rrule(::Type{IntegralProblem{iip}}, f, domain, p; kwargs...) where {iip}

Custom reverse-mode differentiation rule for type-parameterized IntegralProblem construction.

Handles the in-place/out-of-place parameterized constructor.
"""
function ChainRulesCore.rrule(::Type{IntegralProblem{iip}}, f, domain, p=SciMLBase.NullParameters(); kwargs...) where {iip}
    # Forward pass
    prob = IntegralProblem{iip}(f, domain, p; kwargs...)

    # Pullback function
    function IntegralProblem_iip_pullback(Δ)
        if Δ isa Tangent
            ∂f = NoTangent()
            ∂domain = hasproperty(Δ, :domain) ? Δ.domain : NoTangent()
            ∂p = hasproperty(Δ, :p) ? Δ.p : NoTangent()
        elseif Δ isa ChainRulesCore.ZeroTangent || Δ === nothing
            ∂f = NoTangent()
            ∂domain = NoTangent()
            ∂p = NoTangent()
        else
            ∂f = NoTangent()
            ∂domain = NoTangent()
            ∂p = NoTangent()
        end

        # Return gradients for: Type{IntegralProblem{iip}}, f, domain, p
        return (NoTangent(), ∂f, ∂domain, ∂p)
    end

    return prob, IntegralProblem_iip_pullback
end

"""
    ChainRulesCore.rrule(::typeof(getproperty), prob::IntegralProblem, field::Symbol)

Custom rule for accessing fields of IntegralProblem.

Ensures proper gradient propagation when accessing prob.p, prob.domain, etc.
"""
function ChainRulesCore.rrule(::typeof(getproperty), prob::IntegralProblem, field::Symbol)
    # Forward pass
    value = getproperty(prob, field)

    # Pullback
    function getproperty_IntegralProblem_pullback(Δ)
        # Create a Tangent with only the accessed field having a gradient
        if field == :f
            ∂prob = Tangent{typeof(prob)}(f = Δ)
        elseif field == :domain
            ∂prob = Tangent{typeof(prob)}(domain = Δ)
        elseif field == :p
            ∂prob = Tangent{typeof(prob)}(p = Δ)
        elseif field == :kwargs
            ∂prob = Tangent{typeof(prob)}(kwargs = Δ)
        else
            ∂prob = NoTangent()
        end

        return (NoTangent(), ∂prob, NoTangent())
    end

    return value, getproperty_IntegralProblem_pullback
end

[marcobonici]

Opened a related Pr, #282 , @ChrisRackauckas

[marcobonici]

As #282 was merged, this can be closed.