Segfault when using Enzyme as SymbolicRegression backend

[MilesCranmer]

This is on Enzyme 0.13.27, Julia 1.11.2.

To reproduce, run the following code. This is the same code in #2081, but now it hits a segfault whereas before it hit the task switch error.

using SymbolicRegression
using Random: MersenneTwister
using Enzyme
using MLJBase: machine, fit!, predict, report
using Test

X = let rng = MersenneTwister(0), n = 30
    (; x1=randn(rng, n), x2=randn(rng, n), class=rand(rng, 1:2, n))
end

y = let P1 = [0.1, 1.5], P2 = [3.2, 0.5]
    [2 * cos(x2 + P1[class]) + x1^2 - P2[class] for (x1, x2, class) in zip(X.x1, X.x2, X.class)]
end

model = SRRegressor(;
    niterations=100,
    binary_operators=[+, *, /, -],
    unary_operators=[cos, exp],
    populations=30,
    expression_type=ParametricExpression,
    expression_options=(; max_parameters=2),
    autodiff_backend=:Enzyme,
);

mach = machine(model, X, y)
fit!(mach)

[MilesCranmer]

Also noting here that when I use --gcthreads=1 for Julia, I get the same task switch error in #2081.

[wsmoses]

To confirm, what happens on 1.10?

[MilesCranmer]

On 1.10.7 it actually runs for a bit, but then hits a segfault:

[69110] signal (11.2): Segmentation fault: 11 second: 7.00e+04. Press 'q' and then <enter> to stop execution early.
in expression starting at REPL[12]:1
gc_mark_outrefs at /Users/julia/.julia/scratchspaces/a66863c6-20e8-4ff4-8a62-49f30b1f605e/agent-cache/default-honeycrisp-XC9YQX9HH2.0/build/default-honeycrisp-XC9YQX9HH2-0/julialang/julia-release-1-dot-10/src/gc.c:2510 [inlined]
gc_mark_and_steal at /Users/julia/.julia/scratchspaces/a66863c6-20e8-4ff4-8a62-49f30b1f605e/agent-cache/default-honeycrisp-XC9YQX9HH2.0/build/default-honeycrisp-XC9YQX9HH2-0/julialang/julia-release-1-dot-10/src/gc.c:2739
gc_mark_loop_parallel at /Users/julia/.julia/scratchspaces/a66863c6-20e8-4ff4-8a62-49f30b1f605e/agent-cache/default-honeycrisp-XC9YQX9HH2.0/build/default-honeycrisp-XC9YQX9HH2-0/julialang/julia-release-1-dot-10/src/gc.c:2886
jl_gc_mark_threadfun at /Users/julia/.julia/scratchspaces/a66863c6-20e8-4ff4-8a62-49f30b1f605e/agent-cache/default-honeycrisp-XC9YQX9HH2.0/build/default-honeycrisp-XC9YQX9HH2-0/julialang/julia-release-1-dot-10/src/partr.c:142
_pthread_start at /usr/lib/system/libsystem_pthread.dylib (unknown line)

[wsmoses]

Interesting, can you try to reduce the complexity of the MWE for us to debug?

[MilesCranmer]

I've tried before when I first posted #2081 but I couldn't figure out how to break it down further