[CI Broken] DDE test failure due to type conversion error in DelayDiffEq

[ChrisRackauckas-Claude]

Summary

CI is currently broken on the master branch. The delay differential equation (DDE) test is failing with a MethodError related to type conversion in the ODE solver cache.

Which CI job(s) failed

• ubuntu-latest Python 3.10: Failed after 49m12s
• macos-latest Python 3.13: Failed after 1h5m33s
• macos-latest Python 3.10: Still pending
• ubuntu-latest Python 3.13: Still pending

Error Details

Test File

.tox/py/lib/python3.10/site-packages/diffeqpy/tests/test_dde.py::test

Error Message

juliacall.JuliaError: MethodError: Cannot `convert` an object of type 
  OrdinaryDiffEqRosenbrock.Rosenbrock23Cache{...}

The full error is a very long type conversion error between incompatible cache types in the DelayDiffEq solver when using the default algorithm with automatic algorithm switching.

Test Code That Failed

def test():
    f = de.seval("""
    function f(du, u, h, p, t)
      du[1] = 1.1/(1 + sqrt(10)*(h(p, t-20)[1])^(5/4)) - 10*u[1]/(1 + 40*u[2])
      du[2] = 100*u[1]/(1 + 40*u[2]) - 2.43*u[2]
    end""")
    u0 = [1.05767027/3, 1.030713491/3]
    
    h = de.seval("""
    function h(p,t)
      [1.05767027/3, 1.030713491/3]
    end""")
    
    tspan = (0.0, 100.0)
    constant_lags = [20.0]
    prob = de.DDEProblem(f,u0,h,tspan,constant_lags=constant_lags)
    sol = de.solve(prob,saveat=0.1)  # <-- Fails here

Root Cause Analysis

This appears to be a type system incompatibility in the Julia DelayDiffEq solver, specifically:

1. The default algorithm uses CompositeAlgorithm with automatic switching between multiple solvers (Tsit5, Vern7, Rosenbrock23, Rodas5P, FBDF)
2. There's a type mismatch when trying to convert between different cache types during algorithm switching
3. The error involves ForwardDiff.Tag{DiffEqBase.OrdinaryDiffEqTag, Float64} vs ForwardDiff.Tag{..., Nothing} - suggesting an autodiff configuration mismatch
4. This is likely caused by a recent upstream change in one of the Julia packages (DelayDiffEq, OrdinaryDiffEq, ForwardDiff, or ADTypes)

Suggested Fix Approaches

1. Pin Julia package versions - Identify which recent Julia package update broke compatibility
2. Specify explicit algorithm - Update the test to use an explicit DDE solver instead of relying on automatic algorithm selection
3. Update DelayDiffEq - The issue may be fixed in a newer version (or may have been introduced by a newer version)
4. Disable autodiff - Try passing autodiff=false to the solver

Link to Failed CI Run

• Run #20791352209: https://github.com/SciML/diffeqpy/actions/runs/20791352209
• Job (ubuntu-latest Python 3.10): https://github.com/SciML/diffeqpy/actions/runs/20791352209/job/59714006830
• Test PR: CI Health Check - Automated Test #171

Additional Context

This was discovered during an automated CI health check on the master branch. The test was passing previously, indicating this is a regression likely caused by upstream Julia package updates.

Test Summary from CI:

1 failed, 3 passed, 5 skipped in 1796.15s (0:29:56)

cc @ChrisRackauckas