Support SciMLBase v3 / OrdinaryDiffEq v7 ecosystem

[ChrisRackauckas-Claude]

Summary

Allows BifurcationKit to be used alongside the SciMLBase v3 / OrdinaryDiffEq v7 / RecursiveArrayTools v4 ecosystem release that's currently rolling out across SciML.

Compat change

Package	Before	After
SciMLBase	"^1, ^2"	"^1, ^2, 3"

Source change

The new ensemble API changed the prob_func signature from (prob, i, repeat) to (prob, ctx::EnsembleContext) with no deprecation shim. Updated the three _prob_func definitions in src/periodicorbit/FlowDE.jl to accept either signature — detect whether the second arg is an Integer (v1/v2) or an EnsembleContext (v3) and extract the trajectory index accordingly:

_prob_func = (prob, ctx_or_i, _rest...) -> begin
    ii = ctx_or_i isa Integer ? ctx_or_i : ctx_or_i.sim_id
    remake(prob, ...)
end

output_func already had a 2-arg signature (sol, i) whose unused i is benignly replaced by the context object on v3, so no functional change there beyond renaming the throwaway parameter to make the intent obvious.

Other v3 breaking changes checked and not used by BifurcationKit

I went through OrdinaryDiffEq.jl's NEWS.md and grepped the codebase for each migration. None of these patterns are used in BifurcationKit's source:

• sol[i] linear indexing on ODESolution (RAT v4 makes it scalar) — uses sol.u[i] consistently
• length(sol) / eachindex(sol) / for u in sol on a SciML solution — not used
• EnsembleSolution iteration via for sol in es — uses sol.u consistently
• u_modified! → derivative_discontinuity! — not used
• sol.destats / has_destats — not used
• DEAlgorithm / DEProblem / DESolution abstract type renames — not used (uses concrete ODEProblem, EnsembleProblem, SciMLBase.AbstractODEIntegrator)
• sol.retcode == :Symbol — no equality comparisons against Symbol retcodes
• verbose = false on solve — BifurcationKit's own verbose kwargs aren't passed to underlying solve
• concrete_solve — only mentioned in a comment
• alias_u0 / alias_du0 — not used
• gamma/beta1/beta2/qmin/qmax/etc. controller kwargs — not used
• chunk_size / standardtag / diff_type — not used

Verified

julia> using Pkg; Pkg.activate("."); Pkg.add(name="SciMLBase", version="3"); using BifurcationKit
julia> pkgversion(BifurcationKit)
v"0.5.8"

Resolves with: SciMLBase 3.9.1 / RecursiveArrayTools 4.3.0 / DiffEqBase 7.3.0.

Context

Unblocks Catalyst.jl v17 release (SciML/Catalyst.jl#1466) and any other downstream user that wants BifurcationKit alongside the v7 ecosystem.

Please ignore until reviewed by @ChrisRackauckas / the BifurcationKit maintainers.

🤖 Generated with Claude Code

[rveltz]

Thank you Chris for your work (and your co-worker ;) )

I will try to review this PR as fast as I can

[rveltz]

Hi,

I am done with a major overhaul of BK.
I want to merge your changes but the above CI failed for installing OdinaryDiffEq.
Am I missing something?

[ChrisRackauckas-Claude]

@rveltz Sorry, the diff was stale — your 0.6.0/0.7.0 refactor introduced several more v3/v7 break points that hadn't surfaced when I opened this. Rebased on current master and pushed a follow-up commit (047be020). Three things broke beyond the original prob_func signature change:

1. VectorContinuousCallback lost affect_neg! in SciMLBase v3. The signature for affect! changed from (integrator, idx::Int) to (integrator, simultaneous_events::Vector{Int8}) where +1 = upcrossing, -1 = downcrossing, 0 = none. PoincareShooting wants upcrossings only, so its callback now does any(==(Int8(1)), events) && terminate!(integrator). Both call sites in src/periodicorbit/ShootingDE.jl are updated.

2. OrdinaryDiffEq v7 stopped reexporting most solvers. Only Tsit5/Vern*/Rosenbrock23/Rodas5P/FBDF are kept in the umbrella package now. Tests that reach for ODE.Rodas5, ODE.Rodas4, ODE.Rodas4P, ODE.KenCarp4 were hitting UndefVarError. Test files now pull them directly from OrdinaryDiffEqRosenbrock / OrdinaryDiffEqSDIRK, which are added to [extras] and [targets] in Project.toml. Test compat for OrdinaryDiffEq widened to "^6.33, 7".

3. terminate! and EnsembleProblem aren't reexported from OrdinaryDiffEq v7 either. The two tests that referenced ODE.terminate! / ODE.EnsembleProblem now using SciMLBase: terminate!, EnsembleProblem.

The CI failure you saw was the resolver giving up: with RecursiveArrayTools = "4" pinned but the test-env OrdinaryDiffEq capped at ^6, the SciMLBase v3 ↔ OrdinaryDiffEq v6 incompatibility had no satisfying assignment. Widening OrdinaryDiffEq test compat to allow v7 lets the resolver pick the v3-compatible stack.

Verified locally on Julia 1.10 with SciMLBase 3.13.0 / OrdinaryDiffEq 7.0.0 / DiffEqBase 7.5.0 / RecursiveArrayTools 4.3.0:

Test Summary:  |  Pass  Total      Time
BifurcationKit | 15247  15247  99m56.7s
     Testing BifurcationKit tests passed

CI should be green this time — please re-run when you have a moment. Ignore until reviewed by @ChrisRackauckas as usual.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.59%. Comparing base (e8e27df) to head (047be02).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #307   +/-   ##
=======================================
  Coverage   91.59%   91.59%           
=======================================
  Files          69       69           
  Lines       10315    10321    +6     
=======================================
+ Hits         9448     9454    +6     
  Misses        867      867           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.