Bump compat for OrdinaryDiffEq v7 / StochasticDiffEq v7 ecosystem

[ChrisRackauckas-Claude]

Summary

This PR widens compat entries so JumpProcesses works with the upcoming OrdinaryDiffEq v7 / StochasticDiffEq v7 ecosystem, and migrates test code to the SciMLBase v3 / RecursiveArrayTools v4 APIs that shipped in prior PRs.

Compat changes

• OrdinaryDiffEq: "6" → "6, 7" (main, docs, docs/src/assets Project.toml)
• StochasticDiffEq: "6.82" → "6.82, 7" (same files)
• SciMLBase was already "2.115, 3.1" and DiffEqBase was already "6.192, 7" from PRs Prepare source for SciMLBase v3 and widen compat #579 and Bump DiffEqBase / OrdinaryDiffEq / OrdinaryDiffEqCore compat to include v7/v4 #580; no changes needed.

Source / test migrations

• prob_func signature (SciMLBase v3 breaking change): The old 3-arg prob_func(prob, i, repeat) is gone in SciMLBase v3.6.0, which calls prob_func(prob, ctx) where ctx is an EnsembleContext. Updated in test/ensemble_uniqueness.jl and test/thread_safety.jl.
• length(sol) (RecursiveArrayTools v4 breaking change): length(AbstractVectorOfArray) now returns total scalar element count (size product), not the number of trajectories/timesteps. length(sol.u) is unambiguous. Fixed in test/ensemble_problems.jl, test/saveat_regression.jl, test/thread_safety.jl, and test/variable_rate.jl.

No source changes needed

• src/SSA_stepper.jl already handles u_modified! → derivative_discontinuity! rename via an @static if isdefined(SciMLBase, :derivative_discontinuity!) guard (added in PR Prepare source for SciMLBase v3 and widen compat #579).
• u_modified! calls in src/variable_rate.jl and aggregators are still valid: SciMLBase v3 keeps u_modified! as a deprecated re-export through DiffEqBase v7, so these generate deprecation warnings but continue to work.

Related

• OrdinaryDiffEq v7 tracking: Cascade major version bump across all sublibs OrdinaryDiffEq.jl#3562, Revert OrdinaryDiffEqCore and DiffEqDevTools major bumps from #3562 OrdinaryDiffEq.jl#3565

🤖 Generated with Claude Code

[ChrisRackauckas-Claude]

Pushed a follow-up commit (4dbce91) that finishes the SciMLBase v3 rename properly and bumps this to a breaking v10.0.0.

What the new commit does

• Field rename: SSAIntegrator.u_modified::Bool → SSAIntegrator.derivative_discontinuity::Bool. This is the actual fix for the FunctionWrapper test failure — DiffEqBase v7's initialize! (callbacks.jl:25) accesses integrator.derivative_discontinuity as a field, so a method/getproperty shim can't help. The field has to be renamed.
• Drop the pre-v7 ecosystem entirely (the question from the dig was whether to keep both — answer was no, renaming forces this anyway since DiffEqBase v6 callbacks read integrator.u_modified directly and would break the moment we rename). Compats:
  • DiffEqBase: "6.192, 7" → "7"
  • OrdinaryDiffEq: "6, 7" → "7"
  • OrdinaryDiffEqCore: "3, 4" → "4"
  • OrdinaryDiffEqFunctionMap: "1, 2" → "2"
  • RecursiveArrayTools: "3.35, 4" → "4"
  • SciMLBase: "2.115, 3.1" → "3.1"
  • StochasticDiffEq: "6.82, 7" → "7"
• Version bump 9.26.0 → 10.0.0 (breaking ABI on the integrator struct).
• Internal DiffEqBase.u_modified! method removed; SciMLBase.derivative_discontinuity!(::SSAIntegrator, ::Bool) replaces it (no more @static if isdefined guard since SciMLBase v3 is now mandatory).
• Internal u_modified!(integrator, ...) calls in variable_rate.jl, aggregators/coevolve.jl, aggregators/ssajump.jl rewritten to derivative_discontinuity!(integrator, ...). using ... u_modified! → using SciMLBase: derivative_discontinuity!.
• docs/Project.toml and docs/src/assets/Project.toml updated to drop v6 lower bounds.

Companion work

• Patch v9.26.1: tighten OrdinaryDiffEqCore lower bound to 3.17 #583 — patch release v9.26.1 on the v9 line that just tightens OrdinaryDiffEqCore to "3.17, 4" (fixes the precompile bug that blocks SBMLToolkit#199 / SBMLToolkitTestSuite#63 right now without forcing those repos onto v10).
• Tighten JumpProcesses OrdinaryDiffEqCore compat (v9.24–v9.26.0): require ≥ 3.17 JuliaRegistries/General#154093 — retroactive registry compat tightening for already-released v9.24.0 / v9.25.0 / v9.25.1 / v9.26.0, so installs in the wild stop pulling the broken combination.

This PR (now → v10) completes the v7-ecosystem migration; the other two clean up the v9 line.

[isaacsas]

V10 also requires merging #570, so if these changes are merged they should not be released at this time as a new release.

[ChrisRackauckas]

This shouldn't require a v10 of JumpProcesses: nothing about the API here is changed? We can just keep a deprecation path for now.

[ChrisRackauckas-Claude]

Reworked the field-rename approach into a non-breaking alias (force-pushed; previous v10 commit dropped). New plan: this PR stays a minor bump → v9.27.0, no API breakage.

What the new commit does

• Override Base.getproperty / setproperty! / propertynames on SSAIntegrator so :derivative_discontinuity aliases the existing :u_modified field. Both names hit the same underlying storage. v6-era callbacks (which read integrator.u_modified) and v7-era callbacks (which read integrator.derivative_discontinuity) both keep working with no struct rename.
• Field name unchanged — u_modified::Bool stays. No ABI break for any external code that touches the integrator directly.
• Kept both methods: DiffEqBase.u_modified!(::SSAIntegrator, ::Bool) and the @static if isdefined(SciMLBase, :derivative_discontinuity!)-guarded SciMLBase.derivative_discontinuity!(::SSAIntegrator, ::Bool). The SciMLBase v3 @deprecate u_modified! → derivative_discontinuity! alias is bypassed by JumpProcesses' more specific dispatch, so internal u_modified!(integrator, ...) calls don't trigger deprecation warnings.
• Tightened the existing widening from OrdinaryDiffEqCore = "3, 4" to "3.17, 4" so the resolver can't pick the broken v3.0–v3.16 versions (same lower-bound fix as Patch v9.26.1: tighten OrdinaryDiffEqCore lower bound to 3.17 #583).
• Version bump 9.26.0 → 9.27.0 (minor, fully backwards compatible).

Why this works for both ecosystems

DiffEqBase v6 callbacks: integrator.u_modified → hits the real field directly via default getproperty (sym ≠ :derivative_discontinuity).

DiffEqBase v7 callbacks: integrator.derivative_discontinuity → hits the alias branch in our getproperty, returns getfield(integrator, :u_modified).

Same for writes via setproperty!. propertynames reports both for tab-completion / show.

The rest of the PR (compat widening, ensemble prob_func migration, length(sol) fix, OrdinaryDiffEqFunctionMap test dep) is unchanged.

[ChrisRackauckas]

Okay this form is one that I am fine with. There is no breaking update in the source code. But I will note that the tests are using the SciMLBase v3 ensemble code now, so actually on earlier SciMLBase the tests will be broken. I think it will be very good for the ecosystem at least for now to have some backwards compatible versions, so because this is only in the test source code I think it would be good to allow this for now, and then bump the lower bound later. This should unblock a whole lot and the tests confirm the actual source code is good so I think it's best to make that judgement call there.

[ChrisRackauckas]

@isaacsas on the v10 I assume you'll just want to drop the backwards compat path and just use the new derivative_discontinuity naming?

[isaacsas]

Yes, V10 will be SciMLBase 3 only as it makes changes to rng handling that would be unsafe prior to v3.

[isaacsas]

You can see the PR for what is changing here -- I just need to update it to master and check everything is passing. I think it should otherwise be good to go, but I won't have time to play with it and give it a last review for a few weeks.