[ChrisRackauckas-Claude]

Summary

generate_control_function accepted reassemble_alg without error, but the keyword was packaged into Symbolics.CodegenFunctionOptions and never reached the internal mtkcompile call. This change takes reassemble_alg as an explicit keyword (like simplify/split) and forwards it when compiling an unscheduled system, so linear algebraic loops can be eliminated the same way as with a direct mtkcompile(...; reassemble_alg).

Failing before (src stashed)

Test Summary:                                     | Pass  Fail  Error  Total   Time
generate_control_function forwards reassemble_alg |    2     1      1      4  27.8s

Failure: isequal(dvs, [x]) evaluated [x(t), y(t)]; only(f(...)) then errored (multi-element).

Passing after (with fix)

Test Summary:                                     | Pass  Total   Time
generate_control_function forwards reassemble_alg |    4      4  23.5s

Test group / targeted run

Ran the owning test file lib/ModelingToolkitBase/test/input_output_handling.jl via a scratch env that develops this checkout (group InterfaceI owns this file). Exit 0; new testset 4/4. Runic --check and typos on the three touched files: exit 0.

CI triage (PR vs master; not caused by this PR)

Branch is 0 commits behind master (fd0cbadb43). Diff touches only generate_control_function + its regression test. Fetched failing job logs via gh api .../actions/jobs/<id>/logs, stripped ANSI, and compared signatures to master's recent runs.

Failure class	PR jobs	Actual error	Verdict / evidence
InterfaceI (Julia 1 + lts)	sublibrary-ci MTKBase InterfaceI	variable_parsing.jl:66 — ArgumentError: Variable x must have default of matching size. Got [1, 2]	Pre-existing / Symbolics. Identical on master Sublibrary CI run 37486847188 job 112352183435. Local probe on this PR: same ArgumentError. Untouched by this diff.
Extended (Julia 1 + lts)	sublibrary-ci MTKBase Extended	ccompile.jl:11 — CTarget with expression=Val{false} is not supported	Pre-existing / Symbolics. Identical on master job 112352182552. Local probe: same ArgumentError. Untouched by this diff.
RegressionI (Julia 1 + lts)	sublibrary-ci MTKBase RegressionI	latexify.jl:68 — ReferenceTests needs JULIA_REFERENCETESTS_UPDATE	Pre-existing / reference drift. Identical on master job 112352182774. Untouched by this diff.
Optimization (sublibrary + top-level Tests, Julia 1 + lts; downgrade-mtkbase Optimization)	several	optimizationsystem.jl:814 AmplNLWriter BoundsError (1-element Vector{Expr} at [2]); :874 iprob.f.cons_expr.exprs === nothing fails	Pre-existing. Identical on master Tests run 37488901413 job 112358564208 (same lines/errors). Untouched by this diff.
Benchmarks (lts; 1 cancelled)	Benchmarks	failed to clone ... GitError(... early EOF)	Environmental network/git clone on runner.
SciMLBase.jl/SymbolicIndexingInterface/1/	IntegrationTest	SCCNonlinearProblem has no field u0 inside SciMLBase tests	Downstream / unrelated to this PR's MTKBase change.

No PR-caused CI failure found; no additional code change. Our regression reconfirmed locally after triage: 4/4 pass.

Opus review: MERGE, low risk (reviews/opus-1007-5260/REVIEW.md).

What I did NOT verify

• Full GROUP=InterfaceI / Pkg.test() suite for ModelingToolkitBase or the top-level package (CI runs them; failing subsets above are master-shared).
• Downstream ControlSystemsMTK / named_ss integration beyond the issue's algebraic-loop example.
• Forwarding of other mtkcompile keywords beyond reassemble_alg.

Reviewer pushback

• Whether other structural-simplification kwargs should also be peeled out of the codegen kwargs... dump rather than only reassemble_alg.
• Whether nothing as the default sentinel is preferable to always constructing DefaultReassembleAlgorithm().

Fixes #5255

Please ignore this draft until reviewed by @ChrisRackauckas.

Risk assessment

pending independent review

🤖 Generated with Cursor Agent CLI 2026.10.01-14929f9 (model: auto), transcript /log.txt

Risk assessment (independent review, Claude Code 2.1.301 claude-opus-5-5, reviewed head 36b63ae)

• Risk: low
• Blast radius: ModelingToolkitBase.generate_control_function (both the codegen_compat.jl kwargs method and the inputoutput.jl method) gain a reassemble_alg = nothing keyword that is forwarded to mtkcompile only when non-nothing and only when sys is unscheduled. Default path is byte-for-byte the old call. Analysis-point overloads forward kwargs... to the same method, so they pick it up too. Additive public keyword (minor-level API addition, not breaking). No new deps.
• Evidence: Reviewed head 36b63ae2d8f54472c3ac1ec19b61e5e61db7c544. The new testset is gated on @isdefined(ModelingToolkit), so I ran input_output_handling.jl the way MTK's @mtktestset does (module with using ModelingToolkit; import ModelingToolkitBase), Julia 1.13: PR source 125/125 pass; with lib/ModelingToolkitBase/src from master (5abac29) 123 pass, 1 fail, 1 error — isequal(dvs, [x]) evaluated [x(t), y(t)] vs [x(t)] and the f(...) ≈ -0.25 call errors. Fail-before / pass-after confirmed, matching generate_control_function does not pass options of the structural simplification (reassemble_alg) to mtkcompile #5255. CI: root tests / InterfaceI (1, lts, pre, x86) green (https://github.com/SciML/ModelingToolkit.jl/actions/runs/38850879641/job/116819175941), which is where the gated test runs. The 6 reds are pre-existing: sublibrary InterfaceI "Parsing Test" / RegressionI "Latexify recipes Test" under Symbolics 7.45 (https://github.com/SciML/ModelingToolkit.jl/actions/runs/38850879594/job/116818385387), fixed on master by Update MTKBase tests and compat floors for Symbolics 7.43-7.45 #5281; Docs build and SciMLBase.jl/SymbolicIndexingInterface integration also fail on master (https://github.com/SciML/ModelingToolkit.jl/actions/runs/38856355754, https://github.com/SciML/ModelingToolkit.jl/actions/runs/38856355740).
• Independent review: Claude Code 2.1.301 claude-opus-5-5 rated it low, high confidence: one additive keyword forwarded to mtkcompile, default path unchanged, regression verified locally.
• Merge: auto-merge candidate — after merging master so the Symbolics 7.45 test reds clear. It adds a documented public keyword, so the next MTKBase release should be a minor bump rather than a patch.

---

🤖 Posted by an AI agent — harness: Claude Code 2.1.301 · model: claude-opus-5-5 (independent reviewer; appended by Claude Code 2.1.301 · claude-opus-5-5)
Conversation: https://claude.ai/code/session_01VZjwf7Ux9YtTzwHGHMqnsq