Define the independent variable in the motor test

[ChrisRackauckas-Claude]

Please ignore this PR until it has been reviewed by @ChrisRackauckas.

Summary

• define the thermal motor test independent variable with the documented public @independent_variables macro
• import that macro directly from its declaring owner, ModelingToolkitBase, instead of relying on a ModelingToolkit re-export
• avoid the implicit SciCompDSL fallback for t, which is a PleaseImportDynamicQuantities sentinel when DynamicQuantities is absent from the minimum dependency environment
• keep the change limited to the standalone test; runtime code and dependencies are unchanged

Final head: 015b399e, based on upstream d0243bd3.

Root cause

The motor test was the only top-level test using @mtkmodel without defining or importing t. SciCompDSL therefore fell back to ModelingToolkitBase.t. With ModelingToolkit 11.2.0, ModelingToolkitBase 1.0.0, and SciCompDSL 1.0.1, that fallback is ModelingToolkitBase.PleaseImportDynamicQuantities() unless the DynamicQuantities extension is loaded, and @mtkcompile motor = ThermalMotor() fails in _check_if_dde.

A source-history bisect isolated the DynamicQuantities extension boundary:

• first sentinel commit: 2785a86207f9d96fbe4f9395a4346228a43e8dd3 (refactor: move DynamicQuantities to extension)
• parent: ce0436f48c25553bcaa3b6bcd6421d1095e98bd1 (refactor: move DiffEqNoiseProcess to extension)

The historical parent cannot currently be instantiated from the registry because its intermediate SymbolicUtils 4 / Symbolics 6 constraints are unsatisfiable. The released failing dependency stack above was reproduced directly.

@independent_variables is exported by its owner ModelingToolkitBase, has an owner docstring in lib/ModelingToolkitBase/src/independent_variables.jl, and is included in the rendered API reference. ModelingToolkitBase is already a direct dependency of this package.

Local verification

• exact minimum stack before this change: reproduced _check_if_dde(::Vector{Equation}, ::ModelingToolkitBase.PleaseImportDynamicQuantities, ::Vector{System})
• final exact minimum stack after the explicit owner import, with OrdinaryDiffEq preloaded as in native tests: Thermal Motor Demo | Pass 6 / Total 6 (143.3 s, including first-use work)
• final current stack after the explicit owner import: Thermal Motor Demo | Pass 6 / Total 6 (137.0 s, including first-use work)
• GROUP=Core on current upstream passed through Core/isothermal_compressible.jl (27/27), then stopped at the separately audited exact-equality assertion in test/magnetic.jl:66 (2 passed, 1 failed), before reaching motor; the focused runs above execute motor end-to-end
• repository-wide Julia Runic check: exit 0 across all 81 tracked Julia files
• git diff --check: exit 0

No test is skipped, disabled, loosened, or made warn-only by this PR.

Independent minimum-compatibility blocker

On current upstream, the downgrade workflow reaches an earlier ModelingToolkitBase 1.0 array-codegen failure in test/continuous.jl before motor. Focused draft #482 raises the ModelingToolkitBase floor to 1.1 and validates continuous 76/76. That independent fix is deliberately not duplicated here; the exact minimum-stack focused motor run above verifies this PR's test after preloading the native solver stack.

[ChrisRackauckas-Claude]

Investigation scratchpad

1. Synced the fork and branched from clean upstream d0243bd36ca2751c983bb49eddb837aeb7583dbe.
2. Ran the downgrade action logic locally. The current action selected ModelingToolkit 11.0.0 / ModelingToolkitBase 1.0.0 / SciCompDSL 1.0.0, so I also constructed the specifically reported ModelingToolkit 11.2.0 / ModelingToolkitBase 1.0.0 / SciCompDSL 1.0.1 environment.
3. Reproduced the failure at @mtkcompile motor = ThermalMotor() with _check_if_dde(..., ModelingToolkitBase.PleaseImportDynamicQuantities, ...).
4. Traced the fallback through SciCompDSL: because the test had no caller binding for t, model parsing selected ModelingToolkitBase.t, which is the extension sentinel in this dependency generation.
5. Source-history bisection isolated 2785a86207f9d96fbe4f9395a4346228a43e8dd3 as the first commit introducing this sentinel behavior, with parent ce0436f48c25553bcaa3b6bcd6421d1095e98bd1. The parent environment cannot currently resolve because the historical SymbolicUtils 4 / Symbolics 6 combination is unsatisfiable against current registry metadata.
6. Audited two candidate fixes. ModelingToolkitBase.t_nounits is marked public but lacks an owner binding docstring at version 1.0.0, so I did not add a downstream dependency on it. The selected @independent_variables t macro is exported, owner-documented, rendered in the API docs, and passed the exact stack end to end.
7. Verified exact stack 6/6, current stack 6/6, and Runic 81/81. Native Core stopped before motor at the existing magnetic exact-equality failure at test/magnetic.jl:66; no test was disabled, silenced, or loosened.
8. Committed only test/motor.jl as 8421d9e6948ed2215a851425ab6ea73464d820dd with the required co-author trailer.

[ChrisRackauckas-Claude]

API-source follow-up at 015b399e:

• explicitly imported @independent_variables from its declaring/public/documented owner, ModelingToolkitBase, instead of relying on ModelingToolkit's re-export
• reran the exact MTK 11.2.0 / MTKBase 1.0.0 / SciCompDSL 1.0.1 motor stack: 6/6 passed
• reran the current dependency stack: 6/6 passed
• reran repository-wide Julia Runic: exit 0 across all 81 tracked Julia files
• git diff --check: exit 0
• refetched upstream and origin before the normal push; upstream remained d0243bd3, the remote feature branch remained exactly 8421d9e6, and no contributor commits were present

The PR body and CI now point at the final head.

[ChrisRackauckas-Claude]

CI mapping: the downgrade failure is the already-audited ModelingToolkitBase 1.0 symbolic-array indexing defect, not the motor test change. It passes Core/analog.jl (100/100) and Core/chua_circuit.jl (1/1), then errors in StateSpace and TransferFunction with symbolic indices under Core/continuous.jl (54 pass, 2 errors). #482 raises the ModelingToolkitBase floor to 1.1 and adds the focused StateSpace regression; its exact downgraded continuous run passes 76/76. Exact #483 job

[ChrisRackauckas-Claude]

CI mapping for the newly completed red jobs:

• All completed Core variants stop at the exact clean-main magnetic equality in test/magnetic.jl:66 (2 pass, 1 fail), before reaching the motor test changed here. That one-ULP reconstructed-observable regression is independently audited and fixed in focused draft test: bound reconstructed magnetic flux balance by ULPs #484.
• The documentation job reports the same clean-main strict-example failures isolated in stacked drafts Fix callable interpolation builder symtype #480/Repair documentation example setup #481: ambiguous bare @extend OnePort() and the resulting parametrized-interpolation cascade. Repair documentation example setup #481’s hosted documentation job passes after those focused repairs.

Neither failure implicates this PR’s direct owner import; the exact downgraded and current motor tests both pass 6/6 locally.