Fix Moshi bare-singleton deprecation warning on v2-backport

[ChrisRackauckas-Claude]

Please ignore until reviewed by @ChrisRackauckas.

Fixes the precompilation warning reported at https://discourse.julialang.org/t/warning-in-scimlbase-about-continousclock/138534:

┌ SciMLBase
│  ┌ Warning: the bare singleton variant syntax `ContinuousClock` is deprecated, write `ContinuousClock()` instead (near .../SciMLBase/hLfdZ/src/clock.jl:4)
│  │   caller = eval at boot.jl:430 [inlined]
│  └ @ Core ./boot.jl:430
└

Root cause

Moshi v0.3.11 (Roger-luo/Moshi.jl#80, 2026-07-12) deprecated the bare singleton variant syntax inside @data and requires Name(). src/clock.jl on v2-backport declares ContinuousClock and SolverStepClock in the bare form, and the v2 compat entry Moshi = "0.3.6" resolves to v0.3.12, so every user on SciMLBase v2 now gets this warning on precompile. Only one warning shows because Base.depwarn dedupes on (caller frame, funcsym) — SolverStepClock has the same problem.

master (v3) is unaffected: Moshi was dropped entirely in 2837095 ("Replace Moshi with plain Julia structs for Clocks", v3.0.0). Verified — a fresh env with SciMLBase v3.40.0 precompiles clean.

Other Moshi users in the ecosystem (ModelingToolkitBase's MissingGuessValue / StructuralHint, SymbolicUtils' BasicSymbolicImpl) already use the explicit form, so this is the only offender.

Fix

ContinuousClock → ContinuousClock(), SolverStepClock → SolverStepClock(). The explicit form throws ArgumentError: missing fields in variant expression on Moshi < 0.3.11, so the compat floor moves to 0.3.11 alongside the syntax. All @match patterns in clock.jl already used Name() and are unchanged.

Test

Added a regression test that force-recompiles SciMLBase in a subprocess and asserts no deprecation output — the warning is emitted during macroexpansion, so it is invisible unless the package is actually recompiled.

Verified locally on Julia 1.11.9 with Moshi v0.3.12:

• Reproduced the warning on released v2.155.1.
• With the fix, Base.compilecache output is empty and clock behavior is identical (iscontinuous, issolverstepclock, isclock, iseventclock, is_discrete_time_domain, first_clock_tick_time, ==, hash, clock() all as before).
• GROUP=Core Pkg.test() passes: Clocks | 38 pass, 1 broken (pre-existing), Remake | 4414 pass, full group green.
• Confirmed the new test fails on the unpatched source: Expression: !(occursin("deprecated", output)) → Clocks | 37 pass, 1 fail, 1 broken.
• Runic v1 --check clean on both changed files.

Note

Version bumped to 2.155.2. This needs a release off v2-backport to actually reach the users hitting it; if v2 is no longer being registered, the alternative answer for them is to upgrade to SciMLBase v3.

🤖 Generated with Claude Code

https://claude.ai/code/session_01SwhhJmoj8GGx2i2vJ6Rnba

[ChrisRackauckas-Claude]

CI triage

Downstream CI finished with 3 failures. All three are pre-existing on v2-backport and unrelated to this change — none of them touch clocks or Moshi.

Job	Verdict	Evidence
StochasticDiffEq.jl/Interface3/1.10	pre-existing	reproduced locally against released, unmodified SciMLBase v2.155.1
Catalyst/All/1	pre-existing	same failure on #1452 (a comment-spelling-only PR)
StochasticDelayDiffEq.jl/All/1.10	pre-existing	same failure on #1452

StochasticDiffEq — the only one not already red on #1452

test/utility_tests.jl:66,68, asserting a lazy W has not been materialized:

calc_W!: Test Failed at .../test/utility_tests.jl:66
  Expression: convert(AbstractMatrix, integrator.cache.nlsolver.cache.W) != concrete_W
   Evaluated: [-401.0 0.0; 0.0 -200.5] != [-401.0 0.0; 0.0 -200.5]

Reproduced locally on Julia 1.10.11 with StochasticDiffEq master + Pkg.add(name="SciMLBase", version="2.155.1") (the released version, no patch applied):

GROUP=Interface3 Pkg.test()
  Utility Tests | 10 pass, 2 fail
    calc_W!     |  4 pass, 2 fail
  calc_W!: Test Failed at .../utility_tests.jl:66
     Evaluated: [-401.0 0.0; 0.0 -200.5] != [-401.0 0.0; 0.0 -200.5]
  calc_W!: Test Failed at .../utility_tests.jl:68
     Evaluated: [-0.0024937655860349127, -0.004987531172069825] != [-0.0024937655860349127, -0.004987531172069825]

Byte-identical to CI. This is an OrdinaryDiffEqDifferentiation behavior change (calc_W! now materializes the concrete form for an array operator) and is independent of SciMLBase; StochasticDiffEq's own master CI has been red since 2026-05-12.

Catalyst / StochasticDelayDiffEq

Latest downstream releases expect v3-era APIs that v2 does not have:

• UndefVarError: derivative_discontinuity! not defined in SciMLBase (ModelingToolkitBase, Catalyst extensions)
• UndefVarError: stripunits not defined in DiffEqBase
• UndefVarError: handle_callback_modifiers! not defined (StochasticDelayDiffEq precompile)

This is the expected consequence of running a legacy branch's downstream matrix against current releases. ModelingToolkitStandardLibrary.jl/Core/1.10 failed on #1452 but passes here, so there is some run-to-run noise in this matrix as well.

The Core group — which contains the clock tests, including the new regression test — passes.