[ChrisRackauckas-Claude]

Note: this PR should be ignored until reviewed by @ChrisRackauckas.

What changed and why

lib/NonlinearSolveBase/Project.toml: TimerOutputs = "1" → TimerOutputs = "0.5.23, 1" (restoring the range that #1198 dropped on 2026-08-26 as "stale"), plus a patch bump of NonlinearSolveBase to 2.49.2 (master released 2.49.1 via #1221 while this was being prepared; the resolution logs below were taken with the branch at 2.49.1, before rebasing) so the widened compat can be registered.

The Downstream.yml lanes OrdinaryDiffEq.jl/InterfaceI, InterfaceII and Regression_I fail on master since https://github.com/SciML/NonlinearSolve.jl/actions/runs/33303882132 with Unsatisfiable requirements detected for package AlgebraicMultigrid. The constraint chain:

1. Restore weighted preconditioning and tolerances for default Krylov solves OrdinaryDiffEq.jl#4391 (merged 2026-08-30 09:00 UTC, between the last green Downstream run https://github.com/SciML/NonlinearSolve.jl/actions/runs/33290997505 and the first red one) bumped lib/OrdinaryDiffEqDifferentiation to LinearSolve = "5.15". Downstream develops OrdinaryDiffEq master, so the test env can no longer downgrade LinearSolve.
2. LinearSolve ≥ 5.14 has AlgebraicMultigrid = "2" in its weakdeps compat (4 - 5.13.1 still allowed 0.5, 1 - 2). AlgebraicMultigrid is in OrdinaryDiffEq's test target, so it must resolve to 2.0.1.
3. AlgebraicMultigrid 1.2 – 2.0.1 pins TimerOutputs = "0.5.29 - 0.5".
4. NonlinearSolveBase ≥ 2.48.1 (after Drop stale [compat] majors older than one year #1198) requires TimerOutputs = "1" → no version left.

The green run escaped this by resolving LinearSolve v5.13.1 + AlgebraicMultigrid v0.5.1; that escape hatch is gone now. Nothing in the General registry changed for these packages — the trigger was the OrdinaryDiffEq compat bump exposing the compat drop from #1198.

The long-term fix belongs in AlgebraicMultigrid (widen its compat to TimerOutputs 1; it only uses @timeit_debug). Until that is released, NonlinearSolveBase has to keep accepting TimerOutputs 0.5.

Verification

Minimal reproducer (Julia 1.12.7): temp project, Pkg.develop of ., lib/NonlinearSolveBase, lib/SciMLJacobianOperators, lib/NonlinearSolveFirstOrder, lib/SimpleNonlinearSolve, then Pkg.add([AlgebraicMultigrid, LinearSolve@5.15]).

Before (master f561fb5cb):

ERROR: LoadError: Unsatisfiable requirements detected for package TimerOutputs [a759f4b9]:
 TimerOutputs [a759f4b9] log:
 ├─possible versions are: 0.4.0 - 1.2.0 or uninstalled
 ├─restricted to versions 1 by NonlinearSolveBase [be0214bd], leaving only versions: 1.0.0 - 1.2.0
 │ └─NonlinearSolveBase [be0214bd] log:
 │   ├─possible versions are: 2.49.0 or uninstalled
 │   ├─restricted to versions 2.45.0 - 2 by NonlinearSolveFirstOrder [5959db7a], leaving only versions: 2.49.0
 │   └─NonlinearSolveBase [be0214bd] is fixed to version 2.49.0
 └─restricted by compatibility requirements with AlgebraicMultigrid [2169fc97] to versions: 0.5.29 — no versions left
   └─AlgebraicMultigrid [2169fc97] log:
     ├─possible versions are: 0.2.0 - 2.0.1 or uninstalled
     ├─restricted to versions * by an explicit requirement, leaving only versions: 0.2.0 - 2.0.1
     ├─restricted by compatibility requirements with Compat [34da2185] to versions: 0.4.0 - 2.0.1 or uninstalled, leaving only versions: 0.4.0 - 2.0.1
     └─restricted by compatibility requirements with LinearSolve [7ed4a6bd] to versions: 2.0.1 or uninstalled, leaving only versions: 2.0.1
       └─LinearSolve [7ed4a6bd] log:
         ├─possible versions are: 0.1.0 - 5.15.0 or uninstalled
         ├─restricted to versions 5.4.0 - 5 by NonlinearSolveFirstOrder [5959db7a], leaving only versions: 5.4.0 - 5.15.0
         └─restricted to versions 5.15 by an explicit requirement, leaving only versions: 5.15.0

After (this branch), same script:

⌅ [a759f4b9] ↓ TimerOutputs v1.2.0 ⇒ v0.5.29
  [2169fc97] AlgebraicMultigrid v2.0.1
  [7ed4a6bd] LinearSolve v5.15.0
  [be0214bd] NonlinearSolveBase v2.49.1 `~/sandbox/NonlinearSolve-downstream/lib/NonlinearSolveBase`

CI-shaped check: temp project developing the five upstream projects above plus OrdinaryDiffEq master (ff71f2f3f) and its lib/OrdinaryDiffEqDifferentiation, Pkg.add of every non-developed name in OrdinaryDiffEq's test target — resolves with the fix:

  [2169fc97] AlgebraicMultigrid v2.0.1
  [7ed4a6bd] LinearSolve v5.15.0
  [be0214bd] NonlinearSolveBase v2.49.1 `.../lib/NonlinearSolveBase`
  [4302a76b] OrdinaryDiffEqDifferentiation v3.11.1 `.../OrdinaryDiffEq-downstream/lib/OrdinaryDiffEqDifferentiation`

Tests of lib/NonlinearSolveBase with the newly re-allowed version pinned (Pkg.add(PackageSpec(name="TimerOutputs", version="0.5.29")), Julia 1.12.7):

⌃ [a759f4b9] TimerOutputs v0.5.29
GROUP=Core  ->  362 passed of 362, "Testing NonlinearSolveBase tests passed", exit=0
GROUP=QA    ->  QA | 20 20 51.8s, "Testing NonlinearSolveBase tests passed", exit=0

typos on the diff: clean. Runic not applicable (only Project.toml changed).

Not verified

• The Downstream OrdinaryDiffEq lanes themselves (a full OrdinaryDiffEq InterfaceI run); only the environment resolution the lanes die in was reproduced and shown to pass.
• Root/other sublibrary test groups; only lib/NonlinearSolveBase Core and QA were run, since NonlinearSolveBase is the only project in the repo that depends on TimerOutputs.

Things to push back on

• This intentionally reverses part of Drop stale [compat] majors older than one year #1198. The "older than one year" rule was applied to a compat that a live downstream dependency (AlgebraicMultigrid 2.0.1, the only version LinearSolve ≥ 5.14 accepts) still pins to. Once AlgebraicMultigrid releases with TimerOutputs 1 compat, the 0.5.23 part can be dropped again.
• Pre-existing and unrelated to this PR: NonlinearSolveBase.@static_timeit calls TimerOutputs.timer_expr(__module__, false, to, name, expr), but timer_expr has taken a leading source::LineNumberNode since TimerOutputs 0.5.27, so enable_timer_outputs() makes NonlinearSolveBase fail to precompile with any TimerOutputs ≥ 0.5.27 (ArgumentError on 0.5.29, MethodError on 1.2.0). The default (timers disabled) is unaffected. Tracked separately in an issue.

🤖 Generated with Claude Code (Claude Code 2.1.251, model: claude-fable-5)

https://claude.ai/code/session_016LsC6pp9z6s5EABX9DnVjE

[ChrisRackauckas-Claude]

Closing in favor of bumping TimerOutputs in the downstream that still pins 0.5:

• Allow TimerOutputs 1.x JuliaLinearAlgebra/AlgebraicMultigrid.jl#137 (widen to 0.5.29, 1)

Restoring 0.5 compat in NonlinearSolveBase would reverse #1198; AlgebraicMultigrid only uses documented @timeit_debug, so it can take TimerOutputs 1 without code changes.

Separately, the timer_expr/enable_timer_outputs() breakage noted in this PR's "Things to push back on" is fixed in #1225 / #1224.