Gitlab CI does not include unit tests for OrdinaryDiffEqNonlinearSolve

[lte678]

It seems to me that the tests for the subpackage OrdinaryDiffEqNonlinearSolve were never added to the gitlab CI after splitting it off into its own package (commit c399c19 perhaps).

I fixed it up, but the currently existing newton_tests.jl unfortunately don't pass with the following error message:

Newton Tests: Test Failed at /home/lteichro/Projects/scratchpad/ordinarydiffeq_bug1214/dev/OrdinaryDiffEq/lib/OrdinaryDiffEqNonlinearSolve/test/newton_tests.jl:17
  Expression: sol2.stats.nf <= sol1.stats.nf
   Evaluated: 102023 <= 102008

For the following assertion:

for prob in (prob_ode_lorenz, prob_ode_orego)
    sol1 = solve(prob, Trapezoid(), reltol = 1e-12, abstol = 1e-12)
    @test sol1.retcode == DiffEqBase.ReturnCode.Success
    sol2 = solve(prob, Trapezoid(nlsolve = NLNewton(relax = BackTracking())),
        reltol = 1e-12, abstol = 1e-12)
    @test sol2.retcode == DiffEqBase.ReturnCode.Success
    @test sol2.stats.nf <= sol1.stats.nf
end

Is it ok to toss the test or the test case? It seems like a quite mundane failure to me, unless there a reason to expect a strictly lesser amount of function evaluations.

I would like to contribute an unrelated fix to that OrdinaryDiffEqNonlinearSolve , for which I would like to get unit tests running again.

[ChrisRackauckas]

Oh no, that's a bad omission. Thanks for catching this.

I don't think it's an issue for backtracking to take slightly more steps than the other method. There's nothing theoretical backing that assumption, seems to just be a regression test but with the norm change that could definitely just be a random thing that it occurred in the other direction before. I would just add @test sol2.stats.nf+20 <= sol1.stats.nf and that's a sufficient update to the test.

[lte678]

Fixed in #2573
Tests are passing.