SciMLLogging Integration

[jClugstor]

Checklist

• Appropriate tests were added
• Any code changes were done in a way that does not break public API
• All documentation related to code changes were updated
• The new code follows the
  contributor guidelines, in particular the SciML Style Guide and
  COLPRAC.
• Any new documentation only uses public API

Additional context

Add any other context about the problem here.

[jClugstor]

will need SciML/SciMLBase.jl#1159 to fully work

[termi-official]

Since you are already working on the logging system, is there any chance that we can have additional logging messages for the time integration loop at trace level to debug issues in the timestepping itself? Or is that too costly?

[ChrisRackauckas]

We are splitting that to a different step. First let's get the infrastructure together, then add new messages.

Or is that too costly?

Never! If it's super costly, just put it in the All preset only. Semi? Detailed. Since there's individual switching and tons of levels, the goal is to just add tons and tons of default prints and if they are costly just don't have them on by default, but if you switch to All then you just get a lot of information.

[jClugstor]

At this point anything that was behind a verbose is now under a @SciMLMessage and is controlled by an ODEVerbosity.

Currently, the linear_verbosity and nonlinear_verbosity fields are not used because those get set in alg_cache or build_nlsolver which do not have access to the integrator or the DEOptions where the verbosity object is. So setting those in ODEVerbosity will not do anything.

For the Linear solves and Nonlinear solves I set verbose to Minimal() to prevent unnecessary printing from those during an ODE solve.

[ChrisRackauckas]

is this ready to review?

[jClugstor]

Yep, ready to be looked at.

[ChrisRackauckas]

Lots of test failures

[jClugstor]

Yes, because I'm using Minimal() for verbose for the Nonlinear and Linear solves, and they can't handle that until
SciML/NonlinearSolve.jl#647 and
SciML/LinearSolve.jl#756

[jClugstor]

There are a lot of

ERROR: LoadError: Unsatisfiable requirements detected for package OrdinaryDiffEqCore [bbf590c4]:
 OrdinaryDiffEqCore [bbf590c4] log:
 ├─possible versions are: 1.0.0-1.36.0 or uninstalled
 └─restricted to versions 1.37.0-1 by OrdinaryDiffEq [1dea7af3] — no versions left
   └─OrdinaryDiffEq [1dea7af3] log:
     ├─possible versions are: 6.103.0 or uninstalled
     └─OrdinaryDiffEq [1dea7af3] is fixed to version 6.103.0

which is weird because I bumped OrdinaryDiffEq to 1.37 in this PR, is CI not using the versions in this PR?

[ChrisRackauckas]

LTS ignores sources

[ChrisRackauckas]

DiffEqDocs needs a PR that explains it in more detail how it's a solver-package specific thing.

[ChrisRackauckas]

Actually it makes it a bit rough for it to be solver-specific... since then you cannot do things in benchmarks and optimization more generally.

[ChrisRackauckas]

why is this required?

[jClugstor]

Is there anything I can do about errors like this?

Failed to precompile OrdinaryDiffEqAdamsBashforthMoulton [89bda076-bce5-4f1c-845f-551c83cdda9a] to "/home/chrisrackauckas/.julia/compiled/v1.12/OrdinaryDiffEqAdamsBashforthMoulton/jl_QB6FLI".
ERROR: LoadError: Error opening package file /home/chrisrackauckas/.julia/compiled/v1.12/OrdinaryDiffEqCore/SoYJU_894Ui.so: /home/chrisrackauckas/.julia/compiled/v1.12/OrdinaryDiffEqCore/SoYJU_894Ui.so: cannot open shared object file: No such file or directory

[jClugstor]

@ChrisRackauckas I think this is ready, but it's hard to tell with all of the test failures. Most of them are because LTS doesn't use sources, some are from JET, and some are failures to open .so files. But I went through as many of the legitimate failures that I could find and fixed them. I also bumped all of minor versions of every subpackage to reflect that.

The biggest issue here turned out to be that in order to get the Linear and Nonlinear verbosity objects to the linear and nonlinear caches I had to add an argument to every alg_cache function and call, as well as build_nlsolver. This messes up DelayDiffEq and StochasticDiffEq, since DelayDiffEq uses OrdinaryDiffEq.alg_cache and StochasticDiffEq uses build_nlsolver. The solution for that is probably to just update those to SciMLLogging? But that will take a day or two probably.

But yeah not sure if it's mergable in this state.

[ChrisRackauckas]

The biggest issue here turned out to be that in order to get the Linear and Nonlinear verbosity objects to the linear and nonlinear caches I had to add an argument to every alg_cache function and call, as well as build_nlsolver. This messes up DelayDiffEq and StochasticDiffEq, since DelayDiffEq uses OrdinaryDiffEq.alg_cache and StochasticDiffEq uses build_nlsolver. The solution for that is probably to just update those to SciMLLogging? But that will take a day or two probably.

Yes, make the downstream PRs. But this should be easy to make backwards compatible with just the standard tricks right? Just make verbose be the last argument and make a dispatch that has a default verbose=ODEVerbosity(), so that it just adds the arg and passes down. Label it in a deprecated.jl file for future removal. Do this on alg_cache and build_nlsolver. Then downstream will work, just ignoring the verbosity argument until updated.

[ChrisRackauckas]

this branch is already very far behind, continuing in the other PR