Update to MathOptInterface v1.0

[odow]

How does requires work if the requires package introduces breaking changes? I found: JuliaPackaging/Requires.jl#82

It seems like GalacticOptim has some strong compatibility issue with MathOptInterface, and yet there's nothing in the Project.toml to prevent someone installing incompatible versions.

Closes #214
Closes #167

[ChrisRackauckas]

How does requires work if the requires package introduces breaking changes? I found: JuliaPackaging/Requires.jl#82

We need to fix this by moving to subpackages. Requires.jl indeed does not play nicely with compatibilities.

[odow]

Ah. This is going to need the Nonconvex packages updated. cc @mohamed82008

[odow]

• Bump to Ipopt 1.0 JuliaNonconvex/NonconvexIpopt.jl#7
• Bump compat for JuMP packages JuliaNonconvex/NonconvexJuniper.jl#4
• Bump compat bounds JuliaNonconvex/NonconvexPavito.jl#6

[odow]

Let's hold off merging this for a bit. I'm looking into a bug that @ccoffrin found.

[odow]

Now blocked by:

• Prep for v0.9.3 ds4dm/Tulip.jl#116

[ChrisRackauckas]

@blegat and @jonasmac16 if you want to take a look.

[ccoffrin]

@odow, looks like Tulip issue was resolved. Is this branch passing tests now?

[odow]

I give up. There are still a bunch of compatibility issues with the Nonconvex stuff. If we want to test a bunch of different packages like this, it should be done in separate projects. There's no need to attempt to install everything in a single environment.

And the tests don't even pass on master anyway:
https://github.com/SciML/GalacticOptim.jl/runs/5632625030?check_suite_focus=true

[ccoffrin]

FWIW, odow:od/moi1 is working for me and hence is an improvement over master. I support merging even if all tests do not pass.

[Vaibhavdixit02]

@mohamed82008 would you be able to help out with the version issues?

[Vaibhavdixit02]

I support merging even if all tests do not pass.

Let's wait a couple of days and see if we can get some help to get the version conflicts resolved, otherwise we'll go ahead and merge this as is, hope that's ok?

[ChrisRackauckas]

We plan to split the wrappers out to separate subpackages, like how KernelAbstractions.jl is done with the lib being completely different repos. See https://github.com/JuliaGPU/KernelAbstractions.jl/tree/master/lib. This would then get rid of Requires.jl usage entirely (so good for startup times), and would make it easier to version each subpackage separately. The individual subpackages can then be ran as part of the GalacticOptim.jl tests, but be able to participate in version control (since Requires.jl has a pretty bad issue with that 😅).

Our first experiment with doing this is NeuralPDELogging.jl JuliaRegistries/General#57196. It seems like that's working now, so we should do this here.

[ChrisRackauckas]

Let's wait a couple of days and see if we can get some help to get the version conflicts resolved, otherwise we'll go ahead and merge this as is, hope that's ok?

I'm just going to merge so I can do a follow-up PR that splits to version controlled subpackages and fix this madness.

[mohamed82008]

I'm just going to merge so I can do a follow-up PR that splits to version controlled subpackages and fix this madness.

This madness is one of the motivations behind the Nonconvex.jl design of a shell package that loads algorithms from their respective packages.

[jonasmac16]

Let's wait a couple of days and see if we can get some help to get the version conflicts resolved, otherwise we'll go ahead and merge this as is, hope that's ok?

I'm just going to merge so I can do a follow-up PR that splits to version controlled subpackages and fix this madness.

Sorry, I didn't have time to look at the PR (in the last 2 weeks of my PhD atm :( ). Happy to help with the subpackages once that is done as this has been on my todo list for some time.

[ChrisRackauckas]

Note that we moved everything to a subpackage form, so now all of the solvers track their own dependencies in their Project.toml. We need to figure out what's going on with Nonconvex.jl's bindings, but all of the other ones should be good now.