Move solve to OptimalControl.jl

[ocots]

I propose to move the solve function to OptimalControl.jl.

We have to merge

CTDirect.jl/ext/CTSolveExt.jl

Line 69 in 1c0e2c6

function CommonSolve.solve(ocp::OptimalControlModel,

with

https://github.com/control-toolbox/OptimalControl.jl/blob/d139ce5cfc9c59be2e339a6045a5e1654fde2573/src/solve.jl#L67

Note that for the moment, even if in CTDirect.jl the solve function extends the one from CommonSolve.jl we still have the problem of qualifying for OptimalControl .jl. I think it would better to have only one common solve for an OptimalControlModel. In CTDirect.jl, we can keep the solve of a DOCP.

[PierreMartinon]

I agree that we should have the top-level solve in OptimalControl now.
I'll see how I organise the testing at CTDirect level.
The DOCP method can indeed stay in CTDirect, by the way, we could rename it as solve_DOCP ?
That would simplify the extension and would not interfere with the main solve exported in OptimalControl.

I'll start two branches and see how it goes.

[ocots]

If both solve methods extends the one from commonsolve it will not interfere i think. We can keep the name solve for the resolution of a docp. Thanks to multiple dispatxh it will be ok.

[PierreMartinon]

Ok, branches for the first draft are up.

• I added a new test in OptimalControl for the time_grid option, and plan to add more tests from CTDirect later
• Since we currently only use the direct method, I kept the same signature as the former solve in CTDirect. We will probably adjust this when more solving methods are added.
• @ocots help needed for the CommonSolve extension, for whatever reason I don't seem to be able to get it right. Can we schedule a meeting to do it together ? (if you prefer you can do it directly on the branches and I'll catch up later).

[PierreMartinon]

Ok, I think I got the export for solve, now the tests seem to run fine without prefixing, for both the direct and indirect cases.

I probably misunderstood what you said about CommonSolve: in OptimalControl we don't have a package extension (with ext/ and weakdeps etc), we only import CommonSolve.solve, add a new method and export.

[PierreMartinon]

Another small improvement: there is now an error message when trying to call solve without 'using NLPModelsIpopt' beforehand

@ocots @jbcaillau These should fix the current problems with OptimalControl.solve
#176 and control-toolbox/OptimalControl.jl#233

[jbcaillau]

🙏🏽 @PierreMartinon