Added: NonLinMPC and MovingHorizonEstimator now use value_and_gradient!/jacobian! of DI.jl

[franckgaga]

This theoretically more efficient than calling the objective/constraint functions and DI.jacobian!/DI.gradient! separately. For AutoForwardDiff backend, this feature is similar to DiffResult.jl.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.81%. Comparing base (b6341bc) to head (282010a).
Report is 8 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #197      +/-   ##
==========================================
- Coverage   98.90%   98.81%   -0.10%     
==========================================
  Files          25       25              
  Lines        4220     4223       +3     
==========================================
- Hits         4174     4173       -1     
- Misses         46       50       +4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[franckgaga]

This is also more efficient in practice. With the benchmark of the inverted pendulum with SingleShooting and Ipopt, I get on Julia v1.11.5 before this improvement:

btime_NMPC_track_solver_IP = median(bm) = TrialEstimate(432.509 ms)

and after this improvement:

btime_NMPC_track_solver_IP = median(bm) = TrialEstimate(395.618 ms)

[franckgaga]

BTW thanks once more @gdalle for DI.jl !

I wanted to use DiffResults.jl in the past to improve the efficiency of ForwardDiff calls, but I completely forgot about this. Reading through your doc and noticing value_and_jacobian! reminded me to implement it, and it was very easy with DI.jl API.