Changed: RungeKutta no longer uses DiffCache

[franckgaga]

The NonLinModel internal functions are now:

nonlinmodel.solver_f!(xnext, k, x, u, d, p) -> nothing
nonlinmodel.solver_h!(y, x, d, p) -> nothing

in which xnext, k and y are mutated in-place. The k vector is filled in-place with all the intermediate stage values of nonlinmodel.solver differential equation solver. For exemple with RungeKutta(4), the k vector has nx*(4+1) elements. Note that there are 4 intermediate stages for RK4, but the +1 is required because of the supersample keyword argument.

With this modification DiffCaches are no longer needed in RungeKutta. As a side bonus, there is no longer the @warn about the DiffCache chunk size. AFAIK, this number was impossible to set "correctly" (without the warning) at construction since its optimal value is different if the nonlinear model is used to construct MovingHorizonEstimator, NonLinMPC with SingleShooting, NonLinMPC with MultipleShooting, etc.

I can also remove PreallocationTools.jl dependency since the entire package do not use it anymore.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.88%. Comparing base (7d8ccde) to head (3af2012).
Report is 9 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #183   +/-   ##
=======================================
  Coverage   98.87%   98.88%           
=======================================
  Files          25       25           
  Lines        4193     4204   +11     
=======================================
+ Hits         4146     4157   +11     
  Misses         47       47           

☔ 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.