Results are very close on Mac and ubuntu. Agard test number of iterations is 9 if linear solver is GMRES and 4 if PARDISO. Lift and drag do not change significantly.

The test values are different from mac to ubuntu  which is probably related to version difference (python, numpy, scipy, swig and probably others are different).

 - Revert default value of span. - Removed api test for 3D because it could lead to issues when ctests run in parallel (tests would open the same files at the same time which would cause high latency. - Fixed values of Rae 3D case wrt to previous fix

Plus fixed a bug that would set span to 0

Small modifs on validation cases and add them to the validation job on the pipeline

There geo file for naca 0012 was updated by adding the growth ratio as input parameter instead of the farfield mesh size.

Refactor api + updated tests with mode updater and added scalers

API only works for 2D cases

docker with updated VTK version & scipy 1.10

deltaStar was not recomputed for the transition point after averaging solutions, leading to oscillations in the transition region. Test cases have been updated accordingly and deltaStar computation is now tested in the 2D subsonic test case

+ Refactor adjoint and api's

Change potential and mesh derivatives to analytical. Also changed the residuals computation of the viscous solver to the analytical expression instead of solving the linear system. This commit reduces the computational cost of the coupled adjoint methodology by a factor 4, at least.

The api features 3 components: The solver, the geometry and the initial mesh. The api is tested in a aerodynamic shape case where the NACA0012 is constrained at 0deg AoA with cl=0.40 and the objective function to minimize is the drag wrt the max camber and the max camber position. The test case ends with a final shape close to the NACA4412 (camber = 35% & max camber pos = 41%)

Friction drag is now part of the implementation. The rhs of the adjoint system is updated and the test uses total drag. The adjoint test now uses the mesh update capability.

New surface mesh is imposed in the viscous solver in 2D and stagnation point is recomputed at the first coupling iteration

Added adjoint for AoA and mesh in 2D. Nodes indices are now passed to the BoundaryLayer object