GitLab

#739 Solve the memory issue with gcc: destroy manually the singletons so that there is no problem when it's done automatically. #740 has been assigned to periodically check whether the likely gcc issue has been lifted.

#723 FSI/Newton: introduce a new vector to store a quantity that wasn't kept correctly during the imolicit differential calls. Not enough to solve my bug though.

#723 FSI/Newton: introduce a stop criterion. I didn't use exactly the same as freefem, but it should do (especially considering I'll use SNES later on to handle the whole coupling and let it care about that).

#723 FSI/Newton: notwithstanding precision issues, coupling seems to yield the same output as Freefem.

#723 FSI/Newton: apparently I replaced most of the previous solid_displacement. Next step is to deactivate the duplicated calls and the now obsolete attributes.

#723 [WIP] FSI/Newton: rewrite the code using the new data attributes (and at the moment not using private methods to split the Forward(). Currently Dirichlet boundary condition are computed with new data entirely.

#723 FSI/Newton: introduce a new attribute which should replace all solid_displacement_XXX attributes.

#723 FSI/Newton: introduce crude Newton loop. Quantities are not yet updated as they should, so at the moment the same result is yielded each time.

#723 Petsc/ShellMatrix: remove the template argument (when context is required it is inside the functions to provide to SetOperation, and such functions take as argument original Mat object rather than wrapped objects.

#723 FSI/Newton: tyring to use Petsc MatShell feature. Neither Mumps nor UMFPack are actually supported for it; I'll try to use GMRES.