#1256: scheduling MoL_PostStep in Post_Recover_Variables --------------------+------------------------------------------------------- Reporter: rhaas | Owner: Type: task | Status: new Priority: major | Milestone: Component: Cactus | Version: Keywords: MoL | --------------------+------------------------------------------------------- this is not ideal since some routines in MoL_PostStep will always change results (eg. Con2Prim). We already removed MoL_PostStep from the past timelevels since the prolongation cannot be handled properly.
Originally (MoL rev 137, https://trac.einsteintoolkit.org/changeset/137/CactusNumerical/MoL) was introduced to enforce boundary conditions and recompute the ADM variables. Since then the policy has changed and the ADM variables are supposed to be checkpointed (rule of thumb is that anything with more than one timelevels should be checkpointed).
When recovering from a checkpoint the boundary conditions are currently satisfied since CarpetIOHDF5 reads in everything that it can, incl. data in ghost and boundary zones. This method ensures that bit identical data is read in. Unfortunately running MoL_PostStep then changes values.
It seems as if MoL_PostStep is never required so should not be run.
The attached trivial patch adds a parameter to MoL to disable running in Post_Recover_Variables.
Note that this is not yet are request for review but a work item since more testing is required to decide if anything breaks when MoL_PostStep is not run.
Tests done with GRHydro (and changes so that eg. the atmosphere mask is identical in ghost zones [currently it is not valid in ghost zones but also never checked in ghost zones]) gives bit identical results even when changing the number of processors while recovering. This is a very simple test though.