#1499: Error in synchronisation after restriction ---------------------+------------------------------------------------------ Reporter: hinder | Owner: eschnett Type: defect | Status: reopened Priority: major | Milestone: ET_2014_05 Component: Carpet | Version: development version Resolution: | Keywords: ---------------------+------------------------------------------------------
Comment (by hinder):
I have prepared two patches for the ET_2013_11 release of Carpet. The first adds a test case which fails on 2 processes due to the bug. The second reverts the commit which removed the sync, causing the test to pass. I believe that this fixes the regression, unless there was some other commit which causes some other problem. Notes: * I have not addressed the application of boundary conditions for high order restriction operators since this was broken in the previous release as well. If people think this is very important and safe, please implement and test the patch, but I don't really have more time to spend on this, and I think an announcement that cell-centering with boundaries near refinement boundaries is not yet working is sufficient. * I have not included the change to restrict into ghost zones, or the optimisation which avoids the sync for vertex-centered code. As such, the performance should be the same as the last release. The trunk will be faster, but this is a less trivial change. * The commit which removed the sync was from May 2014, not September as we thought in the telecon. This means that anyone using the trunk since then could potentially have been affected by this. * All tests pass on Datura with the new code.
OK to backport to the release?