AHFinderDirect fix on-off issue setting mask when not finding AH every single step

Create issue
Issue #1880 closed
Roland Haas created an issue

when not finding the horizon every single step, the mask is wiped every other step due to interactions of the scheduler and AHFinderDirect storing when the horizon was last found.

https://bitbucket.org/einsteintoolkit/einsteinanalysis/pull-requests/1/fix-on-off-mask/diff

Implements a fix for this.

Keyword: AHFinderDirect

Comments (10)

  1. Roland Haas reporter
    • removed comment

    Are there any more objections to pushing this? I took care of the issues mentioned in the comments in two new commits. Unless there are objections I will push these changes before Saturday.

  2. Roland Haas reporter
    • changed status to open
    • removed comment

    This causes failures in test of Dissipation that use extra dissipation inside of the AH: https://build.barrywardell.net/job/EinsteinToolkitFull/217/testReport/junit/(root)/Dissipation/test_ah_1procs/ and https://build.barrywardell.net/job/EinsteinToolkitFull/217/testReport/junit/(root)/Dissipation/test_ob_1procs/ which are due to changing the scheduling from global (=global-late in ANALYSIS) to global-early. This means that with the change, the apparent horizon is actually recorded correctly in epsdisA. I attach diff files of before (right) and after (left) applying the change. Due to the way loop-local works this is not quite what the code would have seen in EVOL since loop-local loops over all components in all refinement levels not just the ones a local routine would execute on during this timestep.

  3. Frank Löffler
    • removed comment

    After talking to Roland: it seems that regenerating the data is the best option, as the old data is clearly wrong, and the new behavior what people intend. It's a case of bad data in test suits.

  4. Log in to comment