Fixing FlagBuffer & Valgrind warnings

#218 Merged
  1. Sam Skillman

This fixes up the FlagBufferZones which was causing an out-of-bounds error, (and not doing the right thing). Issue is that we start the Index from 1, not the GridStartIndex. This was offsetting the flagging by 2 zones, and running off the edge.

Since I was at it, I also found how to clean up most if not all of the remaining valgrind warnings that occur between when the data is fully loaded and intialized and when it is written out. There are still pieces that have to do with MPI initialization and some read/writing of data that pop up.

Valgrind issues:

  • Mesh Dimension was not ever initialized for < 3D datasets, led to "uninitialized variable warnings"
  • SiblingList[grid1].GridList can either be a single pointer or an array of pointers depending on "NumberOfGrids". This now matches the logic from CreateSiblingList.C and Grid_FastSiblingLocatorFindSiblings.C

Comments (6)

  1. Nathan Goldbaum

    This fixes Issue #23 when I run the tests on my laptop. The valgrind fixes also lgtm.

  2. dcollins4096

    I ran push suite against 7f39d6602fbc. Several tests throw errors. All of the differences relate to differences in output times, which in turn gives differences in field results. There has been a significant amount of work since then, it would be nice to understand the differences that have cropped up in the last 4 months. However none look like they're related to these changes, so I'll go ahead and approve this PR.

    1. Sam Skillman author

      I think that's okay. Specifically the refinement changes with these changes, and therefore could cause a slightly different timestep criteria. Want to go ahead and merge?

  3. dcollins4096

    I agree. The one change that bothers me is MHDCTOrszagTang, which is non AMR. But clearly that can't have been caused by the current PR.