[MOAB-dev] GEOM_SENSE_2 tag
Steve Jackson
sjackson at cae.wisc.edu
Mon Jun 13 13:42:51 CDT 2011
(Mostly for Iulian and Jason's attention.)
There's a problem with the GEOM_SENSE_2 tag that is used by GeomTopoTool and DagMC.
The tag is a sparse tag containing a pair of entity handles. DagMC does not specify a default value for the tag. GeomTopoTool sets {0, 0} as the default value (see GeomTopoTool::check_face_sense_tag()). Using the default value this way introduces an ordering dependence. If DagMC gets this tag handle before GeomTopoTool does, then GeomTopoTool's call to tag_get_handle() will fail, because the first call to tag_get_handle() specifies no default value, and the second call (from GeomTopoTool) specifies a default. Since this tag is sparse, there is no implicit zero default, as there would be for a dense tag.
This issue is causing DagMC::load_file() to fail for any .sat file, because DagMC initializes its tags before loading a file, and ReadCGM uses GeomTopoTool. (Unfortunately, there is no MOAB test for this case right now; the current DagMC tests use h5m files.) This issue also prevents MOAB from successfully loading any .sat file after first loading a mesh file that contains the GEOM_SENSE_2 tag without a default value. Such files include any .sat file that was converted to a mesh using mbconvert or dagmc_preproc before the default value was introduced. (It appears to have been introduced in r4843.)
The simplest fix would be to remove the default value specifier from GeomTopoTool, but doing so causes `make check` to fail. Alternatively, we could make all get_tag_handle() calls that involve this tag specify the same default value. That would fix DagMC::load_file(), but it would not fix the issue where loading old mesh files that do not contain a default value for the tag will confuse future uses of the tag. We could also load the tag using the MB_TAG_DFTOK flag, in order to make tag_get_handle() forgo any checking of default tag values, but that seems like a hack.
What is the best way to fix this?
~S
More information about the moab-dev
mailing list