[brlcad-commits] SF.net SVN: brlcad:[50972] brlcad/trunk/BUGS

[<br...@us...>]

Revision: 50972
          http://brlcad.svn.sourceforge.net/brlcad/?rev=50972&view=rev
Author:   brlcad
Date:     2012-06-07 03:26:01 +0000 (Thu, 07 Jun 2012)
Log Message:
-----------
several hours debugging rt, but still can't find the cause of the crash.  document what I've been able to figure out so far (i.e., not much).  summary seems to be a crash using the noise shader (turb) due to corruption in libbn noise bookkeeping or parallel memory management while rendering our sample terra.g model.

Modified Paths:
--------------
    brlcad/trunk/BUGS

Modified: brlcad/trunk/BUGS
===================================================================
--- brlcad/trunk/BUGS	2012-06-07 02:52:50 UTC (rev 50971)
+++ brlcad/trunk/BUGS	2012-06-07 03:26:01 UTC (rev 50972)
@@ -15,6 +15,18 @@
 Recent Bugs
 -----------
 
+* corruption/crash issue rendering terra.g without terra.dsp,
+  possibly/probably parallel-related in libbn/noise.c's
+  bn_noise_turb() function.  gdb eventually catches a segfault in
+  find_spec_wgt() loop with indications of corrupted memory.  valgrind
+  confirms corrupted memory (8 bytes) after a build_spec_table()
+  realloc within find_spec_wgt() but it's not clear why.  etbl
+  book-keeping seems okay..  corrupted read consistently detected on
+  spec_wgts = ep->spec_wgts line in bn_noise_turb().  observed on
+  64-bit linux system:
+
+  rm -f terra.pix && valgrind bin/rt -o terra.pix -P1 -H3 -s1024 share/brlcad/7.21.0/db/terra.g all.g
+
 * mged 'joint' hold command isn't working. looks like the fp_names
   book-keeping is fuxored or the db_lookup is returning NULL (or isn't
   yet performed).

This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.