[Insight-developers] ITK test failures in Debug only

[Luis Ibanez]

Hi Mathieu,

    Is the Debug build a "clean" build ?

    I mean: Is it starting from scratch every time ?


The messages I see in the failing tests are mostly of the
two following types:


    a) *** Exception executing: Segmentation fault

    b) *** Exception executing: Illegal instruction



The typical suspects therefore are:

   1) You may have CVS updated some headers but
      not recompiled the corresponding libraries

   2) Uninitialized variables (in particular pointers)

   3) Byte swapping misinterpretation errors

   4) Outdated shared libraries getting in the way.
      (do you have ITK shared libraries in your path ?)


Here are some patterns in the failing tests:


     3 are image registration tests
    20 are IO tests (**)
     5 are image resampling tests
     2 are level sets
     4 are various filtering tests


(**) This lead to suspect of:

  a) a byte swapping misinterpretation of your architecture

  b) some relationship with a non rebuild *ITKIO* library
     (note that the ImageIOBase object has changed recently)
     If for some reason your ITKIO library didn't rebuild
     then there is a mismatch on the ImageIO objects virtual
     tables.


A first sanity check will be to do a clean build.

If the error reappear, then we still have the shared
outdated shared library suspect, uninitialized memory
suspect and the byte swapping suspect (this last one
has the alibi of the Release build working...)


The *good news* is that your build is a *Debug* build.


Could you please launch one of these failing test
inside gdb and let us know in what file and line
is the segmentation fault happening ?



     Please let us know what you find,


        Thanks


          Luis


-------------------------
Mathieu Coursolle wrote:
> Hi ITK-developpers,
> 
> We currently have 2 dashboard submissions for ITK, one is
> release: 
> <http://www.itk.org/Testing/Sites/RogueResearch2/MacOSXTiger-univ-
> release/20070912-0100-Nightly/Notes.html>
> 
> and another is debug:
> <http://www.itk.org/Testing/Sites/RogueResearch2/MacOSXTiger-univ-debug/
> 20070912-0740-Experimental/Notes.html>
> 
> We usually have a few (2 to 8) tests failures in release which seem to
> be thread related.
> However, the debug build has around 40 test failures. The current debug
> submission has MINC2 support,
> but this should not be the cause of all those failures.
> 
> I was able to reproduce on another computer as well (i386, OSX 10.4, 2
> dual core), downloaded the
> source and build a Debug and a Release build using the default settings
> (No minc2, no patented, no preview, etc). 
> I always got around 5 tests failures in release, but always more than 35
> in Debug.
> 
> Any idea on why this is happening?
> 
> Thanks!
> 
>