[Debian-med-packaging] ITKv4 build failing on i386

[Gert Wollny]

On Thu, 2014-04-24 at 19:50 -0500, Steve M. Robbins wrote:
> On April 24, 2014 03:34:11 PM Gert Wollny wrote:
> Absolutely: see 
> https://buildd.debian.org/status/fetch.php?pkg=insighttoolkit4&arch=i386&ver=4.5.2-1&stamp=1397463745
> -Steve
Hmm, 


itkSimpleImageRegistrationTestFloat & Double fail with some very strange
error values (same for both tests): 

<DartMeasurement name="ImageError Minimum" type="numeric/double">
  3.0043
</DartMeasurement>
<DartMeasurement name="ImageError Maximum" type="numeric/double">
  2.6022e+292
</DartMeasurement>
<DartMeasurement name="ImageError Mean" type="numeric/double">
  1.8888e+288
</DartMeasurement>

I haven't checked it, but I'm quite certain that the images used in the
tests would not use intensity values that can yield this maximum error
value. 

The compiler is the same version I used so there's not much room for a
systematic error source. The only thing I can try is to run a test build
on another machine, but somehow I doubt that this will give additional
insight, after all, Kitware is running all these tests on a daily basis
and Debian doesn't use some esoteric compiler flags that may introduce
problems like the -march=native did for me.

My only explanation would be that the reference image data used for the
test got corrupted when it was unpacked on the build machine. 

Cheers, 
Gert