Re: [Boost-testing] Darwin/Sandia-intel - looks very problematic

Subject: Re: [Boost-testing] Darwin/Sandia-intel - looks very problematic
From: Artyom Beilis (artyomtnk_at_[hidden])
Date: 2011-07-15 09:12:14

> From: "Belcourt, Kenneth" <kbelco_at_[hidden]>

> To: Running Boost regression tests <boost-testing_at_[hidden]>
> Sent: Thu, July 14, 2011 11:50:19 PM
> Subject: Re: [Boost-testing] Darwin/Sandia-intel - looks very problematic
> 
> Hi Artyom,
> 
> On Jul 14, 2011, at 1:43 AM, Artyom Beilis wrote:
> 
> >  It seems that there is a problem with Darwin/Sandia-intel 11 and 12
> >  compilers - there are lots of faults all over boost libraries
> > that most  of them are segmentation faults even for
> > libraries that has clear record  on other platforms compilers.
> > 
> > It looks like there is a problem  with installation/configuration
> > of these Intel compilers, maybe some ABI  clash of some libraries.
> 
> A year ago I reported to the list and 
> Gennadiy  that there was some kind 
> of infinite recursion in the exception
> handler in  Boost.Test that impacted
> the Sandia Darwin Intel tester.  See
> my email of  last May.  I don't think
> we ever found the actual problem and
> I suspect  it's this same issue impacting
> many of the other Boost libraries tested
> with the  Sandia Darwin Intel tester.

I assume you refer to this:

  http://thread.gmane.org/gmane.comp.lib.boost.testing/6609

However from what I can see there mutiple SIGSEGV in tests on Darwin/Intel
that do not happen on other combinations.

   http://beta.boost.org/development/tests/trunk/developer/test.html

In any case new Boost.Locale that has numerous ERROR 139 (Access violation)
does not use Boost.Test at all.

I'm not trying to tell the Boost.Locale is bug free but at least
it does not use Boost.Test at all, it was tested with Intel
compilers and it was tested on Mac OS X platform (not together), and
numerous Access Violation issues had never seems to 
happen.

Same looks to happen to too many libraries. 

Also take a look for example on UUID

  http://beta.boost.org/development/tests/trunk/developer/uuid.html

It has VERY clean record, it does not use Boost.Test as well
for http://svn.boost.org/svn/boost/trunk/libs/uuid/test/test_serialization.cpp
and it has same ERROR 139 on the same Darwin/Intel combination.

That is why I'm suspicions that it not the problem of Boost.Test.
Maybe some change/code in Boost.Test frequently triggers some common
problem with compiler/libraries but it seems that there may be
some problem that is missed.

Artyom Beilis
--------------
CppCMS - C++ Web Framework:   http://cppcms.sf.net/
CppDB - C++ SQL Connectivity: http://cppcms.sf.net/sql/cppdb/