Re: [Boost-bugs] [Boost C++ Libraries] #7372: test_complex.cpp fails with intel compiler v12 and -ansi-alias

Subject: Re: [Boost-bugs] [Boost C++ Libraries] #7372: test_complex.cpp fails with intel compiler v12 and -ansi-alias
From: Boost C++ Libraries (noreply_at_[hidden])
Date: 2012-09-14 18:58:22

#7372: test_complex.cpp fails with intel compiler v12 and -ansi-alias
----------------------------------------+-----------------------------------
  Reporter:  russelljbrennan@…          |       Owner:  ramey                         
      Type:  Bugs                       |      Status:  new                           
 Milestone:  To Be Determined           |   Component:  serialization                 
   Version:  Boost 1.51.0               |    Severity:  Problem                       
Resolution:                             |    Keywords:  serialization, intel, aliasing
----------------------------------------+-----------------------------------

Comment (by Russell Brennan <russelljbrennan@…>):

 round_error() returns 0.5 on my machine for both double and float, so the
 test for delta between expected and actual values being greater than 1
 seems fine, if not excessively lax.  One thing I just noticed is that
 printing the values out via std::cout causes the error to go away
 (Heisenberg uncertainty?).  Could this indicate that the compiler is
 optimizing away the subtraction?  I just added -O0 and -O1 with successful
 results even with -ansi-alias, but I get failures at -O2 and above, but
 only in conjunction with -ansi-alias.  I am beginning to wonder, is this
 an issue with the boost implementation or my compiler?  How can I make a
 determination either way?

-- 
Ticket URL: <https://svn.boost.org/trac/boost/ticket/7372#comment:2>
Boost C++ Libraries <http://www.boost.org/>
Boost provides free peer-reviewed portable C++ source libraries.