Daniel James wrote:
On Thu, 22 Feb 2007 18:13:58 -0000, Eric Niebler <eric@boost-consulting.com> wrote:
There are two testers running with intel-linux-9.0. The foreach tests pass on one, and fail on the other. How is that possible? If these are different compilers, their toolsets should have different names. As it is, I don't even know how to mark up this failure.
It's the version of gcc they're using with intel linux. The test passes for gcc-3.3, fails for gcc-3.4, gcc-4.0 and gcc-4.1. So the names could be changed to something like intel-linux-9.0-gcc-3.3 and intel-linux-9.0-gcc-3.4. You can see the version details in http://tinyurl.com/2a45wj and http://tinyurl.com/323fb6
Interesting! I'm not familiar with the intel compiler on linux. In what way does it make use of gcc? No matter, I think I can guess what the problem is from the tests you ran. Can you apply the attached patch to boost/foreach.hpp and re-run your tests? Thanks for the tremendous help! -- Eric Niebler Boost Consulting www.boost-consulting.com Index: foreach.hpp =================================================================== RCS file: /cvsroot/boost/boost/boost/foreach.hpp,v retrieving revision 1.35 diff -b -d -u -r1.35 foreach.hpp --- foreach.hpp 10 Jan 2007 23:30:15 -0000 1.35 +++ foreach.hpp 26 Feb 2007 02:16:12 -0000 @@ -29,8 +29,8 @@ // Some compilers let us detect even const-qualified rvalues at compile-time #if BOOST_WORKAROUND(BOOST_MSVC, >= 1310) \ - || BOOST_WORKAROUND(__GNUC__, >= 4) \ - || (BOOST_WORKAROUND(__GNUC__, == 3) && (__GNUC_MINOR__ >= 4)) + || (BOOST_WORKAROUND(__GNUC__, >= 4) && !defined(BOOST_INTEL)) \ + || (BOOST_WORKAROUND(__GNUC__, == 3) && (__GNUC_MINOR__ >= 4) && !defined(BOOST_INTEL)) # define BOOST_FOREACH_COMPILE_TIME_CONST_RVALUE_DETECTION #else // Some compilers allow temporaries to be bound to non-const references.