Re: [Boost-testing] [XL C/C++][IBM_Canada_Ltd][serialization][multi_index]strange error with boost/serialization/force_include.hpp

Subject: Re: [Boost-testing] [XL C/C++][IBM_Canada_Ltd][serialization][multi_index]strange error with boost/serialization/force_include.hpp
From: Christopher Cambly (ccambly_at_[hidden])
Date: 2009-03-30 12:40:14

>
> Thanks, I'll do that if nobody sheds more light on this, but my
> puzzlement stems from the fact that BOOST_USED *shouldn't* be resolving
> to __attribute__ ((used)) as far as I can see: XL C/C++ does *not*
> define __GNUC__ or __INTEL_COMPILER (see http://tinyurl.com/66el78 ),
> and I cannot see how else can BOOST_USED be defined as
> __attribute__ ((used)).
>
> Joaquín M López Muñoz
> Telefónica, Investigación y Desarrollo
>

Perhaps I can clear this up.  It is possible that xlC may support such an
attribute in a future version of the compiler.  However, we are not allowed
to
pre-announce new features or report results to Boost for unreleased
versions of the compiler.

We run our nightly testing on several released versions of the compiler and
some unreleased
versions of the compiler but we only report the results for AIX V 10.1 back
to Boost.  All
of our testing is run off the same Boost code base.

If a future version of the compiler were to support some specific attribute
and we would like
to test the new functionality, we would use the Boost.Test patch mechanism
to patch
force_include.hpp to enable the attribute for our compiler.  However, there
is no way to indicate
which patch file applies to which test run.

This patch mechanism is likely causing the confusion, and the XL compiler
does not define either of
those macros.

Our source is not 100% in sync with Boost since we have not pushed our
local patches upstream into
the main branch.  I can open a ticket and submit the patches for review but
I believe we need to fix
up the force_include.hpp patch to restrict the compiler versions to which
it is applied.

Chris Cambly
XL C++ Compiler Developmen