Re: [Boost-users] [chrono] Intel Composer XE 2011 unable to compile chrono library

12 Jul 2011


      Hi,


2011/7/7 
...
---------- Forwarded message ----------
From: Vicente Botet 
To: boost-users@lists.boost.org
Date: Wed, 6 Jul 2011 15:46:31 -0700 (PDT)
Subject: Re: [Boost-users] [chrono] Intel Composer XE 2011 unable to
compile chrono library
Edward Rankin wrote:
...
Hi,
The purpose of this email is to get some insight on the problem we faced
when trying to compile chrono from the 1.47 Beta 1 package. I have tried
both the latest released compiler (2011 Update 4) and the current SP1
beta.
There is no issue with VS 2010 SP1's compiler.
It was first surfaced here:
http://comments.gmane.org/gmane.comp.lib.boost.user/68421. The error
received can be seen below:
< snip >
chrono.cpp
.\boost/chrono/duration.hpp(94): error: expected an expression
                  &&  (boost::is_convertible<Rep2,
                  ^
.\boost/chrono/duration.hpp(124): error: expected an expression
                  &&  (boost::is_convertible<Rep,
                  ^
< snip >
After some discussion with Vicente and some testing by me, it turns out
that
Intel was not happy with the first operand of the && operator in the
following snippet of code:
template <class Duration, class Rep2,
        bool = (
                    (boost::is_convertible<typename Duration::rep,
                        typename common_type<typename Duration::rep,
Rep2>::type>::value)
                &&  (boost::is_convertible<Rep2,
                        typename common_type<typename Duration::rep,
Rep2>::type>::value)
                )
        >
    struct duration_divide_imp
    {
    };
If we place an additional parenthesis around that first operand like
this:
...
template <class Duration, class Rep2,
        bool = (
                    ( (boost::is_convertible<typename Duration::rep,
                        typename common_type<typename Duration::rep,
Rep2>::type>::value) )
                &&  (boost::is_convertible<Rep2,
                        typename common_type<typename Duration::rep,
Rep2>::type>::value)
                )
        >
    struct duration_divide_imp
    {
    };
It will happily compile. It is rather bizarre that this is necessary,
which
leads Vicente and I to believe that there is a bug in the Intel Compiler.
What do you guys think? I have also posted to the Intel forums, the
thread
...
can be seen here:
http://software.intel.com/en-us/forums/showthread.php?t=84061
Regards,
Edward
Hi Edward,
I committed the change on the trunk. Please, let me know if there is some
issue as the I'm not able to check with Intel compilers, and the regression
test don't have a win Intel compiler.
Thanks,
Vicente
I have checked the release candidate and it compiles okay. Thanks.

Edward