Boost C++ Libraries: Ticket #5112: [config][intel] Patch to detect C++0x features under Intel C++ compiler

attachment set

spillner@… — Mon, 24 Jan 2011 00:39:32 GMT

attachment → intel-c++0x-v1.patch

Patches boost/config/compiler/intel.hpp to detect C++0x features when the -std=c++0x switch is passed to icpc

anonymous — Mon, 24 Jan 2011 18:04:33 GMT

This seems to work OK on Linux - I'm a little wary of using STDC_HOSTED for this as conceptually it could/should be set for C99 conforming mode as well (but appears not to be).

On Win32, I'm a little concerned because their are other ways of disabling MSVC extensions besides -std=c++0x. Traditionally the Intel and MSVC compilers haven't really been usable without MSVC extensions anyway... although I don't know if this is still the case.

For now just applying the Linux patches...

John Maddock — Mon, 24 Jan 2011 18:28:55 GMT

(In [68420]) Add preliminary C++0x support. Refs #5112.

anonymous — Tue, 25 Jan 2011 18:06:11 GMT

I wonder did you test any of the other C++0x features? Intel 11.1 appears not to require:

# undef BOOST_NO_LAMBDAS # undef BOOST_NO_DECLTYPE # undef BOOST_NO_AUTO_DECLARATIONS # undef BOOST_NO_AUTO_MULTIDECLARATIONS

and Intel 12.0 appears not to need:

# undef BOOST_NO_SCOPED_ENUMS # undef BOOST_NO_DELETED_FUNCTIONS # undef BOOST_NO_DEFAULTED_FUNCTIONS

All in C++0x mode of course.

Thanks, John.

spillner@… — Wed, 26 Jan 2011 10:16:26 GMT

I know that there's some decltype support in 11.1, but I received number of compilation warnings (specifically on things like

template < class T > struct Z {

T dummy; typedef decltype(dummy.get()) retval;

};

) that went away when shifting to ICC 12 or GCC, so I enabled that only in 12 and up. This seems consistent with http://software.intel.com/en-us/articles/c0x-features-supported-by-intel-c-compiler/ , which is more conservative than e.g. http://wiki.apache.org/stdcxx/C%2B%2B0xCompilerSupport .

I didn't have a mature use case for the other features, so I confined my attention to the three in the patch. They're also all supported by GCC, which gave me greater confidence in testing. Bottom line, I don't have any reason to believe that the features you named don't work well, I just haven't proven that they do.

status changed; resolution set

John Maddock — Thu, 27 Jan 2011 11:47:15 GMT

status new → closed
resolution → fixed

(In [68472]) Fix up Intel C++0x features. Fixes #5112.