#pragma once

[NN]

The most of modern compilers do support #pragma once feature. This can reduce significantly the compilation time. Boost should use this too.

Proposal: In boost/config.hpp define the BOOST_ONCE macro: (Of course this should be defined in the right compiler configuration header file)

!#cpp
// VC >4.0
// GCC >=4.30
// ICC >10
#if (defined(_MSC_VER) && _MSC_VER > 1000) || \
    (defined(__GNUC__) && ((__GNUC__ == 4 && __GNUC_MINOR__ >= 30) || (__GNUC__ > 4))) || \
    (defined(__ICC) && __ICC > 1000)
#   define BOOST_ONCE once    
#else
#   define BOOST_ONCE message("")
#endif

Using in other files: boost/xxx/yyy.hpp

!#cpp
#ifndef BOOST_XXX_YYY_INCLUDED
#define BOOST_XXX_YYY_INCLUDED

#include <boost/config.hpp>

#pragma BOOST_ONCE

#endif // BOOST_XXX_YYY_INCLUDED

[niels_dekker]

Thanks for your ticket. But the use of #pragma once appears rather controversial. You may want to check out the discussion on comp.std.c++, in December 2007: ​#pragma once in ISO standard yet? (73 postings!)

GCC 4 has already declared #pragma once as being obsolete: ​Obsolete once-only headers

[NN]

Defining the macro is a good solution for this program, because if #pragma once will be declared as unsupported in GCC 5.x it may do nothing.

Anyway #pragma once does decrease the compilation time, so there is no reason not to define the macro and use it. Because it can be simply disabled in a simple and a standard way.

Thanx.

[Shunsuke Sogame <pstade.mb@…>]

Replying to NN:

The most of modern compilers do support #pragma once feature. This can reduce significantly the compilation time.

Seems cool macro. Which compiler would compile faster?

Regards,

[John Maddock]

I thought we did some tests and decided it wasn't worth the effort? Compilers are mostly smart enough now not to need this hint.

In any case the proposed solution won't work: #pragma message isn't supported by anything except MSVC I believe, so every other compiler would emit a warning about an unknown pragma :-(

We could define a BOOST_HAS_PRAGMA_ONCE, but this would basically be synonymous with _MSC_VER I believe and therefore superfluous, especially now the only other case - GCC - has deprecated this.

Closing down for now, please reopen if a better solution presents itself and/or if you have convincing use cases (with data!) :-)

John.