Re: [Boost-build] Default threading model in Linux

Subject: Re: [Boost-build] Default threading model in Linux
From: Joaquin M Lopez Munoz (joaquin_at_[hidden])
Date: 2008-11-09 12:16:42

Vladimir Prus <ghost <at> cs.msu.su> writes:

> 
> On Sunday 09 November 2008 19:16:39 Joaquin M Lopez Munoz wrote:
> > 
> > No, I'm referring to the threading model assumed when, for instance,
> > building a test with
> > 
> >   test-suite "foo" : [ run foo.cpp ] ;
> > 
> > So, threading=single is assumed here, right? 
> 
> I don't understand what "assumed" means. When every target is built,
> all non-optional feature has a specific value -- there's nothing
> to "assume". Unless you did something to force MT, threading=single
> will be in effect.

Here, by "assumed" I meant "by default". So, threading=single
is the default unless otherwise specified.

> > That poses a 
> > small problem, as on the other hand BOOST_HAS_THREADS and
> > BOOST_HAS_PTHREADS seem to be unconditionally defined, 
> 
> By what code? On gcc, those are set only if -pthread is added to gcc
> options during compilation.

Here's where my observations differ: I'm using gmane to write
this, so I can't attach files, the following are pasted inline:

***Jamfile.v2***
test-suite "foo_test" : [ run foo.cpp ] ;

***foo.cpp***
#include <boost/config.hpp>
#include <iostream>

int main()
{
  std::cout<<"BOOST_HAS_THREADS "
#if defined(BOOST_HAS_THREADS)
    "defined\n";
#else
    "undefined\n";
#endif

  std::cout<<"BOOST_HAS_PTHREADS "
#if defined(BOOST_HAS_PTHREADS)
    "defined\n";
#else
    "undefined\n";
#endif

  return 1;
}

Put these two files under a folder foo in boost/libs and
play with bjam and different settings or threading:

GCC under Cygwin:
threading=single
  BOOST_HAS_THREADS defined
  BOOST_HAS_PTHREADS defined
threading=multi
  BOOST_HAS_THREADS defined
  BOOST_HAS_PTHREADS defined

MSVC 8.0:
threading=single
  BOOST_HAS_THREADS defined
  BOOST_HAS_PTHREADS undefined
threading=multi
  BOOST_HAS_THREADS defined
  BOOST_HAS_PTHREADS undefined

So, seems like these Boost.Config macros are not affected
by the threading model selected. I can't try on Linux (no
Linux box at hand right now) but some reports I received lead
me to believe the same problem will show, maybe some kind
reader can try these and tell us the results.

Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo