Boost logo

Boost :

Subject: [boost] Concept check and __LINE__
From: Matthew L. Creech (mlcreech_at_[hidden])
Date: 2010-04-26 17:04:36


I made some trivial changes in a header file today, and suddenly it
refused to compile, throwing out boost::concept_check errors. Here's
an example error:

In file included from
/home/mlcreech/work/PCM-tk/etn_export/include/translator++.hpp:193,
                 from ../../common/source/main.cpp:8:
/home/mlcreech/work/PCM-tk/etn_export/include/translator/handle.hpp:119:
error: conflicting declaration ‘typedef struct
boost::concept::detail::instantiate<boost::concept::requirement<Model>::failed
[with Model = toolkit::translator::Callable<toolkit::translator::CHandle<toolkit::translator::CValArray<toolkit::translator::CSimpleVal<double>
> > >]> toolkit::translator::boost_concept_check119’
/home/mlcreech/work/PCM-tk/etn_export/include/translator/concept/translatable_archetype.hpp:119:
error: ‘toolkit::translator::boost_concept_check119’ has a previous
declaration as ‘typedef struct
boost::concept::detail::instantiate<boost::concept::requirement<Model>::failed
[with Model = toolkit::translator::Translatable<toolkit::translator::translatable_archetype<boost::null_archetype<int>
> >]> toolkit::translator::boost_concept_check119’
make[3]: *** [translate++] Error 1

I had no idea what this was about, but then I went to look up where
the supposed conflicts were and noticed that they were both on line
119. As it turns out, I have 2 different header files which have
BOOST_CONCEPT_ASSERT() for different concepts at line 119. It seems
that Boost internally generates some global names based on the line
number, so it creates 2 identially-named enum definitions. From
"boost/concept_check/general.hpp":

# define BOOST_CONCEPT_ASSERT( ModelInParens ) \
  enum { BOOST_PP_CAT(boost_concept_check,__LINE__) = \
         ::boost::concept_check_<void(*) ModelInParens>::instantiate \
  }

Off-hand I'm not sure how to fix this in Boost (maybe some
preprocessor tricks based on __FILE__?), but it seems like it should
at least be documented as a limitation. Or maybe there's some other
way of using BOOST_CONCEPT_ASSERT() that works around this?

Obviously the immediate fix is trivial - add a few blank lines. I
just thought I'd bring it up in case others hit the same problem.
Thanks

-- 
Matthew L. Creech

Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk