Re: [boost] Re: List bandwidth and compiler problems [was Solaris Forte 6u.1 ...]

From: Beman Dawes (bdawes_at_[hidden])
Date: 2001-09-07 14:41:20

At 01:40 PM 9/7/2001, Gennadiy E. Rozental wrote:

 >--- In boost_at_y..., Beman Dawes <bdawes_at_a...> wrote:
 >> Beyond that, several of the proposed fixes seem to spread compiler
 >> version checking out into individual library files.  The whole point
 >> of config.hpp (and the replacement config system just reviewed) is
 >> to centralize configuration information in one easy to maintain 
location.
 >
 >I agree with you. But Can you propose how to do this?

I'm not sure at what level you're asking that question.

A simplistic answer is just to look at the config docs and header.  See 
http://www.boost.org/libs/config/index.htm or config.htm in John Maddock's 
newly accepted config system.

If you are asking at a somewhat deeper level how to deal with a compiler 
which "sort of" supports a feature, then look at how config.hpp handles 
VC++ member templates, for example:

#   if _MSC_VER <= 1200  // 1200 == VC++ 6.0
...
//    VC++ 6.0 has member templates but they have numerous problems 
including
//    cases of silent failure, so for safety we define:
#     define BOOST_NO_MEMBER_TEMPLATES
//    For VC++ experts wishing to attempt workarounds, we define:
#     define BOOST_MSVC6_MEMBER_TEMPLATES

Then boost code which doesn't want to try to use the feature looks for 
BOOST_NO_MEMBER_TEMPLATES, while code which wants to apply a compiler 
specific workaround looks for BOOST_MSVC6_MEMBER_TEMPLATES.

Does that answer your question?

--Beman