Hi guys, I tried a couple of compilations on the CVS trunk and it appears that gcc's -Wundef option is basically unusable, due to most usages of BOOST_WORKAROUND not being prefixed by a corresponding "defined expression". Now, I'm thinking: if a suitable pragma exists (I'll look it up) could we just disable that warning at the config system level? I know this impacts non-boost user code as well but, whatever that implies, we have a precedent for Visual C++: // turn off the warnings before we #include anything #pragma warning( disable : 4503 ) // warning: decorated name length exceeded Opinions? --Gennaro.
On Wed, 14 Jun 2006 15:45:24 +0200, Gennaro Prota <gennaro_prota@yahoo.com> wrote:
Hi guys,
I tried a couple of compilations on the CVS trunk and it appears that gcc's -Wundef option is basically unusable, due to most usages of BOOST_WORKAROUND not being prefixed by a corresponding "defined expression". Now, I'm thinking: if a suitable pragma exists (I'll look it up) could we just disable that warning at the config system level? I know this impacts non-boost user code as well but, whatever that implies, we have a precedent for Visual C++:
// turn off the warnings before we #include anything #pragma warning( disable : 4503 ) // warning: decorated name length exceeded
Opinions?
Actually there was a similar discussion a couple of years ago. My preferred solution is still the one I proposed at that time: defining *all* compiler identification macros, always: all of them would be zero except the one for the compiler being used. This requires n #defines to zero in one place (where n is the number of compilers we support), plus in each compiler-config file an undef/redef such as: #undef BOOST_THIS_COMPILER #define BOOST_THIS_COMPILER version This would completely solve the problem without any user impact (unless some company decides to ship a compiler having version "zero"! :)) --Gennaro.
participants (1)
-
Gennaro Prota