std::cout and long long problem

[Benjamin Kosnik]

> > WRT OS bits, wouldn't it be reasonable to arrange for any C++
> > translation unit to implicitly #include the libstdc++ config bits
> > upfront?  Adding `-include c++config.h' to the C++ specs doesn't seem
> > too much trouble to me.
> 
> Interesting idea.  In any case, I think it makes sense to revisit this
> discussion now that we have CPLUSPLUS_CPP_SPEC.

I'm pretty much for it too.

I think we'll have to do something like this anyway to get proper 
pre-compiled-header support. So doing it now might make sense, and then 
c++config.pch can just get added instead or something...

This might also offer us a way out of the whole deprecated/backwards 
headers dilema, where we are supposed to warning about inclusion, but only when a g++ 
flag isn't used...

-benjamin