Vincent Lefevre wrote: > > we use the result in .h > > files that should be installable and, after installation, used by a > > potentially different compiler. Simon's and my packages use this feature. > > The best way to get this done correctly are cascades of explicit #ifs. > > But there is a risk of clash if some application uses some library > with this gnulib feature and another library with different > conventions. I think that an installable <stdnoreturn.h> should > be done globally at the system level.
In Simon's and my packages, we rename system headers, i.e. we should not install <stdnoreturn.h> but rather <unistring/stdnoreturn.h> or similar. This avoids clashes at the package system level (we won't install /usr/include/stdnoreturn.h). And clashes regarding the 'noreturn' macro? I don't see any. In a compilation unit, the last definition matters, and since both definitions (from gnulib and from the other package) are supposedly correct, what is the problem? Bruno
