On 02/13/2011 03:43 AM, Bruno Haible wrote: > Yes, I agree gnulib should not interfere with the definition of > __attribute__ that the application has provided in its .h files.
Or, more generally, gnulib should not suppress the application's use of __attribute__. Emacs did not #define __attribute__; it merely used __attribute__. > Here's a proposed patch. That looks good as far as it goes, but while we're at it, shouldn't we use a consistent naming convention? It's confusing that one .h file uses _GL_ATTRIBUTE_NORETURN and another uses ATTRIBUTE_NORETURN, for the same notion. > I think the definitions of __attributes__ in > .c files have a low risk and can be left alone. If nobody objects, I'd like to apply a similar patch to the C files as well, if only to forestall people cutting-and-pasting wrong code from them in the future.
