On 02/28/2012 04:55 AM, Eric Blake wrote: > The trick is that on MSVC, we have to define noreturn in terms of > itself, so that pre-processing doesn't recursively expand things, then > _Noreturn is defined in terms of noreturn.
But the problem isn't <stdnoreturn.h>; it's <stdlib.h> when it's included after <stdnoreturn.h>. For example: $ cat foo.c /* contents of stdnoreturn.h */ #define noreturn __declspec(noreturn) #define _Noreturn noreturn /* contents of stdlib.h */ __declspec(noreturn) void abort (void); $ gcc -E foo.c | tail -n1 __declspec(__declspec(noreturn)) void abort (void); I thought about it some more, and came up with one somewhat-extreme way to try to fix things. <stdnoreturn.h> could do this: #if 1200 <= _MSC_VER # define noreturn #else # define noreturn _Noreturn #endif As I understand it, this would cause MSVC's stdlib.h to expand to this: __declspec( ) void abort (void); which (if Microsoft's grammar is right) should compile. This would mean MSVC would be like any random compiler, where 'noreturn' has no effect on the compiler.
