------- Additional Comments From austern at apple dot com 2004-11-29 20:56 ------- Yep, I see what the problem is now. When we create a builtin in C++, we create the declaration in both the global namespace and namespace std. Later, when we see the declaration of snprintf in the global namespace, we correctly set the assembler name of ::snprintf. We don't set the assembler name of :: snprintf, since it's a separate declaration. Then we see the using-declaration, which is handled by do_nonmember_using_decl. It sees that we have an snprintf in namespace std already, verifies that :: snprintf and std::snprintf are compatible so this is not an error, and then just continues to use the original declaration of std::snprintf that was created by builtin_function. Nothing ever sets the assembler name of std::snprintf to match the one that got set for ::snprintf.
So what to do about this? In principle, I think the answer is that builtin_function is doing something wrong by calling builtin_function_1 twice, once for the global namespace and once for namespace std. If we really must define all builtins in namespace std (what's the rationale?), then at least we should do it with a using-declaration instead of creating two entirely separate decls. This would help compile time performance a little bit: builtin_function_1 is just expensive enough to show up on profiles. If that's too radical a change given the current state of the compiler, then we could special-case builtins in do_nonmember_using_decl to make sure that the assembler name gets copied correctly. This wouldn't be hard. There's already some special-casing for builtins there. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18514
