[clang] [clang][modules] giving the __stddef_ headers their own modules can cause redeclaration errors with -fbuiltin-headers-in-system-modules (PR #84127)

Ian Anderson via cfe-commits cfe-commits at lists.llvm.org
Tue Mar 12 10:38:17 PDT 2024


================
@@ -7,6 +7,11 @@
  *===-----------------------------------------------------------------------===
  */
 
-#ifndef offsetof
+/*
+ * When -fbuiltin-headers-in-system-modules is set this is a non-modular header
+ * and needs to behave as if it was textual.
+ */
+#if !defined(offsetof) ||                                                      \
+    (__has_feature(modules) && !__building_module(_Builtin_stddef))
----------------
ian-twilightcoder wrote:

Oh I see what you mean. When the Darwin module builds, it will build <stddef.h> which will include <__stddef_offsetof.h>. It will bypass the header guard and (re)define `offsetof`. The Darwin module will also build <sys/_types/_offsetof.h>, so which one gets used is undefined, but the Darwin module will be the single definer of `offsetof`. It's a little weird, but this actually matches the LLVM 17 behavior.

https://github.com/llvm/llvm-project/pull/84127


More information about the cfe-commits mailing list