GCC 4.7 apparently complains if you specify the "always_inline" attribute without also specifying "inline", as observed in [1] and [2]. The exact warning is:
# warning: always_inline function might not # be inlinable [-Wattributes] Right now, we define NS_ALWAYS_INLINE as __attribute__((always_inline)) [3], and we never bother to add "inline" alongside it in practice. So we basically trigger an instance of this warning (and fail to inline the function) for each NS_ALWAYS_INLINE usage in GCC 4.7 builds, with a few exceptions[4]. To fix the warning and actually get inlining behavior, I think we need to do either of the following: (a) include "inline" alongside NS_ALWAYS_INLINE in each usage (b) or, change the definition of NS_ALWAYS_INLINE to include "inline" We already effectively do (b) for MOZALLOC_INLINE [5], FWIW. Thoughts? ~Daniel [1] http://comments.gmane.org/gmane.linux.bluez.kernel/22080 [2] https://github.com/zfsonlinux/zfs/commit/10be533e3344f523e1b8d6ab4f0658897a95ac02 [3] https://mxr.mozilla.org/mozilla-central/source/configure.in#3843 [4] One exception: methods that are defined in the body of a class will get automatically marked as inline "as required by ISO C++" per http://gcc.gnu.org/onlinedocs/gcc/Inline.html [5] https://mxr.mozilla.org/mozilla-central/source/memory/mozalloc/mozalloc.h#43 _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
