https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89864
--- Comment #24 from Erik Schnetter <schnetter at gmail dot com> --- On Thu, Apr 4, 2019 at 5:43 AM iains at gcc dot gnu.org < gcc-bugzi...@gcc.gnu.org> wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89864 > > --- Comment #22 from Iain Sandoe <iains at gcc dot gnu.org> --- > (In reply to Erik Schnetter from comment #21) > > https://gcc.gnu.org/ml/gcc-patches/2019-04/msg00162.html > > Additional to the comments on list. > > Perhaps this is just unfixable :( > > I suspect that Apple will want to re-release the SDK, and the best real > fix is > to use the SDK from the previous Xcode command line tools (you can still > use > the latest tools from XC10.2 - just install the older version somewhere and > then use --with-sysroot= and/or --sysroot=) > > (a) there's no guarantee that _Atomic u_long has the same size or > alignment as > volatile u_long. > > C11: 6.2.5 Types > ... > 27 ... The size, representation, and alignment of an atomic type need not > be > the same as those of the corresponding unqualified type. > ... > > .. although it *probably* is for simple types for which there are direct > atomic > ops. > This is for Apple systems, where they presumably control the ABI, or are at least aware of the ABI when writing header files. (b) If we hack around it with "volatile" (assuming that the type happens to > have the same size and alignment), this will silently fail in any case it's > used. > _Atomic is used only in a single struct, which is marked "this structure should not be used outside the kernel", and protected by a "#ifdef __APPLE_API_UNSTABLE" (which unfortunately defaults to being defined). To my knowledge, no part of GCC will use this structure. (c) the <atomic> header is only available from C++11, AFAIR, and GCC is > supposed to be boot-strappable with C++98. Iff Apple were to elect to > declare > that the OS *requires* C++11 to operate, then we should fix the > configuration > for Darwin to ensure that this is enforced. > The header file works for all versions of C and C++. It uses _Atomic for C11, _Atomic for C++11 (the bug we're seeing), and volatile in all other cases. There is no requirement for C++11. (d) In any case, is there any guarantee that the representation of the > u_long > as a C++ atomic is the same size and align as its C11 counterpart? (I've > not > checked this). > This seems to be an ABI question, and I assume Apple checked this on their ABIs (probably only Intel and ARM). I assume that this is generally the case as C11 and C++11 atomics were designed at the same time, so the ABI designers will want to ensure sure. -erik
