On Wed, Dec 24, 2014, Carsten Haitzler wrote: > On Tue, 23 Dec 2014 18:35:20 +0100 (CET) Michelle Legrand > <michelle.legr...@openwide.fr> said: > > > So if I understood correctly, I should replace in > > src/lib/eina/eina_thread.h: > > typedef unsigned long int Eina_Thread; > > by : > > typedef uintptr_t Eina_Thread; > > > > like Nash suggested ? > > or just unsigned long long - which in any sane world will always be 64bit.
It's still completely wrong. It might work, including on Windows with winpthreads (but not pthreads-win32) but it's wrong. As I've mentioned, pthread_t can be anything and the whole issue is about assuming a specific ABI for pthread_t. The proper approach would be to remove this typedef and avoid casts between types which aren't guaranteed to be compatible (i.e. Eina_Thread and pthread_t). It's obviously more work however and the platforms with a non-integer pthread_t are quite uncommon. I've given the example of pthreads-win32 on Windows but it's not that much used nowadays and winpthreads which usually comes with GCC on/for Windows has been created specifically for the applications which violate the specification by assuming an ABI for pthread_t. Realistically I don't see this fixed properly anytime soon and would definitely rather go for the half-fix rather than keeping the win32-based implentation since it probably has many more bugs that are yet to be found and solved. I'd like a check to be added however. Not very long ago I had another ABI issue with libarchive and wrote the following to make compilation fail if the ABIs wouldn't match. It uses C11/C++11's static assert (or that sufficient GCC and LLVM handle it no matter the standard). Code can be seen outside of mail at http://cgit.notk.org/adrien/yypkg/slackware64-current.git/tree/l/libarchive/use-static-asserts-to-guarantee-abi-compatibility.patch /* Take advantage of static_assert/_Static_assert to make sure libarchive is * not built and is not used with a "struct stat" that isn't 64bits. * On Windows I had "stat" #define'd to "stat64" but libarchive #undef'ed it in * its private headers; I then built a program that used libarchive and didn't * have that #undef and got ABI issues: libarchive was storing a 32bit offset * in a "struct stat" but the caller was expecting a 64bit one. * This is a best effort check but which should cover most uses. */ #ifdef __cplusplus #if (__cplusplus >= 201103L) static_assert(sizeof(struct stat) == sizeof(struct stat64), "Not using largefile support"); #endif #else #if ((defined(__STDC_VERSION) && __STDC_VERSION >= 201112L) \ || (defined(__GNUC__) && (__GNUC__ >= 4) && (__GNUC_MINOR__ >= 6)) \ || (defined(__clang__) && (__clang_major__ >= 3) && (__clang_minor__ >= 4))) _Static_assert(sizeof(struct stat) == sizeof(struct stat64), "Not using largefile support"); #endif #endif -- Adrien Nader ------------------------------------------------------------------------------ Dive into the World of Parallel Programming! The Go Parallel Website, sponsored by Intel and developed in partnership with Slashdot Media, is your hub for all things parallel software development, from weekly thought leadership blogs to news, videos, case studies, tutorials and more. Take a look and join the conversation now. http://goparallel.sourceforge.net _______________________________________________ enlightenment-devel mailing list enlightenment-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/enlightenment-devel
