diff options
Diffstat (limited to 'include/ruby/internal/stdalign.h')
| -rw-r--r-- | include/ruby/internal/stdalign.h | 93 |
1 files changed, 52 insertions, 41 deletions
diff --git a/include/ruby/internal/stdalign.h b/include/ruby/internal/stdalign.h index 8c56fbbd69..ec68f6a882 100644 --- a/include/ruby/internal/stdalign.h +++ b/include/ruby/internal/stdalign.h @@ -17,21 +17,19 @@ * recursively included from extension libraries written in C++. * Do not expect for instance `__VA_ARGS__` is always available. * We assume C99 for ruby itself but we don't assume languages of - * extension libraries. They could be written in C++98. + * extension libraries. They could be written in C++98. * @brief Defines #RBIMPL_ALIGNAS / #RBIMPL_ALIGNOF */ #include "ruby/internal/config.h" -#ifdef HAVE_STDALIGN_H -# include <stdalign.h> +#ifdef STDC_HEADERS +# include <stddef.h> #endif #include "ruby/internal/compiler_is.h" -#include "ruby/internal/compiler_since.h" -#include "ruby/internal/has/feature.h" -#include "ruby/internal/has/extension.h" #include "ruby/internal/has/attribute.h" #include "ruby/internal/has/declspec_attribute.h" +#include "ruby/internal/has/feature.h" /** * Wraps (or simulates) `alignas`. This is C++11's `alignas` and is _different_ @@ -67,7 +65,7 @@ #elif RBIMPL_HAS_DECLSPEC_ATTRIBUTE(align) # define RBIMPL_ALIGNAS(_) __declspec(align(_)) -#elif RBIMPL_HAS_ATTRIBUTE(aliged) +#elif RBIMPL_HAS_ATTRIBUTE(aligned) # define RBIMPL_ALIGNAS(_) __attribute__((__aligned__(_))) #else @@ -75,50 +73,63 @@ #endif /** - * Wraps (or simulates) `alignof`. Unlike #RBIMPL_ALIGNAS, we can safely say - * both C/C++ definitions are effective. + * Wraps (or simulates) `alignof`. + * + * We want C11's `_Alignof`. However in spite of its clear language, compilers + * (including GCC and clang) tend to have buggy implementations. We have to + * avoid such things to resort to our own version. + * + * @see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=52023 + * @see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69560 + * @see https://bugs.llvm.org/show_bug.cgi?id=26547 */ -#if defined(__cplusplus) && RBIMPL_HAS_EXTENSION(cxx_alignof) -# define RBIMPL_ALIGNOF __extension__ alignof - -#elif defined(__cplusplus) && (__cplusplus >= 201103L) +#if defined(__DOXYGEN__) # define RBIMPL_ALIGNOF alignof - -#elif defined(__INTEL_CXX11_MODE__) -# define RBIMPL_ALIGNOF alignof - -#elif defined(__GXX_EXPERIMENTAL_CXX0X__) -# define RBIMPL_ALIGNOF alignof - -#elif defined(__STDC_VERSION__) && RBIMPL_HAS_EXTENSION(c_alignof) -# define RBIMPL_ALIGNOF __extension__ _Alignof - -#elif defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) -# define RBIMPL_ALIGNOF _Alignof +#elif defined(__cplusplus) +# /* C++11 `alignof()` can be buggy. */ +# /* see: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69560 */ +# /* But don't worry, we can use templates. */ +# define RBIMPL_ALIGNOF(T) (static_cast<size_t>(ruby::rbimpl_alignof<T>::value)) + +namespace ruby { +template<typename T> +struct rbimpl_alignof { + typedef struct { + char _; + T t; + } type; + + enum { + value = offsetof(type, t) + }; +}; +} #elif RBIMPL_COMPILER_IS(MSVC) +# /* Windows have no alignment glitch.*/ # define RBIMPL_ALIGNOF __alignof -#elif defined(__GNUC__) -# /* At least GCC 2.95 had this. */ -# define RBIMPL_ALIGNOF __extension__ __alignof__ +#elif defined(HAVE__ALIGNOF) +# /* Autoconf detected availability of a sane `_Alignof()`. */ +# define RBIMPL_ALIGNOF(T) RB_GNUC_EXTENSION(_Alignof(T)) -#elif defined(__alignof_is_defined) || defined(__DOXYGEN__) -# /* OK, we can safely take <stdalign.h> definition. */ -# define RBIMPL_ALIGNOF alignof - -#elif RBIMPL_COMPILER_SINCE(SunPro, 5, 9, 0) -# /* According to their manual, Sun Studio 12 introduced __alignof__ for both -# * C/C++. */ -# define RBIMPL_ALIGNOF __alignof__ - -#elif 0 -# /* THIS IS NG, you cannot define a new type inside of offsetof. */ +#else +# /* :BEWARE: This is the last resort. If your compiler somehow supports +# * querying the alignment of a type, you definitely should use that instead. +# * There are 2 known pitfalls for this fallback implementation: +# * +# * First, it is either an undefined behaviour (C) or an explicit error (C++) +# * to define a struct inside of `offsetof`. C compilers tend to accept such +# * things, but AFAIK C++ has no room to allow. +# * +# * Second, there exist T such that `struct { char _; T t; }` is invalid. A +# * known example is when T is a struct with a flexible array member. Such +# * struct cannot be enclosed into another one. +# */ +# /* see: http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2083.htm */ # /* see: http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2350.htm */ # define RBIMPL_ALIGNOF(T) offsetof(struct { char _; T t; }, t) -#else -# error :FIXME: add your compiler here to obtain an alignment. #endif #endif /* RBIMPL_STDALIGN_H */ |