为什么boost log break会提升GMP multiprecision？

Question

我刚刚在Mac OS Mojave 10.14.6上使用Homebrew将我的boost升级到了1.71.0。执行此操作后，以下程序无法编译：

#include <boost/log/sinks/text_ostream_backend.hpp>
#include <boost/multiprecision/gmp.hpp>

int main() {}

$ clang++ -v
Apple LLVM version 10.0.1 (clang-1001.0.46.4)
Target: x86_64-apple-darwin18.7.0
Thread model: posix
InstalledDir: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin

$ clang++ -std=c++14 boost_gmp.cpp -o boost_gmp
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:284:10: error: no template named '_mp_size'; did you mean 'mp11::mp_size'?
      if(mpf_sgn(m_data) == 0)
         ^
/usr/local/include/gmp.h:2239:26: note: expanded from macro 'mpf_sgn'
#define mpf_sgn(F) ((F)->_mp_size < 0 ? -1 : (F)->_mp_size > 0)
                         ^
/usr/local/include/boost/mp11/list.hpp:45:19: note: 'mp11::mp_size' declared here
template<class L> using mp_size = typename detail::mp_size_impl<L>::type;
                  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:284:10: error: no member named '_mp_size' in '__mpf_struct'
      if(mpf_sgn(m_data) == 0)
         ^~~~~~~~~~~~~~~
/usr/local/include/gmp.h:2239:26: note: expanded from macro 'mpf_sgn'
#define mpf_sgn(F) ((F)->_mp_size < 0 ? -1 : (F)->_mp_size > 0)
                    ~~~  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:372:84: error: no template named '_mp_size'; did you mean 'mp11::mp_size'?
      boost::multiprecision::detail::format_float_string(result, e, org_digits, f, mpf_sgn(m_data) == 0);
                                                                                   ^
/usr/local/include/gmp.h:2239:26: note: expanded from macro 'mpf_sgn'
#define mpf_sgn(F) ((F)->_mp_size < 0 ? -1 : (F)->_mp_size > 0)
                         ^
/usr/local/include/boost/mp11/list.hpp:45:19: note: 'mp11::mp_size' declared here
template<class L> using mp_size = typename detail::mp_size_impl<L>::type;
                  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:372:84: error: no member named '_mp_size' in '__mpf_struct'
      boost::multiprecision::detail::format_float_string(result, e, org_digits, f, mpf_sgn(m_data) == 0);
                                                                                   ^~~~~~~~~~~~~~~
/usr/local/include/gmp.h:2239:26: note: expanded from macro 'mpf_sgn'
#define mpf_sgn(F) ((F)->_mp_size < 0 ? -1 : (F)->_mp_size > 0)
                    ~~~  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1323:27: error: no template named '_mp_size'; did you mean 'mp11::mp_size'?
      if((base != 10) && (mpz_sgn(m_data) < 0))
                          ^
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                         ^
/usr/local/include/boost/mp11/list.hpp:45:19: note: 'mp11::mp_size' declared here
template<class L> using mp_size = typename detail::mp_size_impl<L>::type;
                  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1323:27: error: no member named '_mp_size' in '__mpz_struct'
      if((base != 10) && (mpz_sgn(m_data) < 0))
                          ^~~~~~~~~~~~~~~
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                    ~~~  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1364:14: error: no template named '_mp_size'; did you mean 'mp11::mp_size'?
      return mpz_cmp_si(m_data, i);
             ^
/usr/local/include/gmp.h:2249:6: note: expanded from macro 'mpz_cmp_si'
   ? mpz_cmp_ui (Z, __GMP_CAST (unsigned long, SI))                     \
     ^
/usr/local/include/gmp.h:2246:6: note: expanded from macro 'mpz_cmp_ui'
   ? mpz_sgn (Z) : _mpz_cmp_ui (Z,UI))
     ^
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                         ^
/usr/local/include/boost/mp11/list.hpp:45:19: note: 'mp11::mp_size' declared here
template<class L> using mp_size = typename detail::mp_size_impl<L>::type;
                  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1364:14: error: no member named '_mp_size' in '__mpz_struct'
      return mpz_cmp_si(m_data, i);
             ^~~~~~~~~~~~~~~~~~~~~
/usr/local/include/gmp.h:2249:6: note: expanded from macro 'mpz_cmp_si'
   ? mpz_cmp_ui (Z, __GMP_CAST (unsigned long, SI))                     \
     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/usr/local/include/gmp.h:2246:6: note: expanded from macro 'mpz_cmp_ui'
   ? mpz_sgn (Z) : _mpz_cmp_ui (Z,UI))
     ^~~~~~~~~~~
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                    ~~~  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1369:14: error: no template named '_mp_size'; did you mean 'mp11::mp_size'?
      return mpz_cmp_ui(m_data, i);
             ^
/usr/local/include/gmp.h:2246:6: note: expanded from macro 'mpz_cmp_ui'
   ? mpz_sgn (Z) : _mpz_cmp_ui (Z,UI))
     ^
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                         ^
/usr/local/include/boost/mp11/list.hpp:45:19: note: 'mp11::mp_size' declared here
template<class L> using mp_size = typename detail::mp_size_impl<L>::type;
                  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1369:14: error: no member named '_mp_size' in '__mpz_struct'
      return mpz_cmp_ui(m_data, i);
             ^~~~~~~~~~~~~~~~~~~~~
/usr/local/include/gmp.h:2246:6: note: expanded from macro 'mpz_cmp_ui'
   ? mpz_sgn (Z) : _mpz_cmp_ui (Z,UI))
     ^~~~~~~~~~~
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                    ~~~  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1410:11: error: no template named '_mp_size'; did you mean 'mp11::mp_size'?
   return mpz_sgn(val.data()) == 0;
          ^
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                         ^
/usr/local/include/boost/mp11/list.hpp:45:19: note: 'mp11::mp_size' declared here
template<class L> using mp_size = typename detail::mp_size_impl<L>::type;
                  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1410:11: error: no member named '_mp_size' in '__mpz_struct'
   return mpz_sgn(val.data()) == 0;
          ^~~~~~~~~~~~~~~~~~~
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                    ~~~  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1653:11: error: no template named '_mp_size'; did you mean 'mp11::mp_size'?
   return mpz_sgn(val.data());
          ^
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                         ^
/usr/local/include/boost/mp11/list.hpp:45:19: note: 'mp11::mp_size' declared here
template<class L> using mp_size = typename detail::mp_size_impl<L>::type;
                  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1653:11: error: no member named '_mp_size' in '__mpz_struct'
   return mpz_sgn(val.data());
          ^~~~~~~~~~~~~~~~~~~
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                    ~~~  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1657:8: error: no template named '_mp_size'; did you mean 'mp11::mp_size'?
   if (mpz_sgn(val.data()) < 0)
       ^
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                         ^
/usr/local/include/boost/mp11/list.hpp:45:19: note: 'mp11::mp_size' declared here
template<class L> using mp_size = typename detail::mp_size_impl<L>::type;
                  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1657:8: error: no member named '_mp_size' in '__mpz_struct'
   if (mpz_sgn(val.data()) < 0)
       ^~~~~~~~~~~~~~~~~~~
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                    ~~~  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1668:17: error: no template named '_mp_size'; did you mean 'mp11::mp_size'?
      *result = mpz_sgn(val.data()) < 0 ? (std::numeric_limits<long>::min)()  : (std::numeric_limits<long>::max)();
                ^
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                         ^
/usr/local/include/boost/mp11/list.hpp:45:19: note: 'mp11::mp_size' declared here
template<class L> using mp_size = typename detail::mp_size_impl<L>::type;
                  ^
In file included from boost_gmp.cpp:2:
/usr/local/include/boost/multiprecision/gmp.hpp:1668:17: error: no member named '_mp_size' in '__mpz_struct'
      *result = mpz_sgn(val.data()) < 0 ? (std::numeric_limits<long>::min)()  : (std::numeric_limits<long>::max)();
                ^~~~~~~~~~~~~~~~~~~
/usr/local/include/gmp.h:2238:26: note: expanded from macro 'mpz_sgn'
#define mpz_sgn(Z) ((Z)->_mp_size < 0 ? -1 : (Z)->_mp_size > 0)
                    ~~~  ^
18 errors generated.

卸下#include <boost/log/sinks/text_ostream_backend.hpp>即可解决问题。更改includes的顺序也可以解决这个问题。到底怎么回事？

Answer 1

原来是个Clang parser bug。