“自动”推断 hashtable_policy.h 中的类型不正确

Question

我正在构建一对工具，gridlab-d 和 HELICS，前者使用后者的共享库。在成功构建/安装 HELICS 后编译 gridlab-d 时，出现以下错误：

In file included from /usr/include/c++/7/bits/hashtable.h:35:0,
             from /usr/include/c++/7/unordered_map:47,
             from /home/ericsilk/.local/helics-2.3.0/include/helics/external/cereal/archives/../cereal.hpp:36,
             from /home/ericsilk/.local/helics-2.3.0/include/helics/external/cereal/archives/portable_binary.hpp:32,
             from /home/ericsilk/.local/helics-2.3.0/include/helics/application_api/ValueConverter_impl.hpp:20,
             from /home/ericsilk/.local/helics-2.3.0/include/helics/application_api/ValueConverter.hpp:65,
             from /home/ericsilk/.local/helics-2.3.0/include/helics/application_api/ValueFederate.hpp:11,
             from /home/ericsilk/.local/helics-2.3.0/include/helics/application_api/CombinationFederate.hpp:10,
             from connection/helics_msg.h:21,
             from connection/helics_msg.cpp:16:
/usr/include/c++/7/bits/hashtable_policy.h: In member function ‘std::size_t std::__detail::_Power2_rehash_policy::_M_next_bkt(std::size_t)’:
/usr/include/c++/7/bits/hashtable_policy.h:563:40: error: invalid operands of types ‘std::size_t {aka long unsigned int}’ and ‘double’ to binary ‘operator<<’
   const auto __max_bkt = size_t(1) << (__max_width * __CHAR_BIT__ - 1);
                          ~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from /usr/include/c++/7/bits/hashtable.h:35:0,
             from /usr/include/c++/7/unordered_map:47,
             from /home/ericsilk/.local/helics-2.3.0/include/helics/external/cereal/archives/../cereal.hpp:36,
             from /home/ericsilk/.local/helics-2.3.0/include/helics/external/cereal/archives/portable_binary.hpp:32,
             from /home/ericsilk/.local/helics-2.3.0/include/helics/application_api/ValueConverter_impl.hpp:20,
             from /home/ericsilk/.local/helics-2.3.0/include/helics/application_api/ValueConverter.hpp:65,
             from /home/ericsilk/.local/helics-2.3.0/include/helics/application_api/ValueFederate.hpp:11,
             from /home/ericsilk/.local/helics-2.3.0/include/helics/application_api/CombinationFederate.hpp:10,
             from connection/helics_msg.h:21,
             from connection/init.cpp:16:
/usr/include/c++/7/bits/hashtable_policy.h: In member function ‘std::size_t std::__detail::_Power2_rehash_policy::_M_next_bkt(std::size_t)’:
/usr/include/c++/7/bits/hashtable_policy.h:563:40: error: invalid operands of types ‘std::size_t {aka long unsigned int}’ and ‘double’ to binary ‘operator<<’
   const auto __max_bkt = size_t(1) << (__max_width * __CHAR_BIT__ - 1);
                          ~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

调查有问题的标头，这是它抱怨的代码：

// Return a bucket size no smaller than n (as long as n is not above the
// highest power of 2).
std::size_t
_M_next_bkt(std::size_t __n) noexcept
{
  const auto __max_width = std::min<size_t>(sizeof(size_t), 8);
  const auto __max_bkt = size_t(1) << (__max_width * __CHAR_BIT__ - 1);
  std::size_t __res = __clp2(__n);

  if (__res == __n)
__res <<= 1;

  if (__res == 0)
__res = __max_bkt;

  if (__res == __max_bkt)
// Set next resize to the max value so that we never try to rehash again
// as we already reach the biggest possible bucket number.
// Note that it might result in max_load_factor not being respected.
_M_next_resize = std::size_t(-1);
  else
_M_next_resize
  = __builtin_ceil(__res * (long double)_M_max_load_factor);

  return __res;
}

因此，__max_width 的 auto 类型似乎被错误地推导出为 double 类型，而不是 size_t（应该可以从 std::min<size_t> 的类型中轻松推导出）。如果我复制标题，然后对其进行编辑以将auto 更改为size_t，则错误消失，确认这一点。

有趣的是，以下程序可以毫无问题地编译：

#include <iostream>
#include <algorithm>

int main()
{
    const auto __max_width = std::min<size_t>(sizeof(size_t), 8);
    const auto __max_bkt = size_t(1) << (__max_width * __CHAR_BIT__ - 1);

    std::cout << "__max_width: " << __max_width << std::endl;
    std::cout << "__max_bkt: " << __max_bkt << std::endl;

    return 0;
}

我在 Ubuntu 18.04 上，使用 gcc 7.4.0，虽然我在使用 clang 6.0.0 时遇到了同样的错误。我还确认这发生在 chroot 18.04 环境和另一台运行 WSL Ubuntu 18.04 的机器上。 HELICS 是使用 Python 接口和 HELICS_BUILD_CXX_SHARED_LIB 编译的，gridlab-d 是按照说明构建的。

很高兴提供更多信息，如果能在不要求用户修改其标准标题的情况下解决此问题，我们将不胜感激。

更新：这是根文件helics_msg.cpp 的预处理输出，以及它所抱怨的特定功能。我没有看到任何会导致错误的宏有趣的业务。完整文件位于here as a gist，sn-p 从第 72,722 行开始。它是一个大文件（>95k 行）。

    std::size_t
    _M_next_bkt(std::size_t __n) noexcept
    {
      const auto __max_width = std::
# 562 "/usr/include/c++/7/bits/hashtable_policy.h"
                                   fmin
# 562 "/usr/include/c++/7/bits/hashtable_policy.h" 3
                                      <size_t>(sizeof(size_t), 8);
      const auto __max_bkt = size_t(1) << (__max_width * 8 - 1);
      std::size_t __res = __clp2(__n);

      if (__res == __n)
 __res <<= 1;

      if (__res == 0)
 __res = __max_bkt;

      if (__res == __max_bkt)



 _M_next_resize = std::size_t(-1);
      else
 _M_next_resize
   = __builtin_ceil(__res * (long double)_M_max_load_factor);

      return __res;
    }

Answer 1

gridlab-d redefines min to fmin:

#define min fmin /**< min macro */

因此，只要包含 platform.h，min 就会被 fmin 替换。这只是一个有根据的猜测，因为我手头没有确切的包含图。

很遗憾，std::fmin will always return a double on integral values。这是 gridlab-d 的一个缺陷，应该报告——as you already did。