[clang-tidy] argument to operator delete is the address of a local stack variable, which is not memory allocated by 'new'

[matlo607]

clang-tidy on a project brought this to my attention:

/usr/lib/gcc/x86_64-redhat-linux/7/../../../../include/c++/7/ext/new_allocator.h:125:2: error: Argument to operator delete is the address of a local stack variable, which is not memory allocated by 'new' [clang-analyzer-cplusplus.NewDelete,-warnings-as-errors]
        ::operator delete(__p);
        ^
...
my_private_include_path/include/boost/multiprecision/number.hpp:2005:25: note: Calling 'number::str'
   std::string     s  = r.str(d, os.flags());
                        ^
my_private_include_path/include/boost/multiprecision/number.hpp:808:14: note: Calling 'cpp_int_backend::str'
      return m_backend.str(digits, f);
             ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1920:14: note: Calling 'cpp_int_backend::do_get_string'
      return do_get_string(f, trivial_tag());
             ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1822:11: note: Assuming the condition is false
      if ((f & std::ios_base::oct) == std::ios_base::oct)
          ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1822:7: note: Taking false branch
      if ((f & std::ios_base::oct) == std::ios_base::oct)
      ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1824:16: note: Assuming the condition is false
      else if ((f & std::ios_base::hex) == std::ios_base::hex)
               ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1824:12: note: Taking false branch
      else if ((f & std::ios_base::hex) == std::ios_base::hex)
           ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1830:11: note: 'base' is not equal to 8
      if (base == 8 || base == 16)
          ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1830:11: note: Left side of '||' is false
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1830:24: note: 'base' is not equal to 16
      if (base == 8 || base == 16)
                       ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1830:7: note: Taking false branch
      if (base == 8 || base == 16)
      ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1876:10: note: Taking false branch
         if (t.sign())
         ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1881:14: note: Assuming the condition is false
         if (this->size() == 1)
             ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1881:10: note: Taking false branch
         if (this->size() == 1)
         ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1889:13: note: Loop condition is false. Execution continues on line 364
            while (eval_get_sign(t) != 0)
            ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1904:10: note: Calling implicit destructor for 'cpp_int_backend<0, 0, boost::multiprecision::signed_magnitude, boost::multiprecision::unchecked, std::allocator<unsigned long long>>'
         }
         ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:1904:10: note: Calling '~cpp_int_base'
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:364:11: note: Assuming field 'm_internal' is false
      if (!m_internal)
          ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:364:7: note: Taking true branch
      if (!m_internal)
      ^
my_private_include_path/include/boost/multiprecision/cpp_int.hpp:365:10: note: Calling 'new_allocator::deallocate'
         allocator().deallocate(limbs(), capacity());
         ^
/usr/lib/gcc/x86_64-redhat-linux/7/../../../../include/c++/7/ext/new_allocator.h:125:2: note: Argument to operator delete is the address of a local stack variable, which is not memory allocated by 'new'
        ::operator delete(__p);
        ^

I am not sure but it seems that clang-tidy misinterprets the call to the custom allocator scoped_shared_storage.

[boblytton]

I have the same issue: clang-analyzer-cplusplus.NewDelete & clang-analyzer-cplusplus.NewDeleteLeaks

[jzmaddock]

Someone is going to have to convince me that there is a code path that results in m_internal == false and the memory not allocated via the allocator, because I just don't see how that can happen right now.