Re: [lmi] reusing mc_enum<> for serialization into XML

[Greg Chicares]

On 2010-02-02 08:07Z, Vaclav Slavik wrote:
> 
> [...] Raw patch is here:
> http://review.bakefile.org/r/154/diff/raw/

http://svn.savannah.gnu.org/viewvc?view=rev&root=lmi&revision=4776

Here's a detail that deserves discussion (which I'll present in
terms of your original patch, for clarity). In 'mc_enum.tpp':

    BOOST_STATIC_ASSERT(sizeof(typename enum_info::enums_t) == 
sizeof(enum_info::enums()));
    BOOST_STATIC_ASSERT(sizeof(typename enum_info::strings_t) == 
sizeof(enum_info::strings()));
    BOOST_STATIC_ASSERT(enum_info::n == sizeof(enum_info::enums())/sizeof(T));
    BOOST_STATIC_ASSERT(enum_info::n == 
sizeof(enum_info::strings())/sizeof(const char*));

I think it's provably impossible for these assertions to fail,
because enum_info comes from here:

  template<typename T, std::size_t sz, T const (*e)[sz], char const*const 
(*c)[sz]>
  struct mc_enum_type_info_impl

where we probably wouldn't bother to write:

  BOOST_STATIC_ASSERT(sz == sizeof e / sizeof(T));
  BOOST_STATIC_ASSERT(sz == sizeof c / sizeof(char*));

If we made an undisciplined change that defeated the compile-time
checking of array sizes, then we could wind up with:

  some_type some_enums[] = {one, two}; // error--need three

  // Instantiated with sz = 3, whereas the array has sz = 2...
  template<typename T, std::size_t sz, T const (*e)[sz], char const*const 
(*c)[sz]>
  struct ...

but template argument 'e' would be an array of three elements, so it's
too late to check the cardinality there (and attempting to access the
third would cause a SIGSEGV).

However, do you feel that such assertions are valuable as documentation?