Re-add Boost.Array where it is ODR-used

[13steinj]

Determining the size of a type requires its definition; of which there is a "not-C++11-compliant" use and a C++11-agnostic use in .../converter_lexical_streams.hpp.

This was mistakenly forward-declared as part of af5ce2a.

Specifically, sizeof usages here:

• af5ce2a#diff-a0b72e72f56fd9c8d25429567b14e23ef6b31f5091f3377428b8d6661798983eR473-R478

• af5ce2a#diff-a0b72e72f56fd9c8d25429567b14e23ef6b31f5091f3377428b8d6661798983eL684-R687

In case I picked usages that aren't in develop's head (as ctrl-f'ing that file shows more usages than I expect), here's current usage as well:

lexical_cast/include/boost/lexical_cast/detail/converter_lexical_streams.hpp

Lines 446 to 456 in af5ce2a

	// Adding constness to characters. Constness does not change layout
	template <class C, std::size_t N>
	typename boost::disable_if<boost::is_const<C>, bool>::type
	operator<<(boost::array<C, N> const& input) noexcept {
	static_assert(
	sizeof(boost::array<const C, N>) == sizeof(boost::array<C, N>),
	"boost::array<C, N> and boost::array<const C, N> must have exactly the same layout."
	);
	return ((*this) << reinterpret_cast<boost::array<const C, N> const& >(input));
	}

lexical_cast/include/boost/lexical_cast/detail/converter_lexical_streams.hpp

Lines 473 to 485 in af5ce2a

	#ifndef BOOST_NO_CXX11_HDR_ARRAY
	// Making a Boost.Array from std::array
	template <class C, std::size_t N>
	bool operator<<(std::array<C, N> const& input) noexcept {
	static_assert(
	sizeof(std::array<C, N>) == sizeof(boost::array<C, N>),
	"std::array and boost::array must have exactly the same layout. "
	"Bug in implementation of std::array or boost::array."
	);
	return ((*this) << reinterpret_cast<boost::array<C, N> const& >(input));
	}
	#endif
	template <class InStreamable>

Edit: changed GitHub /blame/ -> /blob/ since GitHub renders /blob/ nicely in comments/PRs.

Interestingly enough; in the "non-C++11-compliant" case the ifndef / function was not removed; so the right answer might be to remove that function to drop down to a single usage and then remove that usage or put it as part of some static-compile test instead of adding arrays back. I don't remember if a reinterpret cast is an odr-use.

[apolukhin]

Thanks for the PR and issue report!

The whole reinterpret_casting of arrays has a bad smell. Rewriting that part of the library in #68

[apolukhin]

Fixed in 4bf37fb

Thanks again for the report!