Add TYPE_SAFE_MSC_EMPTY_BASES to all strong_typedef operations

[ZeroMemes]

Expands upon the PR (#128) by adding the TYPE_SAFE_MSC_EMPTY_BASES macro to all strong_typedef operations. This Fixes an issue where the order of inheritance would affect the size of the derived strong_typedef, for example:

using namespace type_safe;

struct A :
    strong_typedef<A, uint16_t>,
    strong_typedef_op::relational_comparison<A>,
    strong_typedef_op::equality_comparison<A>,
    strong_typedef_op::integer_arithmetic<A>
{
    using strong_typedef::strong_typedef;
};
static_assert(sizeof(A) == 2);

In this specific configuration, the code fails to compile due to sizeof(A) actually evaluating to 4. However, transposing the order of equality_comparison and integer_arithmetic has the surprising result of the static_assert succeeding with a class size equal to 2! I believe this rule comes down to the following statement from the Microsoft documentation

However, the default class layout in Visual Studio doesn't take advantage of EBCO in multiple inheritance scenarios [...] The class layout algorithm is adding 1 byte of padding between any two consecutive empty base classes [...]

The following snippet is a more generic case which shows that my proposed patch works, as it only compiles when __declspec(empty_bases) is applied such that no two consecutive base classes are without the modifier.

struct C {
    uint16_t val;
};
// __declspec(empty_bases) must be applied to D and F *or* E
// in order for the static_assert to pass
struct D {};
struct E {};
struct F {};
struct G : C, D, E, F {};

static_assert(sizeof(G) == 2);

[foonathan]

Good idea, thanks.

[BrukerJWD]

@foonathan, are you going to release a new version?

[foonathan]

I've tagged a new release.