Does a phantom type have the same alignment as the original one?

问题

Consider the following struct that contains some environment values:

struct environment_values {
  uint16_t humidity;
  uint16_t temperature;
  uint16_t charging;
};

I would like to add some additional information to those values with a phantom type* and make their types distinct at the same time:

template <typename T, typename P>
struct Tagged {
    T value;
};

// Actual implementation will contain some more features
struct Celsius{};
struct Power{};
struct Percent{};

struct Environment {
  Tagged<uint16_t,Percent> humidity;
  Tagged<uint16_t,Celsius> temperature;
  Tagged<uint16_t,Power>   charging;
};

Is the memory-layout of Environment the same as environment_values? Does this also hold for mixed type layouts, e.g.:

struct foo {
    uint16_t value1;
    uint8_t  value2;
    uint64_t value3;
}

struct Foo {
    Tagged<uint16_t, Foo>  Value1;
    Tagged<uint8_t , Bar>  Value2;
    Tagged<uint64_t, Quux> Value3;
}

For all types I've tried so far, the following assertions held:

template <typename T, typename P = int>
constexpr void check() {
    static_assert(alignof(T) == alignof(Tagged<T,P>), "alignment differs");
    static_assert(sizeof(T)  == sizeof(Tagged<T,P>),  "size differs");
}

// check<uint16_t>(), check<uint32_t>(), check<char>() …

Since the size of the tagged and untagged variants is also the same, I guess the answer should be yes, but I would like to have some certainty.

^{* I have no idea how those tagged values are called in C++. "Strongly typed typedefs"? I've taken the name from Haskell.}

回答1:

The Standard mentions in [basic.align]/1:

Object types have alignment requirements (3.9.1, 3.9.2) which place restrictions on the addresses at which an object of that type may be allocated. An alignment is an implementation-defined integer value representing the number of bytes between successive addresses at which a given object can be allocated. An object type imposes an alignment requirement on every object of that type; stricter alignment can be requested using the alignment specifier (7.6.2).

Moreover, [basic.compound]/3, mentions:

The value representation of pointer types is implementation-defined. Pointers to layout-compatible types shall have the same value representation and alignment requirements (6.11). [Note: Pointers to over-aligned types (6.11) have no special representation, but their range of valid values is restricted by the extended alignment requirement].

As a result, there is a guarantee that layout-compatible types have the same alignment.

struct { T m; } and T are not layout-compatible.

As pointed here, in order for two elements to be layout compatible then they both have to be standard-layout types, and their non-static data members must occur with the same types and in the same order.

struct { T m; } contains just a T, but T is a T so it cannot contain a T as its first non-static data member.

回答2:

According to the letter of the law, size and alignment of types is implementation-defined and the standard gives you few if any guarantees about what sizeof and alignof will return.

template <typename T, typename P>
struct Tagged {
    T value;
};

In theory, the compiler is permitted to add padding to the end of this struct, which would obviously alter the size and probably the alignment as well. In practise, the only time I could envisage this happening is if T was given some sort of compiler-specific "packed" attribute, but Tagged was not (but even then, GCC seems to work okay).

In any case, I'd say it would be a good idea to add some static asserts to ensure that the compiler is being sensible -- which is exactly what you've done :).

回答3:

As mentioned by gsamaras, the standard guarantees the same alignment for layout compatible classes.

Unfortunately, struct { T m; } and T are not layout compatible.

In 12.2.21 the standard lays out the requirements for a layout compatible class:

Two standard-layout struct (Clause 12) types are layout-compatible classes if their common initial sequence comprises all members and bit-fields of both classes (6.9).

And the definition of common initial sequence is in 12.2.20:

The common initial sequence of two standard-layout struct (Clause 12) types is the longest sequence of non-static data members and bit-fields in declaration order, starting with the first such entity in each of the structs, such that corresponding entities have layout-compatible types and either neither entity is a bit-field or both are bit-fields with the same width. [Example:
struct A { int a; char b; };
struct B { const int b1; volatile char b2; };
struct C { int c; unsigned : 0; char b; };
struct D { int d; char b : 4; };
struct E { unsigned int e; char b; };
The common initial sequence of A and B comprises all members of either class. The common initial sequence of A and C and of A and D comprises the first member in each case. The common initial sequence of A and E is empty.
— end example]

So from this we can make the following important observations:

1. Layout compatibility is limited strictly to standard layout classes. (Or enums use the same underlying type or the trivial case when T and T2 are literally the exact same type. See 6.9.11.) In the general case, T is not a standard layout class. In fact, T is not even a class in your example (it is a uint16_t, believe it or not, this matters according to the standard.)*
2. Even if T is guaranteed to be a standard layout class, struct { T m; } does not have a common initial sequence with T. The sequence of struct { T m; } begins with T, whereas the sequence of T begins with whatever T's non-static data members are. This is actually strictly guaranteed not to be a T as a class cannot contain itself by value.

Therefore, the guarantee cannot be held by the letter of the standard. You should continue to perform the static_assertions to ensure your compiler is behaving in the fashion you expect.

* See most questions on union type punning.

来源：https://stackoverflow.com/questions/46425250/does-a-phantom-type-have-the-same-alignment-as-the-original-one