可以在swift中创建静态分配的数组吗？

Question

I want to create a struct in swift that has a small fixed number of values (say 16 floats) as instance data. It is required that this struct not store these values on the heap, so that the address of an instance of the struct is the address of the instance vars. It is also a requirement that these values be accessible internally to the struct via subscript, as Arrays are.

In C you would simply define this kind of thing thusly:

struct Matrix4x4 {
    float elements[16];
    ...
} myMatrix;

With this code, sizeof(Matrix4x4) == 64 and also &myMatrix == &myMatrix.elements[0]; In swift, if I analogously define the elements variable as type [Float] , the matrix instance only contains a pointer to the array, since the Array<Float> instance is an object stored on the heap.

Is there a way in swift to get static allocation of the instance vars without abandoning the convenience and efficiency of array-like subscripting access?

Answer 1

At present, this is not possible in "pure Swift". There is a long discussion on the swift-evolution mailing list starting at

• [swift-evolution] Proposal: Contiguous Variables (AKA Fixed Sized Array Type)

which asks for such a feature, eg to pass a matrix structure to C functions. As far as I can see, the suggestion was well-received, but nothing concrete is planned as of now, and it is not listed in the currently active Swift proposals .

AC array

float elements[16];

is imported to Swift as a tuple with 16 components:

public var elements: (Float, Float, Float, Float, Float, Float, Float, Float, Float, Float, Float, Float, Float, Float, Float, Float)

and at present this seems to be the only way to define a fixed-sized structure with a given memory layout. Joe Groff from Apple writes at [swift-users] Mapping C semantics to Swift

Swift structs have unspecified layout. If you depend on a specific layout, you should define the struct in C and import it into Swift for now.

and later in that discussion :

You can leave the struct defined in C and import it into Swift. Swift will respect C's layout.

If the matrix type is defined in a C header file (for the sake of simplicity I am using a 2x2 matrix as example now)

// matrix.h:
typedef struct Matrix2x2 {
    float elements[4];
} Matrix2x2;

then it is imported to Swift as

public struct Matrix2x2 {
    public var elements: (Float, Float, Float, Float)
    public init()
    public init(elements: (Float, Float, Float, Float))
}

As mentioned above, Swift preserves the C memory layout, so that the matrix, its elements, and the first element all have the same address:

var mat = Matrix2x2(elements: (1, 2, 3, 4))
print(sizeofValue(mat)) // 16
withUnsafePointer(&mat) { print($0) }            // 0x00007fff5fbff808
withUnsafePointer(&mat.elements) { print($0) }   // 0x00007fff5fbff808
withUnsafePointer(&mat.elements.0) { print($0) } // 0x00007fff5fbff808

However, tuples are not subscriptable, and that makes sense if the tuple members have different types. There is another discussion on the swift-evolution mailing list

• [swift-evolution] CollectionType on uniform tuples [forked off Contiguous Variables]

to treat "uniform tuples" as collections, which would allow subscripting. Unfortunately, this hasn't been implemented yet.

There are some methods to access tuple members by index, eg using Mirror() or withUnsafe(Mutable)Pointer() .

Here is a possible solution for Swift 3 (Xcode 8) , which seems to work well and involves only little overhead. The "trick" is to define C functions which return a pointer to the element storage:

// matrix.h:

// Constant pointer to the matrix elements:
__attribute__((swift_name("Matrix2x2.pointerToElements(self:)")))
static inline const float * _Nonnull matrix2x2PointerToElements(const Matrix2x2 * _Nonnull mat)
{
    return mat->elements;
}

// Mutable pointer to the matrix elements:
__attribute__((swift_name("Matrix2x2.pointerToMutableElements(self:)")))
static inline float * _Nonnull pointerToMutableElements(Matrix2x2 * _Nonnull mat)
{
    return mat->elements;
}

We need two variants to make the proper value semantics work (subscript setter requires a variable, subscript getter works with constant or variable). The "swift_name" attribute makes the compiler import these functions as member functions of the Matrix2x2 type, compare

• SE-0044 Import as member

Now we can define the subscript methods in Swift:

extension Matrix2x2 {
    public subscript(idx: Int) -> Float {
        get {
            precondition(idx >= 0 && idx < 4)
            return pointerToElements()[idx]
        }
        set(newValue) {
            precondition(idx >= 0 && idx < 4)
            pointerToMutableElements()[idx] = newValue
        }
    }
}

and everything works as expected:

// A constant matrix:
let mat = Matrix2x2(elements: (1, 2, 3, 4))
print(mat[0], mat[1], mat[2], mat[3]) // 1.0 2.0 3.0 4.0

// A variable copy:
var mat2 = mat
mat2[0] = 30.0
print(mat2) // Matrix2x2(elements: (30.0, 2.0, 3.0, 4.0))

Of course you could also define matrix-like subscript methods

public subscript(row: Int, col: Int) -> Float

in a similar manner.

Answer 2

As the answer above alludes, you can use combo of withUnsafeMutableBytes() and assumingMemoryBound(to:) to treat the C array as a swift array within the scope of the call.

withUnsafeMutableBytes(of: &mymatrix.elements) { rawPtr in
        let floatPtr = rawPtr.baseAddress!.assumingMemoryBound(to: Float.self)
        // Use the floats (with no bounds checking)
        // ...
        for i in 0..<10 {
            floatPtr[i] = 42.0
        }
    }