Interfacing to C++

This document specifies how to interface with C++ directly.

It is also possible to indirectly interface with C++ code, either through a C interface or a COM interface.

The General Idea

Being 100% compatible with C++ means more or less adding a fully functional C++ compiler front end to D. Anecdotal evidence suggests that writing such is a minimum of a 10 man-year project, essentially making a D compiler with such capability unimplementable. Other languages looking to hook up to C++ face the same problem, and the solutions have been:

1. Support the COM interface (but that only works for Windows).
2. Laboriously construct a C wrapper around the C++ code.
3. Use an automated tool such as SWIG to construct a C wrapper.
4. Reimplement the C++ code in the other language.
5. Give up.

D takes a pragmatic approach that assumes a couple modest accommodations can solve a significant chunk of the problem:

• matching C++ name mangling conventions
• matching C++ function calling conventions
• matching C++ virtual function table layout for single inheritance

Global Functions

C++ global functions, including those in namespaces, can be declared and called in D, or defined in D and called in C++.

Calling C++ Global Functions from D

Given a C++ function in a C++ source file:

#include <iostream>

using namespace std;

int foo(int i, int j, int k)
{
    cout << "i = " << i << endl;
    cout << "j = " << j << endl;
    cout << "k = " << k << endl;

    return 7;
}

In the corresponding D code, foo is declared as having C++ linkage and function calling conventions:

extern (C++) int foo(int i, int j, int k);

and then it can be called within the D code:

extern (C++) int foo(int i, int j, int k);

void main()
{
    foo(1, 2, 3);
}

Compiling the two files, the first with a C++ compiler, the second with a D compiler, linking them together, and then running it yields:

i = 1
j = 2
k = 3

There are several things going on here:

• D understands how C++ function names are "mangled" and the correct C++ function call/return sequence.
• Because modules are not part of C++, each function with C++ linkage in the global namespace must be globally unique within the program.
• There are no __cdecl, __far, __stdcall, __declspec, or other such nonstandard C++ extensions in D.
• There are no volatile type modifiers in D.
• Strings are not 0 terminated in D. See "Data Type Compatibility" for more information about this. However, string literals in D are 0 terminated.

Calling Global D Functions From C++

To make a D function accessible from C++, give it C++ linkage:

import std.stdio;

extern (C++) int foo(int i, int j, int k)
{
    writefln("i = %s", i);
    writefln("j = %s", j);
    writefln("k = %s", k);
    return 1;
}

extern (C++) void bar();

void main()
{
    bar();
}

The C++ end looks like:

int foo(int i, int j, int k);

void bar()
{
    foo(6, 7, 8);
}

Compiling, linking, and running produces the output:

i = 6
j = 7
k = 8

C++ Namespaces

C++ symbols that reside in namespaces can be accessed from D. A namespace can be added to the extern (C++) LinkageAttribute:

extern (C++, N) int foo(int i, int j, int k);

void main()
{
    N.foo(1, 2, 3);   // foo is in C++ namespace 'N'
}

Classes

C++ classes can be declared in D by using the extern (C++) attribute on class, struct and interface declarations. extern (C++) interfaces have the same restrictions as D interfaces, which means that Multiple Inheritance is supported to the extent that only one base class can have member fields.

extern (C++) structs do not support virtual functions but can be used to map C++ value types.

Unlike classes and interfaces with D linkage, extern (C++) classes and interfaces are not rooted in Object and cannot be used with typeid.

Using C++ Classes From D

The following example shows binding of a pure virtual function, its implementation in a derived class, a non-virtual member function, and a member field:

#include <iostream>

using namespace std;

class Base
{
    public:
        virtual void print3i(int a, int b, int c) = 0;
};

class Derived : public Base
{
    public:
        int field;
        Derived(int field) : field(field) {}

        void print3i(int a, int b, int c)
        {
            cout << "a = " << a << endl;
            cout << "b = " << b << endl;
            cout << "c = " << c << endl;
        }

        int mul(int factor);
};

int Derived::mul(int factor)
{
    return field * factor;
}

Derived *createInstance(int i)
{
    return new Derived(i);
}

void deleteInstance(Derived *&d)
{
    delete d;
    d = 0;
}

We can use it in D code like:

extern(C++)
{
    interface Base
    {
        void print3i(int a, int b, int c);
    }

    class Derived : Base
    {
        int field;
        @disable this();
        override void print3i(int a, int b, int c);
        final int mul(int factor);
    }

    Derived createInstance(int i);
    void deleteInstance(ref Derived d);
}

void main()
{
    import std.stdio;

    auto d1 = createInstance(5);
    writeln(d1.field);
    writeln(d1.mul(4));

    Base b1 = d1;
    b1.print3i(1, 2, 3);

    deleteInstance(d1);
    assert(d1 is null);

    auto d2 = createInstance(42);
    writeln(d2.field);

    deleteInstance(d2);
    assert(d2 is null);
}

Compiling, linking, and running produces the output:

5
20
a = 1
b = 2
c = 3
42

Note how in the above example, the constructor is not bindable and is instead disabled on the D side; an alternative would be to reimplement the constructor in D. See the section below on lifetime management for more information.

Using D Classes From C++

Given D code like:

extern (C++) int callE(E);

extern (C++) interface E
{
    int bar(int i, int j, int k);
}

class F : E
{
    extern (C++) int bar(int i, int j, int k)
    {
        writefln("i = %s", i);
        writefln("j = %s", j);
        writefln("k = %s", k);
        return 8;
    }
}

void main()
{
    F f = new F();
    callE(f);
}

The C++ code to access it looks like:

class E
{
  public:
    virtual int bar(int i, int j, int k);
};


int callE(E *e)
{
    return e->bar(11, 12, 13);
}

C++ Templates

C++ function and type templates can be bound by using the extern (C++) attribute on a function or type template declaration.

Note that all instantiations used in D code must be provided by linking to C++ object code or shared libraries containing the instantiations.

For example:

#include <iostream>

template<class T>
struct Foo
{
    private:
    T field;

    public:
    Foo(T t) : field(t) {}
    T get();
    void set(T t);
};

template<class T>
T Foo<T>::get()
{
    return field;
}

template<class T>
void Foo<T>::set(T t)
{
    field = t;
}

Foo<int> makeIntFoo(int i)
{
    return Foo<int>(i);
}

Foo<char> makeCharFoo(char c)
{
    return Foo<char>(c);
}

template<class T>
void increment(Foo<T> &foo)
{
    foo.set(foo.get() + 1);
}

template<class T>
void printThreeNext(Foo<T> foo)
{
    for(size_t i = 0; i < 3; ++i)
    {
        std::cout << foo.get() << std::endl;
        increment(foo);
    }
}

// The following two functions ensure that the required instantiations of
// printThreeNext are provided by this code module
void printThreeNexti(Foo<int> foo)
{
    printThreeNext(foo);
}

void printThreeNextc(Foo<char> foo)
{
    printThreeNext(foo);
}

extern(C++):
struct Foo(T)
{
    private:
    T field;

    public:
    @disable this();
    T get();
    void set(T t);
}

Foo!int makeIntFoo(int i);
Foo!char makeCharFoo(char c);
void increment(T)(ref Foo!T foo);
void printThreeNext(T)(Foo!T foo);

extern(D) void main()
{
    auto i = makeIntFoo(42);
    assert(i.get() == 42);
    i.set(1);
    increment(i);
    assert(i.get() == 2);

    auto c = makeCharFoo('a');
    increment(c);
    assert(c.get() == 'b');

    c.set('A');
    printThreeNext(c);
}

Compiling, linking, and running produces the output:

A
B
C

Note: the above does not currently link correctly due to Bugzilla 15372.

Function Overloading

C++ and D follow different rules for function overloading. D source code, even when calling extern (C++) functions, will still follow D overloading rules.

Memory Allocation

C++ code explicitly manages memory with calls to ::operator new() and ::operator delete(). D's new operator allocates memory using the D garbage collector, so no explicit delete is necessary. D's new operator is not compatible with C++'s ::operator new and ::operator delete. Attempting to allocate memory with D's new and deallocate with C++ ::operator delete will result in miserable failure.

D can explicitly manage memory using a variety of library tools, such as with std.experimental.allocator. Additionally, core.stdc.stdlib.malloc and core.stdc.stdlib.free can be used directly for connecting to C++ functions that expect malloc'd buffers.

If pointers to memory allocated on the D garbage collector heap are passed to C++ functions, it's critical to ensure that the referenced memory will not be collected by the D garbage collector before the C++ function is done with it. This is accomplished by:

• Making a copy of the data using std.experimental.allocator or core.stdc.stdlib.malloc and passing the copy instead.
• Leaving a pointer to it on the stack (as a parameter or automatic variable), as the garbage collector will scan the stack.
• Leaving a pointer to it in the static data segment, as the garbage collector will scan the static data segment.
• Registering the pointer with the garbage collector using the core.memory.GC.addRoot or core.memory.GC.addRange functions.

An interior pointer to the allocated memory block is sufficient to let the GC know the object is in use; i.e. it is not necessary to maintain a pointer to the beginning of the allocated memory.

The garbage collector does not scan the stacks of threads not registered with the D runtime, nor does it scan the data segments of shared libraries that aren't registered with the D runtime.

Data Type Compatibility

These equivalents hold when the D and C++ compilers used are companions on the host platform.

Packing and Alignment

D structs and unions are analogous to C's.

C code often adjusts the alignment and packing of struct members with a command line switch or with various implementation specific #pragma's. D supports explicit alignment attributes that correspond to the C compiler's rules. Check what alignment the C code is using, and explicitly set it for the D struct declaration.

D supports bitfields in the standard library: see std.bitmanip.bitfields.

Lifetime Management

C++ constructors, copy constructors, move constructors and destructors cannot be called directly in D code, and D constructors, postblit operators and destructors cannot be directly exported to C++ code. Interoperation of types with these special operators is possible by either 1) disabling the operator in the client language and only using it in the host language, or 2) faithfully reimplementing the operator in the client language. With the latter approach, care needs to be taken to ensure observable semantics remain the same with both implementations, which can be difficult, or in some edge cases impossible, due to differences in how the operators work in the two languages. For example, in D all objects are movable and there is no move constructor.

Special Member Functions

D cannot directly call C++ special member functions, and vice versa. These include constructors, destructors, conversion operators, operator overloading, and allocators.

Runtime Type Identification

D runtime type identification uses completely different techniques than C++. The two are incompatible.

Exception Handling

Exception interoperability is a work in progress.

At present, C++ exceptions cannot be caught in or thrown from D, and D exceptions cannot be caught in or thrown from C++. Additionally, objects in C++ stack frames are not guaranteed to be destroyed when unwinding the stack due to a D exception, and vice versa.

The plan is to support all of the above except throwing D exceptions directly in C++ code (but they will be throwable indirectly by calling into a D function with C++ linkage).

Comparing D Immutable and Const with C++ Const

// Functional:
//ptr to const ptr to const int
const(int*)* p;

// Postfix:
//ptr to const ptr to const int
const int *const *p;

// Yes:
//const ptr to const ptr to const int
const int** p;
**p = 3; // error

// No:
// const ptr to ptr to int
int** const p;
**p = 3;    // ok

// Yes:
// ptr to const int
const(int)* p;
int* q = cast(int*)p; // ok

// Yes:
// ptr to const int
const int* p;
int* q = const_cast<int*>p; //ok

// No:
// ptr to const int
const(int)* p;
int* q = cast(int*)p;
*q = 3;   // undefined behavior

// Yes:
// ptr to const int
const int* p;
int* q = const_cast<int*>p;
*q = 3;   // ok

// Yes:
void foo(int x);
void foo(const int x);  //ok

// No:
void foo(int x);
void foo(const int x);  //error

// Yes:
void foo(const int* x, int* y)
{
    bar(*x); // bar(3)
    *y = 4;
    bar(*x); // bar(4)
}
...
int i = 3;
foo(&i, &i);

// Yes:
void foo(const int* x, int* y)
{
    bar(*x); // bar(3)
    *y = 4;
    bar(*x); // bar(4)
}
...
int i = 3;
foo(&i, &i);

// No:
void foo(immutable int* x, int* y)
{
    bar(*x); // bar(3)
    *y = 4;  // undefined behavior
    bar(*x); // bar(??)
}
...
int i = 3;
foo(cast(immutable)&i, &i);