While D cannot directly compile C source code, it can easily interface to C code, be linked with C object files, and call C functions in DLLs. The interface to C code is normally found in C .h files. So, the trick to connecting with C code is in converting C .h files to D modules. This turns out to be difficult to do mechanically since inevitably some human judgement must be applied. This is a guide to doing such conversions.
Preprocessor
.h files can sometimes be a bewildering morass of layers of macros, #include files, #ifdef's, etc. D doesn't include a text preprocessor like the C preprocessor, so the first step is to remove the need for it by taking the preprocessed output. For DMC (the Digital Mars C/C++ compiler), the command:
dmc -c program.h -e -l
will create a file program.lst which is the source file after all text preprocessing.
Remove all the #if, #ifdef, #include, etc. statements.
Linkage
Generally, surround the entire module with:
extern (C)
{
/* ...file contents... */
}
to give it C linkage.
Types
A little global search and replace will take care of renaming the C types to D types. The following table shows a typical mapping for 32 bit C code:
Mapping C type to D type
C type D type
long double real
unsigned long long ulong
long long long
unsigned long uint
long int
unsigned uint
unsigned short ushort
signed char byte
unsigned char ubyte
wchar_t wchar or dchar
bool bool, byte, int
size_t size_t
ptrdiff_t ptrdiff_t
NULL
NULL and ((void*)0) should be replaced with null.
Numeric Literals
Any ‘L’ or ‘l’ numeric literal suffixes should be removed, as a C long is (usually) the same size as a D int. Similarly, ‘LL’ suffixes should be replaced with a single ‘L’. Any ‘u’ suffix will work the same in D.
String Literals
In most cases, any ‘L’ prefix to a string can just be dropped, as D will implicitly convert strings to wide characters if necessary. However, one can also replace:
L"string"
with:
"string"w // for 16 bit wide characters
"string"d // for 32 bit wide characters
Macros
Lists of macros like:
#define FOO 1
#define BAR 2
#define ABC 3
#define DEF 40
can be replaced with:
enum
{ FOO = 1,
BAR = 2,
ABC = 3,
DEF = 40
}
or with:
const int FOO = 1;
const int BAR = 2;
const int ABC = 3;
const int DEF = 40;
Function style macros, such as:
#define MAX(a,b) ((a) < (b) ? (b) : (a))
can be replaced with functions:
int MAX(int a, int b) { return (a < b) ? b : a; }
The functions, however, won't work if they appear inside static initializers that must be evaluated at compile time rather than runtime. To do it at compile time, a template can be used:
#define GT_DEPTH_SHIFT (0)
#define GT_SIZE_SHIFT (8)
#define GT_SCHEME_SHIFT (24)
#define GT_DEPTH_MASK (0xffU << GT_DEPTH_SHIFT)
#define GT_TEXT ((0x01) << GT_SCHEME_SHIFT)
/* Macro that constructs a graphtype */
#define GT_CONSTRUCT(depth,scheme,size) \
((depth) | (scheme) | ((size) << GT_SIZE_SHIFT))
/* Common graphtypes */
#define GT_TEXT16 GT_CONSTRUCT(4, GT_TEXT, 16)
The corresponding D version would be:
const uint GT_DEPTH_SHIFT = 0;
const uint GT_SIZE_SHIFT = 8;
const uint GT_SCHEME_SHIFT = 24;
const uint GT_DEPTH_MASK = 0xffU << GT_DEPTH_SHIFT;
const uint GT_TEXT = 0x01 << GT_SCHEME_SHIFT;
// Template that constructs a graphtype
template GT_CONSTRUCT(uint depth, uint scheme, uint size)
{
// notice the name of the const is the same as that of the template
const uint GT_CONSTRUCT = (depth | scheme | (size << GT_SIZE_SHIFT));
}
// Common graphtypes
const uint GT_TEXT16 = GT_CONSTRUCT!(4, GT_TEXT, 16);
Declaration Lists
D doesn't allow declaration lists to change the type. Hence:
int *p, q, t[3], *s;
should be written as:
int* p, s;
int q;
int[3] t;
Void Parameter Lists
Functions that take no parameters:
int foo(void);
are in D:
int foo();
Const Type Modifiers
D has const as a storage class, not a type modifier. Hence, just drop any const used as a type modifier:
void foo(const int *p, char *const q);
becomes:
void foo(int* p, char* q);
Extern Global C Variables
Whenever a global variable is declared in D, it is also defined. But if it's also defined by the C object file being linked in, there will be a multiple definition error. To fix this problem, use the extern storage class. For example, given a C header file named foo.h:
struct Foo { };
struct Foo bar;
It can be replaced with the D modules, foo.d:
struct Foo { }
extern (C)
{
extern Foo bar;
}
Typedef
alias is the D equivalent to the C typedef:
typedef int foo;
becomes:
alias int foo;
Structs
Replace declarations like:
typedef struct Foo
{ int a;
int b;
} Foo, *pFoo, *lpFoo;
with:
struct Foo
{ int a;
int b;
}
alias Foo* pFoo, lpFoo;
Struct Member Alignment
A good D implementation by default will align struct members the same way as the C compiler it was designed to work with. But if the .h file has some #pragma's to control alignment, they can be duplicated with the D align attribute:
#pragma pack(1)
struct Foo
{
int a;
int b;
};
#pragma pack()
becomes:
struct Foo
{
align (1):
int a;
int b;
}
Nested Structs
struct Foo
{
int a;
struct Bar
{
int c;
} bar;
};
struct Abc
{
int a;
struct
{
int c;
} bar;
};
becomes:
struct Foo
{
int a;
struct Bar
{
int c;
}
Bar bar;
}
struct Abc
{
int a;
struct
{
int c;
}
}
__cdecl, __pascal, __stdcall
int __cdecl x;
int __cdecl foo(int a);
int __pascal bar(int b);
int __stdcall abc(int c);
becomes:
extern (C) int x;
extern (C) int foo(int a);
extern (Pascal) int bar(int b);
extern (Windows) int abc(int c);
__declspec(dllimport)
__declspec(dllimport) int __stdcall foo(int a);
becomes:
export extern (Windows) int foo(int a);
__fastcall
Unfortunately, D doesn't support the __fastcall convention. Therefore, a shim will be needed, either written in C:
int __fastcall foo(int a);
int myfoo(int a)
{
return foo(int a);
}
and compiled with a C compiler that supports __fastcall and linked in, or compile the above, disassemble it with obj2asm and insert it in a D myfoo shim with inline assembler.
Comments
Post a Comment