C# call C++ DLL passing pointer-to-pointer argument

Question

Could you guys please help me solve the following issue? I have a C++ function dll, and it will be called by another C# application. One of the functions I needed is as follow:

struct DataStruct
{
    unsigned char* data;
    int len;
};

DLLAPI int API_ReadFile(const wchar_t* filename, DataStruct** outData);

I wrote the following code in C#:

class CS_DataStruct
{
    public byte[] data;
    public int len;
}

[DllImport("ReadFile.dll", CallingConvention = CallingConvention.Cdecl, CharSet = CharSet.Unicode)]
private static extern int API_ReadFile([MarshalAs(UnmanagedType.LPWStr)]string filename, ref CS_DataStruct data);

Unfortunately, the above code is not working... I guess that is due to the C++ func takes a pointer-to-pointer of DataStruct, while I just passed a reference of CS_DataStruct in.

May I know how can I pass a pointer-to-pointer to the C++ func? If it is not possible, is there any workaround? (the C++ API is fixed, so changing API to pointer is not possible)

Edit: Memory of DataStruct will be allocated by c++ function. Before that, I have no idea how large the data array should be. (Thanks for the comments below)

Solution 1

I used the following test implementation:

int API_ReadFile(const wchar_t* filename, DataStruct** outData)
{
    *outData = new DataStruct();
    (*outData)->data = (unsigned char*)_strdup("hello");
    (*outData)->len = 5;
    return 0;
}

void API_Free(DataStruct** pp)
{
    free((*pp)->data);
    delete *pp;
    *pp = NULL;
}

The C# code to access those functions are as follows:

[StructLayout(LayoutKind.Sequential)]
struct DataStruct
{
    public IntPtr data;
    public int len;
};

[DllImport("ReadFile.dll", CallingConvention = CallingConvention.Cdecl, CharSet = CharSet.Unicode)]
unsafe private static extern int API_ReadFile([MarshalAs(UnmanagedType.LPWStr)]string filename, DataStruct** outData);

[DllImport("ReadFile.dll", CallingConvention = CallingConvention.Cdecl)]
unsafe private static extern void API_Free(DataStruct** handle);

unsafe static int ReadFile(string filename, out byte[] buffer)
{
    DataStruct* outData;
    int result = API_ReadFile(filename, &outData);
    buffer = new byte[outData->len];
    Marshal.Copy((IntPtr)outData->data, buffer, 0, outData->len);
    API_Free(&outData);
    return result;
}

static void Main(string[] args)
{
    byte[] buffer;
    ReadFile("test.txt", out buffer);
    foreach (byte ch in buffer)
    {
        Console.Write("{0} ", ch);
    }
    Console.Write("\n");
}

The data is now transferred to buffer safely, and there should be no memory leaks. I wish it would help.

Solution 2

It isn't necessary to use unsafe to pass a pointer to an array from a DLL. Here is an example (see the 'results' parameter). The key is to use the ref attribute. It also shows how to pass several other types of data.

As defined in C++/C:

#ifdef __cplusplus
extern "C" {
#endif

#ifdef BUILDING_DLL
#define DLLCALL __declspec(dllexport)
#else
#define DLLCALL __declspec(dllimport)
#endif

static const int DataLength = 10;
static const int StrLen = 16;
static const int MaxResults = 30;

enum Status { on = 0, off = 1 };

struct Result {
    char name[StrLen]; //!< Up to StrLen-1 char null-terminated name
    float location;  
    Status status;
};

/**
* Analyze Data
* @param data [in] array of doubles
* @param dataLength [in] number of floats in data
* @param weight [in]
* @param status [in] enum with data status
* @param results  [out] array of MaxResults (pre-allocated) DLLResult structs.
*                    Up to MaxResults results will be returned.
* @param nResults  [out] the actual number of results being returned.
*/
void DLLCALL __stdcall analyzeData(
      const double *data, int dataLength, float weight, Status status, Result **results, int *nResults);

#ifdef __cplusplus
}
#endif

As used in C#:

private const int DataLength = 10;
private const int StrLen = 16;
private const int MaxThreatPeaks = 30;

public enum Status { on = 0, off = 1 };

[StructLayout(LayoutKind.Sequential, CharSet = CharSet.Ansi)]
public struct Result
{
    [MarshalAs(UnmanagedType.ByValTStr, SizeConst = StrLen)] public string name; //!< Up to StrLen-1 char null-terminated name 
    public float location;  
    public Status status;       
}

[DllImport("dllname.dll", CallingConvention = CallingConvention.StdCall, EntryPoint = "analyzeData@32")] // "@32" is only used in the 32-bit version.
public static extern void analyzeData(
    double[] data,
    int dataLength, 
    float weight, 
    Status status,
    [MarshalAs(UnmanagedType.LPArray, SizeConst = MaxResults)] ref Result[] results, 
    out int nResults
);

Without the extern "C" part, the C++ compiler would mangle the export name in a compiler dependent way. I noticed that the EntryPoint / Exported function name matches the function name exactly in a 64-bit DLL, but has an appended '@32' (the number may vary) when compiled into a 32-bit DLL. Run dumpbin /exports dllname.dll to find the exported name for sure. In some cases you may also need to use the DLLImport parameter ExactSpelling = true. Note that this function is declared __stdcall. If it were not specified, it would be __cdecl and you'd need CallingConvention.Cdecl.

Here is how it might be used in C#:

Status status = Status.on;
double[] data = { -0.034, -0.05, -0.039, -0.034, -0.057, -0.084, -0.105, -0.146, -0.174, -0.167};
Result[] results = new Result[MaxResults];
int nResults = -1; // just to see that it changes (input value is ignored)
analyzeData(data, DataLength, 1.0f, status, ref results, out nResults);

Comments

[LennonLam]

Thanks for the reply but it is not working. When LayoutKind changed to Explicit, compiler says "'DataStruct.data' instance field types marked with StructLayout(LayoutKind.Explicit) must have a FieldOffset attribute"

[Random]

Yongwei Wu - One question here, since Charset is Unicode, single char is not necessarily a size of a byte, am I correct?

[Yongwei Wu]

The size of char is always the size of byte. Charset=Unicode means System.String is marshalled as wchar_t* (instead of char*) by default. Actually Charset=Unicode should have no effect here, as string is explicitly marshalled as UnmanagedType.LPWStr.