Marshaling with C# - Chapter 2: Marshaling Simple Types


Read the full book here.

Overview

This chapter discusses the nitty-gritty part of marshaling process. It is the base for the rest of discussion about marshaling. It is about marshaling simple data types.

The first section of this chapter breaks data types into two categories, simple and compound. Simple types (integers, booleans, etc.) are those that are not made of other types. On the contrary, compound types (structures and classes) are those types that require special handling and made of other types.

After that, we will dig into the discussion of simple types and we will break them into two categories, blittable and non-blittable.

Before we end this chapter, we will discuss the passing mechanism and handles in .NET Framework.

Simple and Compound Data Types

There are two kinds of data types:

  • Simple (primitive/basic)
  • Compound (complex)
Primitive data types are those that are not defined in terms of other data types. They are the basis for all other types. Examples of managed primitives are numbers like System.Byte, System.Int32, System.UInt32, and System.Double, strings like System.Char and System.String, and handles like System.IntPtr.

Compound data types are those that built up of other data types. For example a class or a structure that encapsulates simple types and other compound types.

We will use terms simple, primitive, and basic types to refer to base types like integers, strings, etc. Terms compound, and complex types also will be used interchangeably to refer to classes and structures.

Some considers that strings are not primitives.

Blittable and Non-Blittable Data Types

Most data types have common representations in both managed and unmanaged memory and do not require special handling. These types are called blittable types because they do not require special handling when passed between managed and unmanaged code. Other types that require special handling are called non-blittable types. You can think that most of simple types are blittable and all of compound types are non-blittable.

The following table lists the blittable data types exist in .NET (their counterparts in unmanaged code will be covered soon):

Table 2.1 Blittable Types
 
Description Managed Type
System.SByte 8-bit signed integer.
System.Byte 8-bit unsigned integer
System.Int16 16-bit signed integer.
System.UInt16 16-bit unsigned integer
System.Int32 32-bit signed integer
System.UInt32 32-bit unsigned integer
System.Int64 64-bit signed integer
System.UInt64 64-bit unsigned integer
System.IntPtr Signed pointer
System.UIntPtr Unsigned pointer

More information about pointers later in this chapter.

Marshaling Blittable Data Types

You can marshal an unmanaged simple data type by tracking its definition then finding its counterpart (marshaling type) in the managed environment based on its definition (we will see how soon.)

Numeric Data Types

The following table lists some of the unmanaged data types in Windows, their C/C++ keywords, and their counterparts (marshaling types) in .NET. As you might guess, by tracking each of these unmanaged types, we were able to find its managed counterpart. Notice that so

Table 2.2 Numeric Data Types
Description Windows Type C/C++ Keyword Managed Type C# Keyword
8-bit unsigned integer BYTE unsigned char System.Byte Byte
16-bit signed integer SHORT Short System.UInt16 ushort
16-bit unsigned integer WORD and USHORT unsigned short System.Int16 short
32-bit signed integer INT, INT32, LONG, and LONG32 int, long System.UInt32 Uint
32-bit unsigned integer DWORD, DWORD32, UINT, and UINT32 unsigned int, unsigned long System.Int32 int
64-bit signed integer INT64, LONGLONG, and LONG64 __int64, long long System.UInt64 ulong
64-bit unsigned integer DWORDLONG, DWORD64, ULONGLONG, and UINT64 unsigned __int64, unsigned long long System.Int64 long
Floating-point integer FLOAT float System.Double double

Notice that long and int defer from a platform to another and from a compiler to another. In 32-bit versions of Windows, most compilers refer to both long and int as 32-bit integers.

Know that there is no difference between Windows data types and C/C++ data types. Windows data types are just aliases for the actual C types.

Do not be confused with the many types that refer to one thing, they are all just names (aliases.) INT, INT32, LONG, and LONG32 are all 32-bit integers for instance.

To keep things simple, we will focus on Windows API in our examples.

Although, some unmanaged types have names similar to names of some managed types, they have different meanings. An example is LONG, it has similar name as System.Long. However, LONG is 32-bit and System.Long is 64-bit!

If you need to learn more about these types, check out the article "Windows Data Types" in MSDN library.

Textual Data Types

In addition to the numeric data types, you will need to know how to marshal unmanaged textual data types (a single character or a string.) However, these types are non-blittable, so they require special handling.

The following table lists briefly unmanaged textual data types.

Table 2.3 Textual Data Types
Description Unmanaged Type(s) Managed Type
8-bit ANSI character CHAR System.Char
16-bit Unicode character WCHAR System.Char
8-bit ANSI string of characters LPSTR, LPCSTR, PCSTR, and PSTR System.String
16-bit Unicode string of characters LPCWSTR, LPWSTR, PCWSTR, and PWSTR System.String

Soon we will cover textual data types in details.

Examining Type Definition

As we have said, for the sake of simplicity, we will use Windows API as the base for our discussion in this book. Therefore, you need to know that all Windows Data Types (INT, DWORD, etc.) are just names (technically, typedefs) for the actual C types. Therefore, many names may refer to one thing just as INT and LONG.

Thus, we can say that LONG is defined as C int and DWORD is defined as C unsigned long.

INT and LONG are easy to marshal. However, there are primitive types that you will need to track their definitions to know how to marshal it.

Remember that we will use MSDN documentation (specially the article "Windows Data Types") when tracking unmanaged data types (Windows data types specially.)

The next are some of the types defined as another types. You can think of these types as aliases for the base types. Yet, some are platform-specific, and others not.
  • HRESULT: As you will see, plenty of functions return a HRESULT to represent the status of the operation. If HRESULT equals to zero, then the function succeeded, otherwise it represents the error code or status information for the operation. HRESULT defined as LONG, and LONG in turn defined as a 32-bit signed integer. Therefore, you can marshal HRESULT as System.Int32.
     
  • BOOL and BOOLEAN: Both are Boolean types, that means that they take either TRUE (non-zero) or FALSE (zero.) The big difference between BOOL and BOOLEAN is that BOOL is defined as INT, thus occupies 4 bytes. BOOLEAN on the other hand is defined as BYTE, thus occupies only 1 byte. Booleans are covered soon.
     
  • HFILE: A handle to a file opened using one of the Windows File IO functions like OpenFile() function. This type is defined as INT, and INT in turn is defined as a 32-bit signed integer. Therefore, you can marshal HFILE as System.Int32. Although, HFILE defined as INT, handles should be marshaled as System.IntPtr, which is internally encapsulates the raw handle. To be clear, you would better marshal an unmanaged handle as a System.Runtime.InteropServices.SafeHandle or CriticalHandle, this is the ideal marshaling type for any handle. Hence, file handles best marshaled as Microsoft.Win32.SafeHandles.SafeFileHandle that is derived from SafeHandleZeroOrMinusOneIsInvalid that is in turn derived from the abstract class System.Runtime.InteropServices.SafeHandle. For more details about handles, refer to the section "Marshaling Handles" later in this chapter.

In addition, there are types that are variable based on the operating system. Examples are:
  • INT_PTR: A pointer to a signed integer. Defined as INT64 if this is a 64-bit OS, or INT otherwise.
  • LONG_PTR: A pointer to a signed long. Defined as INT64 if this is a 64-bit OS, or LONG otherwise.
  • UINT_PTR: A pointer to an unsigned integer. Defined as DWORD64 if this is a 64-bit OS, or DWORD otherwise.
  • ULONG_PTR: A pointer to an unsigned long. Defined as DWORD64 if this is a 64-bit OS, or DWORD otherwise.

Keep in mind that there is a big difference between a variable and a pointer to a variable. A variable refers directly to its value into the memory. However, a pointer contains an address of another value into the memory. Consider the following illustration, Figure 2.1:

marsheling1.jpg

In the illustration above, the variable I contains the value 320 and you can get the value from the variable directly. The pointer ptr on the other hand contains the address of the variable I. Thus, it indirectly contains the value of the variable i. That is why we cannot get the value of the pointer directly. We need to dereference it first before retrieving its value.

More on pointers later in this chapter. Memory management is discussed in details in chapter 6.

In addition, for textual data types, there are types variable based on Unicode definition (strings and buffers are covered soon.) Examples are:
  • TBYTE and TCHAR: Defined as WCHAR if UNICODE defined, otherwise CHAR.
  • LPCTSTR, LPTSTR, and PCTSTR: All defined as LPCWSTR if UNICODE defined, otherwise LPCSTR.
  • PTSTR: Defined as PWSTR if UNICODE defined, otherwise PSTR.
More on textual data types and Unicode later in this chapter.

Notice that some types have special characters in their names. For example, A in textual data types stands for ANSI, and W in stands for Wide, which means Unicode. In addition, the letter T in textual information too means it varies based on OS. Another example is the prefix P (lowercase,) it means a pointer, and LP means a long pointer. LPC stands for long pointer to a constant.

Variants

In addition, Win32 API defines the types VOID, LPVOID, and LPCVOID. VOID indicates that the function does accept no arguments. Consider the following function:

DWORD GetVersion(VOID);

It is required to tag the function with VOID if it does not accept any arguments (that is one of the specifications of C89.) Notice that VOID is defined as void.

LPVOID and LPCVOID are defined as any type (variant). That means that they can accept any value. They can be marshaled as integers, strings, handles, or even compound types, anything you want. In addition, you can marshal them as System.IntPtr, so you can set them to the address of any object in memory. In addition, you can marshal them as pointers to object. For example, marshaling a LPCVOID as System.Int32* (a pointer to an integer) in unsafe code. Moreover, you can use unsafe code and marshal them as void*. Furthermore, you can marshal them as System.Object, so you can set them to any type (refer to chapter 6 for more information about memory management and unsafe code.)

It is worth mentioning that when working with VOIDs it is recommended decorating your variable with MarshalAsAttribute attribute specifying UnmanagedType.AsAny which tells the compiler to work out the marshaling process and sets the type of the argument at runtime. Refer to the last chapter: "Controlling the Marshaling Process" for more information about this attribute.

If you have worked with traditional Visual Basic, thinking about LPVOID and LOCVOID as a Variant could help too much.

If you are interoperating with the traditional Visual Basic code, you can use the same way we did on marshaling LPVOID and LPCVOID in marshaling the type Variant.

Try It Out!

Now, we will try to create the PInvoke method for the MessageBoxEx() function. The example demonstrates how to control precisely the marshaling process using the MarshalAsAttribute attribute. We will cover this attribute and more in the last chapter of this book: "Controlling the Marshaling Process." Handles are covered in the section: "Marshaling Handles" of this chapter.

The following example creates the PInvoke method for the MessageBoxEx() function and calls it to display a friendly message to the user.

The definition of the MessageBoxEx() function is as following:

Listing 2.1 MessageBoxEx() Unmanaged Signature

        int MessageBoxEx
        ( HWND hWnd,LPCTSTR lpText,LPCTSTR lpCaption,UINT uType,WORD wLanguageId);

And here is the managed signature (the PInvoke method) of this function:

In order for the example to run you must add a using statement to System.Runtime.InteropServices namespace. Be sure to add it for all examples throughout this book.

Listing 2.2 MessageBoxEx() Managed Signature

// CharSet.Unicode defines the UNICODE.
// Use either this way to control
// the whole function, or you can control
// the parameters individually using the
// MarshalAsAttribute attribute
[DllImport("User32.dll", CharSet = CharSet.Unicode)]
[return: MarshalAs(UnmanagedType.I4)]
static extern Int32 MessageBoxEx
(IntPtr hWnd,
// Marshaling as Unicode characters
[param: MarshalAs(UnmanagedType.LPTStr)]
String lpText,
// Marshaling as Unicode characters
[param: MarshalAs(UnmanagedType.LPTStr)]
String lpCaption,
// Marshaling as 4-bytes (32-bit) unsigned integer
[param: MarshalAs(UnmanagedType.U4)]
UInt32 uType,
// Marshaling as 2-bytes (16-bit) unsigned integer
[param: MarshalAs(UnmanagedType.U2)]
UInt16 wLanguageId);

For more information about marshaling strings, see section "Marshaling Strings and Buffers" later in this chapter.

A Rule of Thumb

Keep in mind that. .NET Framework allows you to take a granular level of control over the marshaling process and that would be very complicated. However, things can be so simple.

You can ignore attributes in most cases and just use the counterparts and CLR will do its best. Likely, you are not required to use managed signed integers for unmanaged equivalents. You can use managed signed integers for unmanaged unsigned integers and vice versa. You can also marshal a SHORT as System.Char!

The key point is that as long as the managed marshal type occupies the same memory size as the unmanaged type, you are in safe. However, keeping things in its right position helps avoiding undesirable errors that maybe very difficult to know and handle.

Another thing that you should keep in mind that the information in this book can be applied to any unmanaged environment. You can apply this information when interoperating with Windows API, C/C++ libraries, Visual Basic, COM, OLE, ActiveX, etc. However, for the sake of simplicity, we will talk about the Windows API as the source of the unmanaged code.

Marshaling Booleans

The Two Types

In general, marshaling simple data types is very easy and booleans are no exception. However, Booleans are non-blittable types. Therefore, they require some handling.

There are some notes about marshaling booleans in the managed environment. The first thing to mention about is that Windows defines two types of Boolean variables:
  1. BOOL: Defined as INT, therefore, it is 4-bytes wide.
  2. BOOLEAN: Defined as BYTE, therefore it is only 1-byte.
Both can be set to non-zero to indicate a true (TRUE) value, and zero otherwise (FALSE.)

Again, the two types exist only in the Windows SDK. Other environments may define other types with similar names.

While it is true that BOOL and BOOLEAN are best marshaled as System.Boolean, BOOL can be marshaled as System.Int32 too, because it is defined as a 32-bit integer. On the other hand, BOOLEAN can be marshaled as System.Byte or System.U1, because it is defined as 8-bits integer. Do you remember the rule of thumb?

Take into consideration that whether you are marshaling your Boolean type to System.Boolean, System.Int32, or System.Byte, it is recommended that you apply MarshalAsAttribute attribute to the variable to specify the underlying unmanaged type. For example, to specify that the underlying type is BOOL, specify UnmanagedType.Bool (recommended) or UnmanagedType.I4 in the MarshalAsAttribute constructor. On the other hand, BOOLEAN can be specified as UnmanagedType.U1. If you omit MarshalAsAttribute, CLR assumes the default behavior for System.Boolean, which is 2 bytes wide. For more information about MarshalAsAttribute attribute, see the last chapter: "Controlling the Marshaling Process."

Try It Out!

Fortunately, plenty of functions return BOOL indicating whether the function succeeded (TRUE) or failed (FALSE.)

The following is the definition of the famous CloseHandle() function:

Listing 2.3 CloseHandle() Unmanaged Signature

BOOL CloseHandle(HANDLE hObject);

The managed version of CloseHandle() is as following:

Listing 2.4 CloseHandle() Managed Signature

[DllImport("Kernel32.dll")]
[return: MarshalAs(UnmanagedType.Bool)]
// In addition, you can marshal it as:
// [return: MarshalAs(UnmanagedType.I4)]
// Moreover, You can change System.Boolean to System.Int32
static extern Boolean CloseHandle(IntPtr hObject)

Handles covered soon. For now, it is OK to know that all handles marshaled to System.IntPtr.

Marshaling Textual Data Types

How to Marshal Strings and Buffers

This section discusses how to marshal strings and buffers. We will use the terms string and buffer interchangeably to refer to a sequence of characters.

Two types exist in the managed environment for marshaling unmanaged string buffers. They are System.String and System.Text.StringBuilder. Of course, they both hold character sequences. However, StringBuilder is more advantageous because it is very efficient working with mutable strings than System.String.

Every time you use one of the methods of System.String class or you pass a System.String to a function, normally, you create a new string object in memory, which requires a new allocation of memory space for the new object. In addition, if the function changes the string you will not get the results back. That is why System.String is called immutable. On the other hand, StringBuilder does not require re-allocating of space unless you exceed its capacity. Besides the talk about marshaling, you should use StringBuilder to accommodate performance issues if you often change the same string many times.

To keep System.String immutable, the marshaler copies the contents of the string to another buffer before calling the function, and then it passes that buffer to the function. If you were passing the string by reference, the marshaler copies the contents of the buffer into the original string when returning from the function.

Conversely, when using StringBuilder, it passes a reference to the internal buffer of StringBuilder if passed by value. Passing a StringBuilder by reference actually passes a pointer to the StringBuilder object into memory to the function not a pointer to the buffer itself.

Read more about passing a type by value or by reference in the section "Passing Mechanism" later in this chapter.

Another feature of StringBuilder is its ability to specify buffer capacity. As we will see, this can be very helpful in plenty of cases.

To summarize, System.String is preferable when working with immutable strings, especially for input (In) arguments. On the other hand, System.Text.StringBuilder is preferable with changeable strings especially output (Out) arguments.

Noteworthy to say that StringBuilder cannot be used inside compound types. Therefore, you will need to use String instead.

Another point to mention is that you can pass array of System.Char in place of a System.String or System.Text.StringBuilder. In other words, you can marshal unmanaged strings as managed arrays of System.Char (or System.Int16, do you remember?)

Compound types discussed in the next chapter.

Handling Character Encoding

Encoding of a character is very important because it determines the value that the character can hold and the size it occupies into memory. For example, if the character is ANSI-encoded it can be one of only 256 characters. Likewise, if it is Unicode-encoded, it can hold one of 65536 characters, which is very good for most languages.

If you need more information about Unicode, you can check the official site of Unicode, www.Unicode.org. In addition, Programming Windows 5th by Charles Petzold includes a must-read introduction of Unicode and character sets.

For controlling character encoding when marshaling unmanaged types, you may take one of two approaches or you can combine them as needed. You can control the encoding of the overall function (i.e. at the function level,) or you can drill down and control the encoding process at a granular level by controlling every argument separately (the second approach is required in certain cases e.g. MultiByteToWideChar() function.)

For changing the encoding of the overall function, DllImportAttribute offers the property CharSet that indicates the encoding (character set) for the strings and arguments of the function. This property can take one of several values:
  • CharSet.Auto (CLR Default): Strings encoding varies based on operating system; it is Unicode-encoded on Windows NT and ANSI-encoded on other versions of Windows.
  • CharSet.Ansi (C# Default): Strings are always 8-bit ANSI-encoded.
  • CharSet.Unicode: Strings are always 16-bit Unicode-encoded.
  • CharSet.None: Obsolete. Has the same behavior as CharSet.Ansi.

    Take into consideration that if you have not set the CharSet property, CLR automatically sets it to CharSet.Auto. However, some languages override the default behavior. For example, C# defaults to CharSet.Ansi
It is worth mentioning that plenty of functions that accept strings and buffers are just names (technically typedefs)! They are not real functions, they are entry-points (aliases) for the real functions. For example, ReadConsole() function is nothing except an entry point redirects the call to the right function, either ReadConsoleA() if ANSI is defined, or ReadConsoleW() if Unicode is defined (A stands for ANSI, and W stands for Wide which means Unicode.) Therefore, you can actually bypass this entry-point by changing the PInvoke method name to match the right function or by changing DllImportAttribute.EntryPoint to the name of the required function. In both cases, setting DllImportAttribute.CharSet along with is no use.

If you want to control the encoding at a granular level, you can apply the MarshalAsAttribute attribute to the argument specifying the underlying unmanaged type.

Usually, you will need to unify the character encoding of all your native functions and types. This is, all the functions should be either Unicode or ANSI. Under rare occasions, some functions would be different in character encoding.

It is worth mentioning that, for fixed-length strings you will need to set the SizeConst property of MarshalAsAttribute to the buffer length.

These techniques are not limited to arguments only! You can use them with variables of compound types too. We will look at compound types in the following chapter.

Try It Out!

Now we will look on both ReadConsole() and FormatConsole() unmanaged functions and how to call them from your managed environment. Next is the definition of both functions and other functions required for the example:

Listing 2.5 GetStdHandle(), ReadConsole(), GetLastError(), and FormatMessage() Unmanaged Signature

HANDLE GetStdHandle(
DWORD nStdHandle);

BOOL ReadConsole(
HANDLE hConsoleInput,
[out] LPVOID lpBuffer,
DWORD nNumberOfCharsToRead,
[out] LPDWORD lpNumberOfCharsRead,
LPVOID lpReserved);

DWORD GetLastError(void);

DWORD FormatMessage(
DWORD dwFlags,
LPCVOID lpSource,
DWORD dwMessageId,
DWORD dwLanguageId,
[out] LPTSTR lpBuffer,
DWORD nSize,
va_list* Arguments);

And this is the managed version along with the test code.

Listing 2.6 Reading from the Console Screen Buffer Example

    // For retrieving a handle to a specific console device
    [DllImport("Kernel32.dll")]
    static extern IntPtr GetStdHandle(
    [param: MarshalAs(UnmanagedType.U4)]
int nStdHandle);

    // Used with GetStdHandle() for retrieving console input buffer
    const int STD_INPUT_HANDLE = -10;

    // Specifying the DLL along with the character set
    [DllImport("Kernel32.dll", CharSet = CharSet.Unicode)]
    [return: MarshalAs(UnmanagedType.Bool)]
    static extern bool ReadConsole(
        // Handle to the input device
    IntPtr hConsoleInput,
        // The buffer of which to write input to
    [param: MarshalAs(UnmanagedType.LPTStr), Out()]
// [param: MarshalAs(UnmanagedType.AsAny)]
StringBuilder lpBuffer,
        // Number of characters to read
    [param: MarshalAs(UnmanagedType.U4)]
uint nNumberOfCharsToRead,
        // Outputs the number of characters read
    [param: MarshalAs(UnmanagedType.U4), Out()]
out uint lpNumberOfCharsRead,
        // Reserved = Always set to NULL
    [param: MarshalAs(UnmanagedType.AsAny)]
uint lpReserved);

    // For getting the code for the last error occurred
    [DllImport("Kernel32.dll")]
    [return: MarshalAs(UnmanagedType.U4)]
    static extern uint GetLastError();

    // Retrieves error messages
    [DllImport("Kernel32.dll", CharSet = CharSet.Unicode)]
    [return: MarshalAs(UnmanagedType.U4)]
    static extern uint FormatMessage(
        // Options
    [param: MarshalAs(UnmanagedType.U4)]
uint dwFlags,
        // Source to get the message from
        // [param: MarshalAs(UnmanagedType.AsAny)]
    [param: MarshalAs(UnmanagedType.U4)]
uint lpSource,
        // Message code = error code
    [param: MarshalAs(UnmanagedType.U4)]
uint dwMessageId,
        // Language ID (Reserved)
    [param: MarshalAs(UnmanagedType.U4)]
uint dwLanguageId,
        // Outputs the error message
    [param: MarshalAs(UnmanagedType.LPTStr), Out()]
out string lpBuffer,
        // Size of error message
    [param: MarshalAs(UnmanagedType.U4)]
uint nSize,
        // Additional options
    [param: MarshalAs(UnmanagedType.U4)]
uint Arguments);

    // Message Options
    const uint FORMAT_MESSAGE_ALLOCATE_BUFFER = 0x0100;
    const uint FORMAT_MESSAGE_IGNORE_INSERTS = 0x0200;
    const uint FORMAT_MESSAGE_FROM_SYSTEM = 0x1000;
    const uint FORMAT_MESSAGE_FLAGS =
    FORMAT_MESSAGE_ALLOCATE_BUFFER |
    FORMAT_MESSAGE_IGNORE_INSERTS |
    FORMAT_MESSAGE_FROM_SYSTEM;

    // Message Source
    public const int FORMAT_MESSAGE_FROM_HMODULE = 0x0800;

    static void Main()
    {
        // Handle to input buffer
        IntPtr handle = GetStdHandle(STD_INPUT_HANDLE);

        const int maxCount = 256;

        uint noCharacters;
        StringBuilder builder = new StringBuilder(maxCount);

        if (ReadConsole(handle, builder, (uint)maxCount,
        out noCharacters, 0) == false) // false = non-zero = failed
        {
            string errMsg;
            FormatMessage(FORMAT_MESSAGE_FLAGS,
            FORMAT_MESSAGE_FROM_HMODULE,
            GetLastError(),
            0, // Means NULL
            out errMsg,
            0, // Maximum length
            0); // Means NULL
 
            Console.WriteLine("ERROR:\n{0}", errMsg);
        }
        else // true = zero = succeeded
            // Writing user input withour the newline
            Console.WriteLine("User wroted: = " +
            builder.ToString().Substring(0,
            builder.Length - Environment.NewLine.Length));
 
        Console.WriteLine(new string('-', 25));

        builder = new StringBuilder(maxCount);

        // Invalid handle
        handle = GetStdHandle(12345);

        if (ReadConsole(handle, builder, (uint)maxCount,
        out noCharacters, 0) == false) // false = non-zero = failed
        {
            string errMsg;
            FormatMessage(FORMAT_MESSAGE_FLAGS,
            FORMAT_MESSAGE_FROM_HMODULE,
            GetLastError(),
            0, // Means NULL
            out errMsg,
            0, // Maximum length
            0); // Means NULL

            Console.WriteLine("ERROR: {0}", errMsg);
        }
        else // true = zero = succeeded
            // Exculding the newline characters
            Console.WriteLine("User wroted: = " +
            builder.ToString().Substring(0,
            builder.Length - Environment.NewLine.Length));
    }

The last code demonstrates other useful techniques:

  • Until now, handles should be marshaled as System.IntPtr. The following section talks in details about handles.
  • Because LPVOID and LPCVOID are both defined as a pointer to a variant (i.e. any type,) you can set them to any type you want. They are very similar to System.Object in the .NET methodology or Variant for people who are familiar with the traditional Visual Basic. In our example, we have marshaled LPVOID as System.UInt32 and set it to zero. Again, you are free to play with the marshaling types. LPVOID and LPCVOID are both 32-bit integer. Why not just marshaling them as any of the 32-bit managed types and forgetting about them? In addition, you can marshal it as System.IntPtr, and pass it System.IntPtr.Zero to indicate a NULL value. Moreover, you can marshal it as System.Object, and set it to any value, even null to indicate the NULL value. Variant has been discussed in details previously in the section "Marshaling Blittable Data Types."
  • va_list* is a pointer to an array of specific arguments. You can marshal it as an array, or System.IntPtr. System.IntPtr is preferred if you intend to pass it a NULL value.
  • If the function requires a parameter passed by value or by reference you can add the required modifiers like ref and out to the parameter, and decorate the parameter with either InAttribute or OutAttribute, or both. The section "Passing an Argument by Value or by Reference" later in this chapter discusses by-value and by-reference parameters.
  • While DWORD is defined as unsigned 32-bit integer and it should be marshaled as System.UInt32, we find that the GetStdHandle() can take one of three values: -10 for the input device, -11 for the output device, and -12 for the error device (usually is the output device.) Although System.UInt32 does not support negative values, Windows handles this for you. It converts the signed value to its equivalent unsigned value. Therefore, you should not worry about the value passed. However, keep in mind that the unsigned values are too different (from the perspective of most developers.) For example, the unsigned value of -11 is 0xFFFFFFF5! Does this seem strange for you? Start by consulting the documentation about binary notation.
Marshaling Handles

Generic Handles

What is a handle? A handle is a pointer to some resource loaded in memory, such as handles to the console standard input, output, and error devices, the handle for the window, and the handle to a device context (DC.)

There are plenty of type handles in unmanaged code, here is some of them:
  • HANDLE: This is the most widely used handle type in the unmanaged environment. It represents a generic handle.
  • HWND: Most widely used with Windows application. It is a handle to a window or a control.
  • HDC, HGDIOBJ, HBITMAP, HICON, HBRUSH, HPEN, and HFONT: If you have worked with GDI, you will be familiar with these handles. HDC is a handle to a device context (DC) object that will be used for drawing. HGDIOBJ is a handle for any GDI object. HBITMAP is a handle to a bitmap, while HICON is a handle to an icon. HBRUSH is a handle to a brush, HPEN is a handle to pen, and HFONT is a handle to a font.
  • HFILE: A handle to a file opened by any of Windows File IO functions like OpenFile() function.
  • HMENU: A handle to a menu or menu item.

    Again, from all you have seen, you may have noticed that most types identified by a prefix or a suffix. For example, handles prefixed with the letter H, while some pointers have the suffix _PTR, or the prefix P or LP. While strings with letter W are Unicode-encoded, and strings with letter T are OS-based.
Handles can be marshaled as the managed type System.IntPtr that represents a pointer to an object into memory. It is worth mentioning that because System.IntPtr represents a pointer to an object no matter what the object is, you can use System.IntPtr for marshaling any type not handles only, but that is not recommended because it is more difficult to work with, and it is not very flexible, but it provides more control over the object in memory. For more information about memory management, see chapter 6: "Memory Management."

In addition, starting from version 2.0, new managed types for working with unmanaged handles added to the .NET Framework. A new namespace Microsoft.Win32.SafeHandles that contains most of the new types has been added too. Other types exist in System.Runtime.InteropServices. These types called managed handles.

Managed handles allow you to pass, to unmanaged code, a handle to an unmanaged resource (such as DC) wrapped by managed class.

There are two kinds of managed handles safe and critical handles.

Safe handles

Safe handles represented by the abstract System.Runtime.InteropServices.SafeHandle. Safe handles provide protection from recycling security attacks by perform reference counting (and that makes safe handles slower.) In addition, it provides critical finalization for handle resources. As a refresher, finalization means releasing the object and its resources from the memory, and critical finalization ensures object finalization under any circumstances.

Figure 2.2 shows the definition of SafeHandle and its descendants.

marsheling2.gif


As the diagram illustrates, SafeHandle is the base class that represents any safe handle. It inherits from System.Runtime.ConstrainedExecution.CriticalFinalizerObject that ensures the finalization process. The following are the most common members of SafeHandle:
  • IsClosed: Returns a value indicates whether the handle is closed.
  • IsInvalid: Abstract. If overridden, returns a value indicates whether the handle is invalid or not.
  • Close() and Dispose(): Both close the handle and dispose its resources. Internally, they rely on the abstract method ReleaseHandle() for releasing the handle. Therefore, classes inherit from SafeHandle must implement this member. Be aware that Dispose() is inherited from System.IDispose interface that is implemented by SafeHandle, and Close() does not do anything except calling the Dispose() method. Therefore, you strictly should dispose (close) the handle as soon as you finish your work with it.
  • ReleaseHandle(): Protected Abstract. Use to provide handle clean-up code. This function should returns true if successfully released, or false otherwise. In the case of false, it generates a ReleaseHandleFailed Managed Debugging Assistant (MDA) exception that will not interrupt your code but provides you with a bad sign about it. Keep in mind that ReleaseHandle() called internally by Dispose().
  • SetHandle(): Protected. Sets the handle to the specified pre-existing handle.
  • SetHandleAsInvalid(): Sets the handle as invalid so it is no longer used.
  • DangerousGetHandle(): Returns System.IntPtr that represents the handle. Beware that if you have called SetHandleAsInvalid() before calling DangerousGetHandle(), it returns the original handle not the invalid one.
  • DangerousRelease(): Manually releasing the handle in unsafe manner. It is recommended using Close() or Dispose() methods instead.
  • DangerousAddRef(): Increments the reference count of the handle. It is not recommended using neither DangerousRelease() nor DangerousAddRef(), use safe methods instead. However, when working with COM, you will find yourself using these functions.

    Do not use unsafe methods unless you really need to use it because they pass the protection level offered by safe handles.

Because SafeHandle is abstract, you must either implement it or use one of its implementation classes. Only two classes from the new namespace Microsoft.Win32.SafeHandles implement SafeHandle, both are abstract too:
  • SafeHandleMinusOneIsInvalid: Represents a safe handle of which a value of -1 indicates that the handle is invalid. Therefore, IsInvalid returns true only if the handle equals to -1.
  • SafeHandleZeroOrMinusOneIsInvalid: Represents a safe handle of which a value of 0 or -1 indicates that the handle is invalid. So, IsInvalid returns true only if the handle equals to 0 or -1.

Notice that, choosing between the two implementations is up to the type of the underlying handle. If it considered invalid if set to -1, use SafeHandleMinusOneIsInvalid. If it considered invalid if set to 0 or -1, use SafeHandleZeroOrMinusOneIsInvalid. Using the right class for the handle ensures that methods like IsInvalid() returns correct results. It also ensures that CLR will mark the handle as garbage only if it is invalid.

If you need to provide a safe handle for your object, you will need to inherit from SafeHandleMinusOneIsInvalid, SafeHandleZeroOrMinusOneIsInvalid, or even from SafeHandle. Be aware that, you will always need to override the ReleaseHandle() method because neither SafeHandleMinusOneIsInvalid nor SafeHandleZeroOrMinusOneIsInvalid does override it.

As the diagram illustrates, two concrete classes inherit from SafeHandleZeroOrMinusOneIsInvalid:
  • SafeFileHandle: A wrapper class for an IO device handle (e.g. HFILE.) This class internally overrides the ReleaseHandle() and calls the unmanaged CloseHandle() function to close the handle. Use when working with HFILE handles in Windows File IO functions like OpenFile() and CreateFile(). Internally, System.FileStream uses a HFILE as SafeFileHandle, and it exposes a constructor that accepts SafeFileHandle.
  • SafeWaitHandle: If you are working with unmanaged thread synchronization objects like a Mutex or an Event, then this should be the desired marshaling type for synchronization objects' handles.
Now, we are going to create a file using CreateFile() function with SafeFileHandle for the marshaling process. The definition of CreateFile() is as following:

Listing 2.7 CreateFile() Unmanaged Signature

HANDLE CreateFile(
LPCTSTR lpFileName,
DWORD dwDesiredAccess,
DWORD dwShareMode,
LPSECURITY_ATTRIBUTES lpSecurityAttributes,
DWORD dwCreationDisposition,
DWORD dwFlagsAndAttributes,
HANDLE hTemplateFile
);

In addition, here is the .NET code:

Listing 2.8 Create File Example

    [DllImport("Kernel32.dll", CharSet = CharSet.Auto, SetLastError = true)]
    static extern SafeFileHandle CreateFile(
    string lpFileName,
    uint dwDesiredAccess,
    uint dwShareMode,
        // Because we are going to set the argument
        // to NULL we marshaled it as IntPtr
        // so we can set it to IntPtr.Zero
        // to represent a NULL value
    IntPtr lpSecurityAttributes,
    uint dwCreationDisposition,
    uint dwFlagsAndAttributes,
        // A handle for a template file
        // we are going to set it to NULL
        // so e can marshal it as System.IntPtr
        // and pass IntPtr.Zero for the NULL value
        // But, this is another way
    SafeFileHandle hTemplateFile);

    // Accessing the file for writing
    const uint GENERIC_WRITE = 0x40000000;
    // Do now allow file sharing
    const uint FILE_SHARE_NONE = 0x0;
    // Create the file and overwrites it if exists
    const uint CREATE_ALWAYS = 0x2;
    // Normal file, no attribute set
    const uint FILE_ATTRIBUTE_NORMAL = 0x80;

    static void Main()
    {
        SafeFileHandle handle =
        CreateFile("C:\\MyFile.txt",
        GENERIC_WRITE,
        FILE_SHARE_NONE,
        IntPtr.Zero, // NULL
        CREATE_ALWAYS,
        FILE_ATTRIBUTE_NORMAL,
        new SafeFileHandle(IntPtr.Zero, true));

        // Because SafeFileHandle inherits
        // SafeHandleZeroOrMinusOneIsInvalid
        // IsInvalid returns true only if
        // the handle equals to 0 or -1
        if (handle.IsInvalid) // 0 or -1
        {
            Console.WriteLine("ERROR: {0}", Marshal.GetLastWin32Error());
            return;
            // Marshal.GetLastWin32Error() returns the last error only
            // if DllImportAttribute.SetLastError is set to true
        }

        FileStream stream = new FileStream(handle, FileAccess.Write);
        StreamWriter writer = new StreamWriter(stream);
        writer.WriteLine("Hello, World!");
        writer.Close();

        /*
        * Order of methods called by
        * StreamWriter by this example:
        *
        * StreamWriter.Close()
        * - StreamWriter.BaseStream.Close()
        * - - FileStream.SafeFileHandle.Close()
        * - - - SafeHandleZeroOrMinusOneIsInvalid
        * .Close()
        * - - - - SafeHandle.Close()
        * - - - - - SafeHandle.ReleaseHandle()
        */
    }

Although, you can use IntPtr instead of SafeFileHandle, the FileStream constructor that accepts the IntPtr is considered obsolete (.NET 2.0 and higher) and you should use the constructor that accepts the SafeFileHandle.

The next example demonstrates how to create your custom safe handle. This custom safe handle represents a handle invalid only if equals to zero. Although, you can extend the functionality of either SafeHandleMinusOneIsInvalid or SafeHandleZeroOrMinusOneIsInvalid, we have inherited SafeHandle directly. Code is very simple:

Listing 2.9 Custom Safe Handle Example

    public sealed class SafeHandleZeroIsInvalid : SafeHandle
    {
        [DllImport("Kernel32.dll")]
        [return: MarshalAs(UnmanagedType.Bool)]
        private static extern bool CloseHandle(IntPtr hObject);

        // If ownsHandle equals true handle will
        // be automatically released during the
        // finalization process, otherwise, you
        // will have the responsibility to
        // release it outside the class.
        // Automatic releasing means calling
        // the ReleaseHandle() method.
        public SafeHandleZeroIsInvalid
        (IntPtr preexistingHandle, bool ownsHandle)
            : base(IntPtr.Zero, ownsHandle)
        {
            this.SetHandle(preexistingHandle);
        }

        public override bool IsInvalid
        {
            get
            {
                // this.handle.ToInt32() == 0
                // this.handle == new IntPtr(0)
                return this.handle == IntPtr.Zero;
            }
        }

        protected override bool ReleaseHandle()
        {
            return CloseHandle(this.handle);
        }
    }

Until now, I do not have an answer for why a handle could be invalid only if it is set to zero! Maybe you will need this for your custom handles. However, this is just an illustration.

Critical Handles

Critical handles are the same as safe handles, except that they do not perform reference counting, so they do not provide protection from recycling security attacks.

Use critical handles instead of safe handles to address performance considerations, but you will be required to provide necessary synchronization for reference counting yourself.

Critical handles represented by the abstract System.Runtime.InteropServices.CriticalHandle. Figure 2.3 shows the definition of CriticalHandle and its descendants.

marsheling3.gif

As the diagram illustrates, CriticalHandle is the base class that represents any critical handle. It inherits from System.Runtime.ConstrainedExecution.CriticalFinalizerObject that ensures the finalization process. The members of CriticalHandle are the same as SafeHandle, except that it does not include the Dangerous-prefixed methods because critical handles themselves are dangerous because they do not provide the necessary protection. For more information about CriticalHandle members, refer to members of SafeHandle discussed previously.

Because CriticalHandle is abstract, you must either implement it or use one of its implementation classes. Only two classes from the new namespace Microsoft.Win32.SafeHandles implement CriticalHandle, both are abstract too:

  • CriticalHandleMinusOneIsInvalid: Represents a critical handle of which a value of -1 indicates that the handle is invalid. Therefore, IsInvalid returns true only if the handle equals to -1.
  • CriticalHandleZeroOrMinusOneIsInvalid: Represents a critical handle of which a value of 0 or -1 indicates that the handle is invalid. So, IsInvalid returns true only if the handle equals to 0 or -1.
Examples are the same as SafeHandle, only to change the type name.
Passing Mechanism

When passing an argument to a function, the function may require either passing the argument by value or by reference. If the function intends to change argument value, it requires it to be passed by reference, otherwise, by value. This is what called passing mechanism.

Value arguments (i.e. input/In arguments,) when passed to a function, a copy of the argument is sent to the function. Therefore, any changes to the argument do not affect the original copy. On the other hand, reference arguments, when passed to a function, the argument itself is passed to the function. Therefore, the caller sees any changes happen inside the function.

Arguments passed by reference can be either In/Out (Input/Output) or only Out (Output.) In/Out arguments are used for passing input to the function and returning output. On the other hand, Out arguments used for returning output only. Therefore, In/Out arguments must be initialized before they are passed to the function. Conversely, Out arguments do not require pre-initialization.

When passing an argument by value, no changes to the PInvoke method are required. Conversely, passing an argument by reference requires two additional changes. The first is adding the ref modifier to the argument if it is In/Out argument, or the out modifier if it is Out argument. The second is decorating your argument with both InAttribute and OutAttribute attributes if it is In/Out argument or only OutAttribute if it is Out argument. To be honest, applying those attributes is not required, the modifiers are adequate in most cases. However, applying them gives the CLR a notation about the passing mechanism.

As you have seen, when marshaling a string, you can marshal it as a System.String or as a System.Text.StringBuilder. By default, StringBuilder is passed by reference (you do not need to apply any changes.) System.String on the other hand is passed by value.

It is worth mentioning that Windows API does not support reference arguments. Instead, if a function requires an argument to be passed by reference, it declares it as a pointer so that caller can see the applied changes. Other code such as COM libraries can require either a pointer or a reference argument. In either cases, you can safely apply the changes required. You can also marshal a pointer argument as System.IntPtr or as the unsafe void* for example.

Many of the previous examples demonstrated only functions those require arguments to be passed by value. Some functions require one or more arguments to be passed by reference. A good example of a function requires In/Out argument is GetVersionEx() which returns version information of the current system. It requires a single reference (In/Out) argument. The argument is of the structure OSVERSIONINFOEX. For our discussion, we will leave this function to the next chapter in the discussion of compound types.

A great deal of functions require Out arguments specially for returning results or status information. Good examples are ReadConsole() and WriteConsole() that require by-reference Out arguments for returning the characters read/written. The following is the unmanaged signature for the WriteConsole() function.

Listing 2.10 WriteConsole() Unmanaged Signature

BOOL WriteConsole(
HANDLE hConsoleOutput,
VOID lpBuffer,
DWORD nNumberOfCharsToWrite,
LPDWORD lpNumberOfCharsWritten,
LPVOID lpReserved
);

And this is the managed version along with the test code:

Listing 2.11
Writing to Console Screen Example

    [DllImport("Kernel32.dll", CharSet = CharSet.Unicode)]
    [return: MarshalAs(UnmanagedType.Bool)]
    static extern bool WriteConsole(
    IntPtr hConsoleOutput,
    String lpBuffer,
    [param: MarshalAs(UnmanagedType.U4)]
UInt32 nNumberOfCharsToWrite,
    [param: MarshalAs(UnmanagedType.U4)]
out UInt32 lpNumberOfCharsWritten,
    [param: MarshalAs(UnmanagedType.AsAny)]
object lpReserved);

    [DllImport("Kernel32.dll")]
    static extern IntPtr GetStdHandle(
    [param: MarshalAs(UnmanagedType.U4)]
Int32 nStdHandle);

    const int STD_OUTPUT_HANDLE = -11;

    static void Main()
    {
        IntPtr handle = GetStdHandle(STD_OUTPUT_HANDLE);

        String textToWrite = "Hello, World!" + Environment.NewLine;
        uint noCharactersWritten;

        WriteConsole(handle,
        textToWrite,
        (uint)textToWrite.Length,
        out noCharactersWritten,
        null);

        Console.WriteLine("No. Characters written = {0}",
        noCharactersWritten);
    }

Finally yet importantly, chapter 6 provides you with more granular and down-level details about the memory management and the passing mechanism.

Additional Techniques

Here we will talk about techniques that should be taken into consideration when working with unmanaged code, they are encapsulation, creating wrappers, working with nullable arguments, and working out CLS problem.

Encapsulation

If the function requires an argument that can be set to a value or more, you can define these values (constants or typedefs) in an enumeration so you can easily access every set of values separately; that technique called encapsulation (grouping.) The following example shows the MessageBoxEx() example, the most suitable function for the example:

Listing 2.12 Message Box Example

    [DllImport("User32.dll", CharSet = CharSet.Unicode)]
    [return: MarshalAs(UnmanagedType.I4)]
    static extern UInt32 MessageBoxEx
    (IntPtr hWnd,
    [param: MarshalAs(UnmanagedType.LPTStr)]
String lpText,
    [param: MarshalAs(UnmanagedType.LPTStr)]
String lpCaption,
    [param: MarshalAs(UnmanagedType.U4)]
UInt32 uType,
    [param: MarshalAs(UnmanagedType.U2)]
UInt16 wLanguageId);

    public enum MB_BUTTON : uint
    {
        MB_OK = 0x0,
        MB_OKCANCEL = 0x1,
        MB_ABORTRETRYIGNORE = 0x2,
        MB_YESNOCANCEL = 0x3,
        MB_YESNO = 0x4,
        MB_RETRYCANCEL = 0x5,
        MB_HELP = 0x4000,
    }
    public enum MB_ICON : uint
    {
        MB_ICONHAND = 0x10,
        MB_ICONQUESTION = 0x20,
        MB_ICONEXCLAMATION = 0x30,
        MB_ICONASTERISK = 0x40,
        MB_ICONERROR = MB_ICONHAND,
        MB_ICONSTOP = MB_ICONHAND,
        MB_ICONWARNING = MB_ICONEXCLAMATION,
        MB_ICONINFORMATION = MB_ICONASTERISK,
    }
    public enum MB_DEF_BUTTON : uint
    {
        MB_DEFBUTTON1 = 0x0,
        MB_DEFBUTTON2 = 0x100,
        MB_DEFBUTTON3 = 0x200,
        MB_DEFBUTTON4 = 0x300,
    }
    public enum MB_MODAL : uint
    {
        MB_APPLMODAL = 0x0,
        MB_SYSTEMMODAL = 0x1000,
        MB_TASKMODAL = 0x2000,
    }
    public enum MB_SPECIAL : uint
    {
        MB_SETFOREGROUND = 0x10000,
        MB_DEFAULT_DESKTOP_ONLY = 0x20000,
        MB_SERVICE_NOTIFICATION_NT3X = 0x40000,
        MB_TOPMOST = 0x40000,
        MB_RIGHT = 0x80000,
        MB_RTLREADING = 0x100000,
        MB_SERVICE_NOTIFICATION = 0x200000,
    }
    public enum MB_RETURN : uint
    {
        IDOK = 1,
        IDCANCEL = 2,
        IDABORT = 3,
        IDRETRY = 4,
        IDIGNORE = 5,
        IDYES = 6,
        IDNO = 7,
        IDCLOSE = 8,
        IDHELP = 9,
        IDTRYAGAIN = 10,
        IDCONTINUE = 11,
    }

    static void Main()
    {
        UInt32 result = MessageBoxEx(IntPtr.Zero, // NULL
        "Do you want to save changes before closing?",
        "MyApplication",
        (UInt32)MB_BUTTON.MB_YESNOCANCEL |
        (UInt32)MB_ICON.MB_ICONQUESTION |
        (UInt32)MB_DEF_BUTTON.MB_DEFBUTTON3 |
        (UInt32)MB_SPECIAL.MB_TOPMOST,
        0);// Reserved
 
        if (result == 0) // error occurred
            Console.WriteLine("ERROR");
        else
        {
            MB_RETURN ret = (MB_RETURN)result;

            if (ret == MB_RETURN.IDYES)
                Console.WriteLine("User clicked Yes!");
            else if (ret == MB_RETURN.IDNO)
                Console.WriteLine("User clicked No!");
            else if (ret == MB_RETURN.IDCANCEL)
                Console.WriteLine("User clicked Cancel!");
        }
    }

You could also change the names of the constants to friendly names.

Figure 2.4 shows the message box resulted from running of the last code.

marsheling4.jpg

In addition, you can marshal an argument as an enumeration which of the argument type of course. The following example demonstrates this:

Listing 2.13 Console Standard Devices Example

    [DllImport("Kernel32.dll")]
    static extern IntPtr GetStdHandle(
    [param: MarshalAs(UnmanagedType.U4)]
CONSOLE_STD_HANDLE nStdHandle);

    public enum CONSOLE_STD_HANDLE
    {
        STD_INPUT_HANDLE = -10,
        STD_OUTPUT_HANDLE = -11,
        STD_ERROR_HANDLE = -12
    }

    static void Main()
    {
        IntPtr handle;
        handle =
        GetStdHandle(CONSOLE_STD_HANDLE.STD_INPUT_HANDLE);
        if (handle == IntPtr.Zero)
            Console.WriteLine("Failed!");
        else
            Console.WriteLine("Succeeded!");
    }

Creating Wrappers

Exposing PInvoke methods to the outside the assembly is not a good practice. It is always recommended that you group your PInvoke methods into an internal class, and that class should be named as NativeMethods, SafeNativeMethods or UnsafeNativeMethods. For more information about this, check Code Analyzing Rules in MSDN documentation. Read "Move PInvokes to Native Methods Class" article.

The following code segment illustrates the wrapper method for our MessageBoxEx() function:

Listing 2.14 Message Box Example Revised

    public static MB_RETURN MessageBox
    (IntPtr handle, string text, string title,
    MB_BUTTON buttons, MB_ICON icon, MB_DEF_BUTTON defaultButton,
    MB_MODAL modality, MB_SPECIAL options)
    {
        UInt32 result = MessageBoxEx(handle,
        "Do you want to save changes before closing?",
        "MyApplication",
        (UInt32)buttons |
        (UInt32)icon |
        (UInt32)defaultButton |
        (UInt32)modality |
        (UInt32)options,
        0);

        if (result == 0)
            // Not recommended throwing System.Exception
            // throw a derived exception instead
            throw new Exception("FAILED");
        return (MB_RETURN)result;
    }

In addition, it is recommended changing the type of enumerations to any CLS-compliant type like System.Int32. Check the last technique in this section.

Working with Nullable Arguments

Some function arguments are nullable. Means that they can take a NULL (null in C#) value. To pass a NULL value to an argument, you can marshal this argument as System.IntPtr, so you can set it to System.IntPtr.Zero to represent a NULL value. Another trick here is creating an overload for the function, in which the first is marshaled as the argument type, and the other is marshaled as System.IntPtr. Thus, if you pass a System.IntPtr.Zero, CLR directs the call to the function with System.IntPtr. Conversely, passing a value to the argument, directs the call to the function with the correct type. The following code segment demonstrates this technique:

Code abbreviated for clarity.

Listing 2.15 ScrollConsoleScreenBuffer() Managed Signature

    [DllImport("Kernel32.dll", CharSet = CharSet.Auto)]
    [return: MarshalAs(UnmanagedType.Bool)]
    static extern bool ScrollConsoleScreenBuffer(
    IntPtr hConsoleOutput,
    SMALL_RECT lpScrollRectangle,
    SMALL_RECT lpClipRectangle,
    COORD dwDestinationOrigin,
    CHAR_INFO lpFill);

    [DllImport("Kernel32.dll", CharSet = CharSet.Auto)]
    [return: MarshalAs(UnmanagedType.Bool)]
    static extern bool ScrollConsoleScreenBuffer(
    IntPtr hConsoleOutput,
    SMALL_RECT lpScrollRectangle,
    IntPtr lpClipRectangle,
    COORD dwDestinationOrigin,
    CHAR_INFO lpFill);

Working Out the CLS Problem

You should know that some types are non-CLS-compliant and you should avoid exposing them outside the assembly. For example, the famous System.UInt32 is non-CLS-compliant, and you strictly should not expose it.

Being non-CLS-compliant means that the type violates with CLS (Common Language Specifications) specifications. Following CLS specifications helps the interoperation of .NET languages. It helps avoiding some actions like declaring specific types or following uncommon naming conventions.

Why to avoid such these acts? This helps the big goal of .NET Framework, the interoperation of .NET languages. Some languages for example does not support variable names beginning with an underscore (_) others do. Therefore, following the CLS specifications allows your assembly to be callable from any other assembly build with any language easily.

To force the check of CLS specification, you can decorate the assembly with System.CLSCompliantAttribute attribute -specifying true,- and that would result in compiler warnings whenever you try to expose non-CLS-compliant type out.

To work out this CLS dilemma, for functions require UInt32 as an argument, you can create a wrapper that behaves as an entry-point to the private non-CLS-compliant method. That wrapper method accepts, for instance, System.Int32 and converts it internally to System.UInt32.

For structures, you can declare the structure as internal and continue using it the normal way.

Again, you could replace all non-CLS-compliant types like System.UInt32 with CLS-compliant equivalents like System.Int32 and take advantage of easily distributing your types and assembly. However, that would not be easy in all cases.

It is very helpful consulting the documentation about System.CLSCompliantAttribute attribute.

Real-World Examples

In this chapter, we have covered many aspects of marshaling in many examples. However, most of all were just for illustration.

The following are some real-world examples that solve problems that you might face while developing your application. Those problems can be solved only via interoperability with unmanaged code.

Programmatically Swapping Mouse Buttons

The following code swaps mouse buttons programmatically. It makes the left button acts like the right button (e.g. opens the context menu) and vice versa.

Listing 2.16 Swapping Mouse Buttons Sample

    [DllImport("user32.dll")]
    [return: MarshalAs(UnmanagedType.Bool)]
    public static extern bool SwapMouseButton
    ([param: MarshalAs(UnmanagedType.Bool)] bool fSwap);

    public void MakeRightButtonPrimary()
    {
        SwapMouseButton(true);
    }

    public void MakeLeftButtonPrimary()
    {
        SwapMouseButton(false);
    }

Programmatically Turning On the Screen Saver

The following code shows how to turn on the screen saver programmatically.

Listing 2.19 Dragging a Form without a Title Bar Sample

    [DllImport("User32.dll")]
    public static extern int SendMessage
    (IntPtr hWnd,
    uint Msg,
    uint wParam,
    uint lParam);

    public const uint WM_SYSCOMMAND = 0x112;
    public const uint SC_SCREENSAVE = 0xF140;

    public enum SpecialHandles
    {
        HWND_DESKTOP = 0x0,
        HWND_BROADCAST = 0xFFFF
    }

    public static void TurnOnScreenSaver()
    {
        SendMessage(
        new IntPtr((int)SpecialHandles.HWND_BROADCAST),
        WM_SYSCOMMAND,
        SC_SCREENSAVE,
        0);
    }


Dragging a Form without a Title Bar

The following code allows the form to be dragged from its body. This code is a good example for the wrapper creating technique discussed earlier.

Listing 2.18 Dragging a Form without a Title Bar Sample

SafeNativeMethods.cs

    internal static class SafeNativeMethods
    {
        [DllImport("user32.dll")]
        [return: MarshalAs(UnmanagedType.I4)]
        public static extern int SendMessage(
        IntPtr hWnd,
        [param: MarshalAs(UnmanagedType.U4)]
uint Msg,
        [param: MarshalAs(UnmanagedType.U4)]
uint wParam,
        [param: MarshalAs(UnmanagedType.I4)]
int lParam);

        [DllImport("user32.dll")]
        [return: MarshalAs(UnmanagedType.Bool)]
        public static extern bool ReleaseCapture();

        public const uint WM_NCLBUTTONDOWN = 0xA1; // 161
        public const uint HTCAPTION = 2;
    }


HelperMethods.cs

    internal static class HelperMethods
    {
        public static void MoveObject(IntPtr hWnd)
        {
            SafeNativeMethods.ReleaseCapture();
            SafeNativeMethods.SendMessage
            (hWnd, SafeNativeMethods.WM_NCLBUTTONDOWN,
            SafeNativeMethods.HTCAPTION, 0);
        }
    }


MainForm.cs

    // In the form, write the following code
    // in the handler of the MouseDown event

    private void MainForm_MouseDown(object sender, MouseEventArgs e)
    {
        HelperMethods.MoveObject(this.Handle);
    }

Summary

The last word to say is that MarshalAsAttribute is not required all the time. Sometimes it is optional, and other times it is required.

For example, if you marshal blittable data types like DWORD, you can safely ignore MarshalAsAttribute. Conversely, if you are marshaling non-blittable data types like booleans and strings, you will need to use the MarshalAsAttribute to ensure correct marshaling process. However, it is always better giving the CLR and other developers a notation about the underlying data type by apply the MarshalAsAttribute attribute to blittable data types too.

Finally yet importantly, this chapter was the key for the gate to the interoperation with unmanaged environments. It discussed the most important part of the marshaling process, marshaling the simple types, which you will always need to keep it into your mind.

Next, you will learn how to work with compound types and marshal them in your managed environment.


Similar Articles