共用方式為


MFC Debugging Techniques

If you are debugging an MFC program, these debugging techniques may be useful.

In this topic

AfxDebugBreak

The TRACE macro

Memory leak detection in MFC

  • Tracking memory allocations

  • Enabling memory diagnostics

  • Taking memory snapshots

  • Viewing memory statistics

  • Taking object dumps

    • Interpreting memory dumps

    • Customizing object dumps

    Reducing the size of an MFC Debug build

    • Building an MFC app with debug information for selected modules

AfxDebugBreak

MFC provides a special AfxDebugBreak function for hard-coding breakpoints in source code:

AfxDebugBreak( );

On Intel platforms, AfxDebugBreak produces the following code, which breaks in source code rather than kernel code:

_asm int 3

On other platforms, AfxDebugBreak merely calls DebugBreak.

Be sure to remove AfxDebugBreak statements when you create a release build or use #ifdef _DEBUG to surround them.

In this topic

The TRACE macro

To display messages from your program in the debugger Output window, you can use the ATLTRACE macro or the MFC TRACE macro. Like assertions, the trace macros are active only in the Debug version of your program and disappear when compiled in the Release version.

The following examples show some of the ways you can use the TRACE macro. Like printf, the TRACE macro can handle a number of arguments.

int x = 1;
int y = 16;
float z = 32.0;
TRACE( "This is a TRACE statement\n" );

TRACE( "The value of x is %d\n", x );

TRACE( "x = %d and y = %d\n", x, y );

TRACE( "x = %d and y = %x and z = %f\n", x, y, z );

The TRACE macro appropriately handles both char* and wchar_t* parameters. The following examples demonstrate the use of the TRACE macro together with different types of string parameters.

TRACE( "This is a test of the TRACE macro that uses an ANSI string: %s %d\n", "The number is:", 2);

TRACE( L"This is a test of the TRACE macro that uses a UNICODE string: %s %d\n", L"The number is:", 2);

TRACE( _T("This is a test of the TRACE macro that uses a TCHAR string: %s %d\n"), _T("The number is:"), 2);

For more information on the TRACE macro, see Diagnostic Services.

In this topic

Detecting memory leaks in MFC

MFC provides classes and functions for detecting memory that is allocated but never deallocated.

Tracking memory allocations

In MFC, you can use the macro DEBUG_NEW in place of the new operator to help locate memory leaks. In the Debug version of your program, DEBUG_NEW keeps track of the file name and line number for each object that it allocates. When you compile a Release version of your program, DEBUG_NEW resolves to a simple new operation without the file name and line number information. Thus, you pay no speed penalty in the Release version of your program.

If you do not want to rewrite your entire program to use DEBUG_NEW in place of new, you can define this macro in your source files:

#define new DEBUG_NEW

When you do an object dump, each object allocated with DEBUG_NEW will show the file and line number where it was allocated, allowing you to pinpoint the sources of memory leaks.

The Debug version of the MFC framework uses DEBUG_NEW automatically, but your code does not. If you want the benefits of DEBUG_NEW, you must use DEBUG_NEW explicitly or #define new as shown above.

In this topic

Enabling memory diagnostics

Before you can use the memory diagnostics facilities, you must enable diagnostic tracing.

To enable or disable memory diagnostics

  • Call the global function AfxEnableMemoryTracking to enable or disable the diagnostic memory allocator. Because memory diagnostics are on by default in the debug library, you will typically use this function to temporarily turn them off, which increases program execution speed and reduces diagnostic output.

To select specific memory diagnostic features with afxMemDF

  • If you want more precise control over the memory diagnostic features, you can selectively turn individual memory diagnostic features on and off by setting the value of the MFC global variable afxMemDF. This variable can have the following values as specified by the enumerated type afxMemDF.

    Value

    Description

    allocMemDF

    Turn on diagnostic memory allocator (default).

    delayFreeMemDF

    Delay freeing memory when calling delete or free until program exits. This will cause your program to allocate the maximum possible amount of memory.

    checkAlwaysMemDF

    Call AfxCheckMemory every time memory is allocated or freed.

    These values can be used in combination by performing a logical-OR operation, as shown here:

    afxMemDF = allocMemDF | delayFreeMemDF | checkAlwaysMemDF;
    

In this topic

Taking memory snapshots

  1. Create a CMemoryState object and call the CMemoryState::Checkpoint member function. This creates the first memory snapshot.

  2. After your program performs its memory allocation and deallocation operations, create another CMemoryState object and call Checkpoint for that object. This gets a second snapshot of memory usage.

  3. Create a third CMemoryState object and call its CMemoryState::Difference member function, supplying as arguments the two previous CMemoryState objects. If there is a difference between the two memory states, the Difference function returns a nonzero value. This indicates that some memory blocks have not been deallocated.

    This example shows what the code looks like:

    // Declare the variables needed
    #ifdef _DEBUG
        CMemoryState oldMemState, newMemState, diffMemState;
        oldMemState.Checkpoint();
    #endif
    
        // Do your memory allocations and deallocations.
        CString s("This is a frame variable");
        // The next object is a heap object.
       CPerson* p = new CPerson( "Smith", "Alan", "581-0215" );
    
    #ifdef _DEBUG
        newMemState.Checkpoint();
        if( diffMemState.Difference( oldMemState, newMemState ) )
        {
            TRACE( "Memory leaked!\n" );
        }
    #endif
    

    Notice that the memory-checking statements are bracketed by #ifdef_DEBUG/ #endif blocks so that they are compiled only in Debug versions of your program.

    Now that you know a memory leak exists, you can use another member function, CMemoryState::DumpStatistics that will help you locate it.

In this topic

Viewing memory statistics

The CMemoryState::Difference function looks at two memory-state objects and detects any objects not deallocated from the heap between the beginning and end states. After you have taken memory snapshots and compared them using CMemoryState::Difference, you can call CMemoryState::DumpStatistics to get information about the objects that have not been deallocated.

Consider the following example:

if( diffMemState.Difference( oldMemState, newMemState ) )
{
   TRACE( "Memory leaked!\n" );
   diffMemState.DumpStatistics();
}

A sample dump from the example looks like this:

0 bytes in 0 Free Blocks
22 bytes in 1 Object Blocks
45 bytes in 4 Non-Object Blocks
Largest number used: 67 bytes
Total allocations: 67 bytes

Free blocks are blocks whose deallocation is delayed if afxMemDF was set to delayFreeMemDF.

Ordinary object blocks, shown on the second line, remain allocated on the heap.

Non-object blocks include arrays and structures allocated with new. In this case, four non-object blocks were allocated on the heap but not deallocated.

Largest number used gives the maximum memory used by the program at any time.

Total allocations gives the total amount of memory used by the program.

In this topic

Taking object dumps

In an MFC program, you can use CMemoryState::DumpAllObjectsSince to dump a description of all objects on the heap that have not been deallocated. DumpAllObjectsSince dumps all objects allocated since the last CMemoryState::Checkpoint. If no Checkpoint call has taken place, DumpAllObjectsSince dumps all objects and nonobjects currently in memory.

Note

Before you can use MFC object dumping, you must enable diagnostic tracing.

Note

MFC automatically dumps all leaked objects when your program exits, so you do not need to create code to dump objects at that point.

The following code tests for a memory leak by comparing two memory states and dumps all objects if a leak is detected.

if( diffMemState.Difference( oldMemState, newMemState ) )
{
   TRACE( "Memory leaked!\n" );
   diffMemState.DumpAllObjectsSince();
}

The contents of the dump look like this:

Dumping objects ->

{5} strcore.cpp(80) : non-object block at $00A7521A, 9 bytes long
{4} strcore.cpp(80) : non-object block at $00A751F8, 5 bytes long
{3} strcore.cpp(80) : non-object block at $00A751D6, 6 bytes long
{2} a CPerson at $51A4

Last Name: Smith
First Name: Alan
Phone #: 581-0215

{1} strcore.cpp(80) : non-object block at $00A7516E, 25 bytes long

The numbers in braces at the beginning of most lines specify the order in which the objects were allocated. The most recently allocated object has the highest number and appears at the top of the dump.

To get the maximum amount of information out of an object dump, you can override the Dump member function of any CObject-derived object to customize the object dump.

You can set a breakpoint on a particular memory allocation by setting the global variable _afxBreakAlloc to the number shown in the braces. If you rerun the program the debugger will break execution when that allocation takes place. You can then look at the call stack to see how your program got to that point.

The C run-time library has a similar function, _CrtSetBreakAlloc, that you can use for C run-time allocations.

In this topic

Interpreting memory dumps

Look at this object dump in more detail:

{5} strcore.cpp(80) : non-object block at $00A7521A, 9 bytes long
{4} strcore.cpp(80) : non-object block at $00A751F8, 5 bytes long
{3} strcore.cpp(80) : non-object block at $00A751D6, 6 bytes long
{2} a CPerson at $51A4

Last Name: Smith
First Name: Alan
Phone #: 581-0215

{1} strcore.cpp(80) : non-object block at $00A7516E, 25 bytes long

The program that generated this dump had only two explicit allocations—one on the stack and one on the heap:

// Do your memory allocations and deallocations.
CString s("This is a frame variable");
// The next object is a heap object.
CPerson* p = new CPerson( "Smith", "Alan", "581-0215" );

The CPerson constructor takes three arguments that are pointers to char, which are used to initialize CString member variables. In the memory dump, you can see the CPerson object along with three nonobject blocks (3, 4, and 5). These hold the characters for the CString member variables and will not be deleted when the CPerson object destructor is invoked.

Block number 2 is the CPerson object itself. $51A4 represents the address of the block and is followed by the contents of the object, which were output by CPerson::Dump when called by DumpAllObjectsSince.

You can guess that block number 1 is associated with the CString frame variable because of its sequence number and size, which matches the number of characters in the frame CString variable. Variables allocated on the frame are automatically deallocated when the frame goes out of scope.

Frame Variables

In general, you should not worry about heap objects associated with frame variables because they are automatically deallocated when the frame variables go out of scope. To avoid clutter in your memory diagnostic dumps, you should position your calls to Checkpoint so that they are outside the scope of frame variables. For example, place scope brackets around the previous allocation code, as shown here:

oldMemState.Checkpoint();
{
    // Do your memory allocations and deallocations ...
    CString s("This is a frame variable");
    // The next object is a heap object.
    CPerson* p = new CPerson( "Smith", "Alan", "581-0215" );
}
newMemState.Checkpoint();

With the scope brackets in place, the memory dump for this example is as follows:

Dumping objects ->

{5} strcore.cpp(80) : non-object block at $00A7521A, 9 bytes long
{4} strcore.cpp(80) : non-object block at $00A751F8, 5 bytes long
{3} strcore.cpp(80) : non-object block at $00A751D6, 6 bytes long
{2} a CPerson at $51A4

Last Name: Smith
First Name: Alan
Phone #: 581-0215

Nonobject Allocations

Notice that some allocations are objects (such as CPerson) and some are nonobject allocations. "Nonobject allocations" are allocations for objects not derived from CObject or allocations of primitive C types such as char, int, or long. If the CObject-derived class allocates additional space, such as for internal buffers, those objects will show both object and nonobject allocations.

Preventing Memory Leaks

Notice in the code above that the memory block associated with the CString frame variable has been deallocated automatically and does not show up as a memory leak. The automatic deallocation associated with scoping rules takes care of most memory leaks associated with frame variables.

For objects allocated on the heap, however, you must explicitly delete the object to prevent a memory leak. To clean up the last memory leak in the previous example, delete the CPerson object allocated on the heap, as follows:

{
    // Do your memory allocations and deallocations.
    CString s("This is a frame variable");
    // The next object is a heap object.
    CPerson* p = new CPerson( "Smith", "Alan", "581-0215" );
    delete p;
}

In this topic

Customizing object dumps

When you derive a class from CObject, you can override the Dump member function to provide additional information when you use DumpAllObjectsSince to dump objects to the Output window.

The Dump function writes a textual representation of the object's member variables to a dump context (CDumpContext). The dump context is similar to an I/O stream. You can use the append operator (<<) to send data to a CDumpContext.

When you override the Dump function, you should first call the base class version of Dump to dump the contents of the base class object. Then output a textual description and value for each member variable of your derived class.

The declaration of the Dump function looks like this:

class CPerson : public CObject
{
public:
#ifdef _DEBUG
    virtual void Dump( CDumpContext& dc ) const;
#endif

    CString m_firstName;
    CString m_lastName;
    // And so on...
};

Because object dumping only makes sense when you are debugging your program, the declaration of the Dump function is bracketed with an #ifdef _DEBUG / #endif block.

In the following example, the Dump function first calls the Dump function for its base class. It then writes a short description of each member variable along with the member's value to the diagnostic stream.

#ifdef _DEBUG
void CPerson::Dump( CDumpContext& dc ) const
{
    // Call the base class function first.
    CObject::Dump( dc );

    // Now do the stuff for our specific class.
    dc << "last name: " << m_lastName << "\n"
        << "first name: " << m_firstName << "\n";
}
#endif

You must supply a CDumpContext argument to specify where the dump output will go. The Debug version of MFC supplies a predefined CDumpContext object named afxDump that sends output to the debugger.

CPerson* pMyPerson = new CPerson;
// Set some fields of the CPerson object.
//...
// Now dump the contents.
#ifdef _DEBUG
pMyPerson->Dump( afxDump );
#endif

In this topic

Reducing the size of an MFC Debug build

The debug information for a large MFC application can take up a lot of disk space. You can use one of these procedures to reduce the size:

  1. Rebuild the MFC libraries using the /Z7, /Zi, /ZI (Debug Information Format) option, instead of /Z7. These options build a single program database (PDB) file that contains debug information for the entire library, reducing redundancy and saving space.

  2. Rebuild the MFC libraries without debug information (no /Z7, /Zi, /ZI (Debug Information Format) option). In this case, the lack of debug information will prevent you from using most debugger facilities within the MFC library code, but because the MFC libraries are already thoroughly debugged, this may not be a problem.

  3. Build your own application with debug information for selected modules only as described below.

In this topic

Building an MFC app with debug information for selected modules

Building selected modules with the MFC debug libraries enables you to use stepping and the other debug facilities in those modules. This procedure makes use of both the Debug and Release modes of the Visual C++ makefile, thus necessitating the changes described in the following steps (and also making a "rebuild all" necessary when a full Release build is required).

  1. In Solution Explorer, select the project.

  2. From the View menu, select Property Pages.

  3. First, you will create a new project configuration.

    1. In the <Project> Property Pages dialog box, click the Configuration Manager button.

    2. In the Configuration Manager dialog box, locate your project in the grid. In the Configuration column, select <New...>.

    3. In the New Project Configuration dialog box, type a name for your new configuration, such as "Partial Debug", in the Project Configuration Name box.

    4. In the Copy Settings from list, choose Release.

    5. Click OK to close the New Project Configurationdialog box.

    6. Close the Configuration Manager dialog box.

  4. Now, you will set options for the entire project.

    1. In the Property Pages dialog box, under the Configuration Properties folder, select the General category.

    2. In the project settings grid, expand Project Defaults (if necessary).

    3. Under Project Defaults, find Use of MFC. The current setting appears in the right column of the grid. Click on the current setting and change it to Use MFC in a Static Library.

    4. In the left pane of the Properties Pages dialog box, open the C/C++ folder and select Preprocessor. In the properties grid, find Preprocessor Definitions and replace "NDEBUG" with "_DEBUG".

    5. In the left pane of the Properties Pages dialog box, open the Linker folder and select the Input Category. In the properties grid, find Additional Dependencies. In the Additional Dependencies setting, type "NAFXCWD.LIB" and "LIBCMT."

    6. Click OK to save the new build options and close the Property Pages dialog box.

  5. From the Build menu, select Rebuild. This removes all debug information from your modules but does not affect the MFC library.

  6. Now you must add debug information back to selected modules in your application. Remember that you can set breakpoints and perform other debugger functions only in modules you have compiled with debug information. For each project file in which you want to include debug information, carry out the following steps:

    1. In Solution Explorer, open the Source Files folder located under your project.

    2. Select the file you want to set debug information for.

    3. From the View menu, select Property Pages.

    4. In the Property Pages dialog box, under the Configuration Settings folder, open the C/C++ folder then select the General category.

    5. In the properties grid, find Debug Information Format.

    6. Click the Debug Information Format settings and select the desired option (usually /ZI) for debug information.

    7. If you are using an application wizard-generated application or have precompiled headers, you have to turn off the precompiled headers or recompile them before compiling the other modules. Otherwise, you will receive warning C4650 and error message C2855. You can turn off precompiled headers by changing the Create/Use Precompiled Headers setting in the <Project> Properties dialog box (Configuration Properties folder, C/C++ subfolder, Precompiled Headers category).

  7. From the Build menu, select Build to rebuild project files that are out of date.

As an alternative to the technique described in this topic, you can use an external makefile to define individual options for each file. In that case, to link with the MFC debug libraries, you must define the _DEBUG flag for each module. If you want to use MFC release libraries, you must define NDEBUG. For more information on writing external makefiles, see the NMAKE Reference.

In this topic

See Also

Other Resources

Debugging Visual C++