Muokkaa

Jaa

TN038: MFC/OLE IUnknown Implementation

  • Artikkeli

Note

The following technical note has not been updated since it was first included in the online documentation. As a result, some procedures and topics might be out of date or incorrect. For the latest information, it is recommended that you search for the topic of interest in the online documentation index.

At the heart of OLE 2 is the "OLE Component Object Model", or COM. COM defines a standard for how cooperating objects communicate to one another. This includes the details of what an "object" looks like, including how methods are dispatched on an object. COM also defines a base class, from which all COM compatible classes are derived. This base class is IUnknown. Although the IUnknown interface is referred to as a C++ class, COM is not specific to any one language — it can be implemented in C, PASCAL, or any other language that can support the binary layout of a COM object.

OLE refers to all classes derived from IUnknown as "interfaces." This is an important distinction, since an "interface" such as IUnknown carries with it no implementation. It simply defines the protocol by which objects communicate, not the specifics of what those implementations do. This is reasonable for a system that allows for maximum flexibility. It is MFC's job to implement a default behavior for MFC/C++ programs.

To understand MFC's implementation of IUnknown you must first understand what this interface is. A simplified version of IUnknown is defined below:

class IUnknown
{
public:
    virtual HRESULT QueryInterface(REFIID iid, void** ppvObj) = 0;
    virtual ULONG AddRef() = 0;
    virtual ULONG Release() = 0;
};

Note

Certain necessary calling convention details, such as __stdcall are left out for this illustration.

The AddRef and Release member functions control memory management of the object. COM uses a reference counting scheme to keep track of objects. An object is never referenced directly as you would in C++. Instead, COM objects are always referenced through a pointer. To release the object when the owner is done using it, the object's Release member is called (as opposed to using operator delete, as would be done for a traditional C++ object). The reference counting mechanism allows for multiple references to a single object to be managed. An implementation of AddRef and Release maintains a reference count on the object — the object is not deleted until its reference count reaches zero.

AddRef and Release are fairly straightforward from an implementation standpoint. Here is a trivial implementation:

ULONG CMyObj::AddRef()
{
    return ++m_dwRef;
}

ULONG CMyObj::Release()
{
    if (--m_dwRef == 0)
    {
        delete this;
        return 0;
    }
    return m_dwRef;
}

The QueryInterface member function is a little more interesting. It is not very interesting to have an object whose only member functions are AddRef and Release — it would be nice to tell the object to do more things than IUnknown provides. This is where QueryInterface is useful. It allows you to obtain a different "interface" on the same object. These interfaces are usually derived from IUnknown and add additional functionality by adding new member functions. COM interfaces never have member variables declared in the interface, and all member functions are declared as pure-virtual. For example,

class IPrintInterface : public IUnknown
{
public:
    virtual void PrintObject() = 0;
};

To get an IPrintInterface if you only have an IUnknown, call QueryInterface using the IID of the IPrintInterface. An IID is a 128-bit number that uniquely identifies the interface. There is an IID for each interface that either you or OLE define. If pUnk is a pointer to an IUnknown object, the code to retrieve an IPrintInterface from it might be:

IPrintInterface* pPrint = NULL;
if (pUnk->QueryInterface(IID_IPrintInterface, (void**)&pPrint) == NOERROR)
{
    pPrint->PrintObject();
    pPrint->Release();
    // release pointer obtained via QueryInterface
}

That seems fairly easy, but how would you implement an object supporting both the IPrintInterface and IUnknown interface In this case it is simple since the IPrintInterface is derived directly from IUnknown — by implementing IPrintInterface, IUnknown is automatically supported. For example:

class CPrintObj : public CPrintInterface
{
    virtual HRESULT QueryInterface(REFIID iid, void** ppvObj);
    virtual ULONG AddRef();
    virtual ULONG Release();
    virtual void PrintObject();
};

The implementations of AddRef and Release would be exactly the same as those implemented above. CPrintObj::QueryInterface would look something like this:

HRESULT CPrintObj::QueryInterface(REFIID iid, void FAR* FAR* ppvObj)
{
    if (iid == IID_IUnknown || iid == IID_IPrintInterface)
    {
        *ppvObj = this;
        AddRef();
        return NOERROR;
    }
    return E_NOINTERFACE;
}

As you can see, if the interface identifier (IID) is recognized, a pointer is returned to your object; otherwise an error occurs. Also note that a successful QueryInterface results in an implied AddRef. Of course, you'd also have to implement CEditObj::Print. That is simple because the IPrintInterface was directly derived from the IUnknown interface. However, if you wanted to support two different interfaces, both derived from IUnknown, consider the following:

class IEditInterface : public IUnkown
{
public:
    virtual void EditObject() = 0;
};

Although there are a number of different ways to implement a class supporting both IEditInterface and IPrintInterface, including using C++ multiple inheritance, this note will concentrate on the use of nested classes to implement this functionality.

class CEditPrintObj
{
public:
    CEditPrintObj();

    HRESULT QueryInterface(REFIID iid, void**);
    ULONG AddRef();
    ULONG Release();
    DWORD m_dwRef;

    class CPrintObj : public IPrintInterface
    {
    public:
        CEditPrintObj* m_pParent;
        virtual HRESULT QueryInterface(REFIID iid, void** ppvObj);
        virtual ULONG AddRef();
        virtual ULONG Release();
    } m_printObj;

    class CEditObj : public IEditInterface
    {
    public:
        CEditPrintObj* m_pParent;
        virtual ULONG QueryInterface(REFIID iid, void** ppvObj);
        virtual ULONG AddRef();
        virtual ULONG Release();
    } m_editObj;
};

The entire implementation is included below:

CEditPrintObj::CEditPrintObj()
{
    m_editObj.m_pParent = this;
    m_printObj.m_pParent = this;
}

ULONG CEditPrintObj::AddRef()
{
    return ++m_dwRef;
}

CEditPrintObj::Release()
{
    if (--m_dwRef == 0)
    {
        delete this;
        return 0;
    }
    return m_dwRef;
}

HRESULT CEditPrintObj::QueryInterface(REFIID iid, void** ppvObj)
{
    if (iid == IID_IUnknown || iid == IID_IPrintInterface)
    {
        *ppvObj = &m_printObj;
        AddRef();
        return NOERROR;
    }
    else if (iid == IID_IEditInterface)
    {
        *ppvObj = &m_editObj;
        AddRef();
        return NOERROR;
    }
    return E_NOINTERFACE;
}

ULONG CEditPrintObj::CEditObj::AddRef()
{
    return m_pParent->AddRef();
}

ULONG CEditPrintObj::CEditObj::Release()
{
    return m_pParent->Release();
}

HRESULT CEditPrintObj::CEditObj::QueryInterface(REFIID iid, void** ppvObj)
{
    return m_pParent->QueryInterface(iid, ppvObj);
}

ULONG CEditPrintObj::CPrintObj::AddRef()
{
    return m_pParent->AddRef();
}

ULONG CEditPrintObj::CPrintObj::Release()
{
    return m_pParent->Release();
}

HRESULT CEditPrintObj::CPrintObj::QueryInterface(REFIID iid, void** ppvObj)
{
    return m_pParent->QueryInterface(iid, ppvObj);
}

Notice that most of the IUnknown implementation is placed into the CEditPrintObj class rather than duplicating the code in CEditPrintObj::CEditObj and CEditPrintObj::CPrintObj. This reduces the amount of code and avoids bugs. The key point here is that from the IUnknown interface it is possible to call QueryInterface to retrieve any interface the object might support, and from each of those interfaces it is possible to do the same. This means that all QueryInterface functions available from each interface must behave exactly the same way. In order for these embedded objects to call the implementation in the "outer object", a back-pointer is used (m_pParent). The m_pParent pointer is initialized during the CEditPrintObj constructor. Then you would implement CEditPrintObj::CPrintObj::PrintObject and CEditPrintObj::CEditObj::EditObject as well. Quite a bit of code was added to add one feature — the ability to edit the object. Fortunately, it is quite uncommon for interfaces to have only a single member function (although it does happen) and in this case, EditObject and PrintObject would usually be combined into a single interface.

That's a lot of explanation and a lot of code for such a simple scenario. The MFC/OLE classes provide a simpler alternative. The MFC implementation uses a technique similar to the way Windows messages are wrapped with Message Maps. This facility is called Interface Maps and is discussed in the next section.

MFC Interface Maps

MFC/OLE includes an implementation of "Interface Maps" similar to MFC's "Message Maps" and "Dispatch Maps" in concept and execution. The core features of MFC's Interface Maps are as follows:

  • A standard implementation of IUnknown, built into the CCmdTarget class.

  • Maintenance of the reference count, modified by AddRef and Release

  • Data driven implementation of QueryInterface

In addition, interface maps support the following advanced features:

  • Support for creating aggregatable COM objects

  • Support for using aggregate objects in the implementation of a COM object

  • The implementation is hookable and extensible

For more information on aggregation, see the Aggregation topic.

MFC's interface map support is rooted in the CCmdTarget class. CCmdTarget "has-a" reference count as well as all the member functions associated with the IUnknown implementation (the reference count for example is in CCmdTarget). To create a class that supports OLE COM, you derive a class from CCmdTarget and use various macros as well as member functions of CCmdTarget to implement the desired interfaces. MFC's implementation uses nested classes to define each interface implementation much like the example above. This is made easier with a standard implementation of IUnknown as well as a number of macros that eliminate some of the repetitive code.

Interface Map Basics

To implement a class using MFC's interface maps

  1. Derive a class either directly or indirectly from CCmdTarget.

  2. Use the DECLARE_INTERFACE_MAP function in the derived class definition.

  3. For each interface you wish to support, use the BEGIN_INTERFACE_PART and END_INTERFACE_PART macros in the class definition.

  4. In the implementation file, use the BEGIN_INTERFACE_MAP and END_INTERFACE_MAP macros to define the class's interface map.

  5. For each IID supported, use the INTERFACE_PART macro between the BEGIN_INTERFACE_MAP and END_INTERFACE_MAP macros to map that IID to a specific "part" of your class.

  6. Implement each of the nested classes that represent the interfaces you support.

  7. Use the METHOD_PROLOGUE macro to access the parent, CCmdTarget-derived object.

  8. AddRef, Release, and QueryInterface can delegate to the CCmdTarget implementation of these functions (ExternalAddRef, ExternalRelease, and ExternalQueryInterface).

The CPrintEditObj example above could be implemented as follows:

class CPrintEditObj : public CCmdTarget
{
public:
    // member data and member functions for CPrintEditObj go here

// Interface Maps
protected:
    DECLARE_INTERFACE_MAP()

    BEGIN_INTERFACE_PART(EditObj, IEditInterface)
        STDMETHOD_(void, EditObject)();
    END_INTERFACE_PART(EditObj)

    BEGIN_INTERFACE_PART(PrintObj, IPrintInterface)
        STDMETHOD_(void, PrintObject)();
    END_INTERFACE_PART(PrintObj)
};

The above declaration creates a class derived from CCmdTarget. The DECLARE_INTERFACE_MAP macro tells the framework that this class will have a custom interface map. In addition, the BEGIN_INTERFACE_PART and END_INTERFACE_PART macros define nested classes, in this case with names CEditObj and CPrintObj (the X is used only to differentiate the nested classes from global classes which start with "C" and interface classes which start with "I"). Two nested members of these classes are created: m_CEditObj, and m_CPrintObj, respectively. The macros automatically declare the AddRef, Release, and QueryInterface functions; therefore you only declare the functions specific to this interface: EditObject and PrintObject (the OLE macro STDMETHOD is used so that _stdcall and virtual keywords are provided as appropriate for the target platform).

To implement the interface map for this class:

BEGIN_INTERFACE_MAP(CPrintEditObj, CCmdTarget)
    INTERFACE_PART(CPrintEditObj, IID_IPrintInterface, PrintObj)
    INTERFACE_PART(CPrintEditObj, IID_IEditInterface, EditObj)
END_INTERFACE_MAP()

This connects the IID_IPrintInterface IID with m_CPrintObj and IID_IEditInterface with m_CEditObj respectively. The CCmdTarget implementation of QueryInterface (CCmdTarget::ExternalQueryInterface) uses this map to return pointers to m_CPrintObj and m_CEditObj when requested. It is not necessary to include an entry for IID_IUnknown; the framework will use the first interface in the map (in this case, m_CPrintObj) when IID_IUnknown is requested.

Even though the BEGIN_INTERFACE_PART macro automatically declared the AddRef, Release and QueryInterface functions for you, you still need to implement them:

ULONG FAR EXPORT CEditPrintObj::XEditObj::AddRef()
{
    METHOD_PROLOGUE(CEditPrintObj, EditObj)
    return pThis->ExternalAddRef();
}

ULONG FAR EXPORT CEditPrintObj::XEditObj::Release()
{
    METHOD_PROLOGUE(CEditPrintObj, EditObj)
    return pThis->ExternalRelease();
}

HRESULT FAR EXPORT CEditPrintObj::XEditObj::QueryInterface(
    REFIID iid,
    void FAR* FAR* ppvObj)
{
    METHOD_PROLOGUE(CEditPrintObj, EditObj)
    return (HRESULT)pThis->ExternalQueryInterface(&iid, ppvObj);
}

void FAR EXPORT CEditPrintObj::XEditObj::EditObject()
{
    METHOD_PROLOGUE(CEditPrintObj, EditObj)
    // code to "Edit" the object, whatever that means...
}

The implementation for CEditPrintObj::CPrintObj, would be similar to the above definitions for CEditPrintObj::CEditObj. Although it would be possible to create a macro that could be used to automatically generate these functions (but earlier in MFC/OLE development this was the case), it becomes difficult to set break points when a macro generates more than one line of code. For this reason, this code is expanded manually.

By using the framework implementation of message maps, there are a number of things that were not necessary to do:

  • Implement QueryInterface

  • Implement AddRef and Release

  • Declare either of these built-in methods on both of your interfaces

In addition, the framework uses message maps internally. This allows you to derive from a framework class, say COleServerDoc, that already supports certain interfaces and provides either replacements or additions to the interfaces provided by the framework. You can do this because the framework fully supports inheriting an interface map from a base class. That is the reason why BEGIN_INTERFACE_MAP takes as its second parameter the name of the base class.

Note

It is generally not possible to reuse the implementation of MFC's built-in implementations of the OLE interfaces just by inheriting the embedded specialization of that interface from the MFC version. This is not possible because the use of the METHOD_PROLOGUE macro to get access to the containing CCmdTarget-derived object implies a fixed offset of the embedded object from the CCmdTarget-derived object. This means, for example, you cannot derive an embedded XMyAdviseSink from MFC's implementation in COleClientItem::XAdviseSink, because XAdviseSink relies on being at a specific offset from the top of the COleClientItem object.

Note

You can, however, delegate to the MFC implementation for all of the functions that you want MFC's default behavior. This is done in the MFC implementation of IOleInPlaceFrame (XOleInPlaceFrame) in the COleFrameHook class (it delegates to m_xOleInPlaceUIWindow for many functions). This design was chosen to reduce the runtime size of objects which implement many interfaces; it eliminates the need for a back-pointer (such as the way m_pParent was used in the previous section).

Aggregation and Interface Maps

In addition to supporting stand-alone COM objects, MFC also supports aggregation. Aggregation itself is too complex a topic to discuss here; refer to the Aggregation topic for more information on aggregation. This note will simply describe the support for aggregation built into the framework and interface maps.

There are two ways to use aggregation: (1) using a COM object that supports aggregation, and (2) implementing an object that can be aggregated by another. These capabilities can be referred to as "using an aggregate object" and "making an object aggregatable". MFC supports both.

Using an Aggregate Object

To use an aggregate object, there needs to be some way to tie the aggregate into the QueryInterface mechanism. In other words, the aggregate object must behave as though it is a native part of your object. So how does this tie into MFC's interface map mechanism In addition to the INTERFACE_PART macro, where a nested object is mapped to an IID, you can also declare an aggregate object as part of your CCmdTarget derived class. To do so, the INTERFACE_AGGREGATE macro is used. This allows you to specify a member variable (which must be a pointer to an IUnknown or derived class), which is to be integrated into the interface map mechanism. If the pointer is not NULL when CCmdTarget::ExternalQueryInterface is called, the framework will automatically call the aggregate object's QueryInterface member function, if the IID requested is not one of the native IIDs supported by the CCmdTarget object itself.

To use the INTERFACE_AGGREGATE macro

  1. Declare a member variable (an IUnknown*) which will contain a pointer to the aggregate object.

  2. Include an INTERFACE_AGGREGATE macro in your interface map, which refers to the member variable by name.

  3. At some point (usually during CCmdTarget::OnCreateAggregates), initialize the member variable to something other than NULL.

For example:

class CAggrExample : public CCmdTarget
{
public:
    CAggrExample();

protected:
    LPUNKNOWN m_lpAggrInner;
    virtual BOOL OnCreateAggregates();

    DECLARE_INTERFACE_MAP()
    // "native" interface part macros may be used here
};

CAggrExample::CAggrExample()
{
    m_lpAggrInner = NULL;
}

BOOL CAggrExample::OnCreateAggregates()
{
    // wire up aggregate with correct controlling unknown
    m_lpAggrInner = CoCreateInstance(CLSID_Example,
        GetControllingUnknown(), CLSCTX_INPROC_SERVER,
        IID_IUnknown, (LPVOID*)&m_lpAggrInner);

    if (m_lpAggrInner == NULL)
        return FALSE;
    // optionally, create other aggregate objects here
    return TRUE;
}

BEGIN_INTERFACE_MAP(CAggrExample, CCmdTarget)
    // native "INTERFACE_PART" entries go here
    INTERFACE_AGGREGATE(CAggrExample, m_lpAggrInner)
END_INTERFACE_MAP()

The m_lpAggrInner variable is initialized in the constructor to NULL. The framework ignores a NULL member variable in the default implementation of QueryInterface. OnCreateAggregates is a good place to actually create your aggregate objects. You'll have to call it explicitly if you are creating the object outside of the MFC implementation of COleObjectFactory. The reason for creating aggregates in CCmdTarget::OnCreateAggregates as well as the usage of CCmdTarget::GetControllingUnknown will become apparent when creating aggregatable objects is discussed.

This technique will give your object all of the interfaces that the aggregate object supports plus its native interfaces. If you only want a subset of the interfaces that the aggregate supports, you can override CCmdTarget::GetInterfaceHook. This allows you very low-level hookability, similar to QueryInterface. Usually, you want all the interfaces that the aggregate supports.

Making an Object Implementation Aggregatable

For an object to be aggregatable, the implementation of AddRef, Release, and QueryInterface must delegate to a "controlling unknown." In other words, for it to be part of the object, it must delegate AddRef, Release, and QueryInterface to a different object, also derived from IUnknown. This "controlling unknown" is provided to the object when it is created, that is, it is provided to the implementation of COleObjectFactory. Implementing this carries a small amount of overhead, and in some cases is not desirable, so MFC makes this optional. To enable an object to be aggregatable, you call CCmdTarget::EnableAggregation from the object's constructor.

If the object also uses aggregates, you must also be sure to pass the correct "controlling unknown" to the aggregate objects. Usually this IUnknown pointer is passed to the object when the aggregate is created. For example, the pUnkOuter parameter is the "controlling unknown" for objects created with CoCreateInstance. The correct "controlling unknown" pointer can be retrieved by calling CCmdTarget::GetControllingUnknown. The value returned from that function, however, is not valid during the constructor. For this reason, it is suggested that you create your aggregates only in an override of CCmdTarget::OnCreateAggregates, where the return value from GetControllingUnknown is reliable, even if created from the COleObjectFactory implementation.

It is also important that the object manipulate the correct reference count when adding or releasing artificial reference counts. To ensure this is the case, always call ExternalAddRef and ExternalRelease instead of InternalRelease and InternalAddRef. It is rare to call InternalRelease or InternalAddRef on a class that supports aggregation.

Reference Material

Advanced usage of OLE, such as defining your own interfaces or overriding the framework's implementation of the OLE interfaces requires the use of the underlying interface map mechanism.

This section discusses each macro and the APIs which is used to implement these advanced features.

CCmdTarget::EnableAggregation — Function Description

void EnableAggregation();

Remarks

Call this function in the constructor of the derived class if you wish to support OLE aggregation for objects of this type. This prepares a special IUnknown implementation that is required for aggregatable objects.

CCmdTarget::ExternalQueryInterface — Function Description

DWORD ExternalQueryInterface(
    const void FAR* lpIID,
    LPVOIDFAR* ppvObj
);

Parameters

lpIID
A far pointer to an IID (the first argument to QueryInterface)

ppvObj
A pointer to an IUnknown* (second argument to QueryInterface)

Remarks

Call this function in your implementation of IUnknown for each interface your class implements. This function provides the standard data-driven implementation of QueryInterface based on your object's interface map. It is necessary to cast the return value to an HRESULT. If the object is aggregated, this function will call the "controlling IUnknown" instead of using the local interface map.

CCmdTarget::ExternalAddRef — Function Description

DWORD ExternalAddRef();

Remarks

Call this function in your implementation of IUnknown::AddRef for each interface your class implements. The return value is the new reference count on the CCmdTarget object. If the object is aggregated, this function will call the "controlling IUnknown" instead of manipulating the local reference count.

CCmdTarget::ExternalRelease — Function Description

DWORD ExternalRelease();

Remarks

Call this function in your implementation of IUnknown::Release for each interface your class implements. The return value indicates the new reference count on the object. If the object is aggregated, this function will call the "controlling IUnknown" instead of manipulating the local reference count.

DECLARE_INTERFACE_MAP — Macro Description

DECLARE_INTERFACE_MAP

Remarks

Use this macro in any class derived from CCmdTarget that will have an interface map. Used in much the same way as DECLARE_MESSAGE_MAP. This macro invocation should be placed in the class definition, usually in a header (.H) file. A class with DECLARE_INTERFACE_MAP must define the interface map in the implementation file (.CPP) with the BEGIN_INTERFACE_MAP and END_INTERFACE_MAP macros.

BEGIN_INTERFACE_PART and END_INTERFACE_PART — Macro Descriptions

BEGIN_INTERFACE_PART(localClass, iface);
END_INTERFACE_PART(localClass)

Parameters

localClass
The name of the class that implements the interface

iface
The name of the interface that this class implements

Remarks

For each interface that your class will implement, you need to have a BEGIN_INTERFACE_PART and END_INTERFACE_PART pair. These macros define a local class derived from the OLE interface that you define as well as an embedded member variable of that class. The AddRef, Release, and QueryInterface members are declared automatically. You must include the declarations for the other member functions that are part of the interface being implemented (those declarations are placed between the BEGIN_INTERFACE_PART and END_INTERFACE_PART macros).

The iface argument is the OLE interface that you wish to implement, such as IAdviseSink, or IPersistStorage (or your own custom interface).

The localClass argument is the name of the local class that will be defined. An 'X' will automatically be prepended to the name. This naming convention is used to avoid collisions with global classes of the same name. In addition, the name of the embedded member, the same as the localClass name except it is prefixed by 'm_x'.

For example:

BEGIN_INTERFACE_PART(MyAdviseSink, IAdviseSink)
    STDMETHOD_(void, OnDataChange)(LPFORMATETC, LPSTGMEDIUM);
    STDMETHOD_(void, OnViewChange)(DWORD, LONG);
    STDMETHOD_(void, OnRename)(LPMONIKER);
    STDMETHOD_(void, OnSave)();
    STDMETHOD_(void, OnClose)();
END_INTERFACE_PART(MyAdviseSink)

would define a local class called XMyAdviseSink derived from IAdviseSink, and a member of the class in which it is declared called m_xMyAdviseSink.Note:

Note

The lines beginning with STDMETHOD_ are essentially copied from OLE2.H and modified slightly. Copying them from OLE2.H can reduce errors that are hard to resolve.

BEGIN_INTERFACE_MAP and END_INTERFACE_MAP — Macro Descriptions

BEGIN_INTERFACE_MAP(theClass, baseClass)
END_INTERFACE_MAP

Parameters

theClass
The class in which the interface map is to be defined

baseClass
The class from which theClass derives from.

Remarks

The BEGIN_INTERFACE_MAP and END_INTERFACE_MAP macros are used in the implementation file to actually define the interface map. For each interface that is implemented there is one or more INTERFACE_PART macro invocations. For each aggregate that the class uses, there is one INTERFACE_AGGREGATE macro invocation.

INTERFACE_PART — Macro Description

INTERFACE_PART(theClass, iid, localClass)

Parameters

theClass
The name of the class that contains the interface map.

iid
The IID that is to be mapped to the embedded class.

localClass
The name of the local class (less the 'X').

Remarks

This macro is used between the BEGIN_INTERFACE_MAP macro and the END_INTERFACE_MAP macro for each interface your object will support. It allows you to map an IID to a member of the class indicated by theClass and localClass. The 'm_x' will be added to the localClass automatically. Note that more than one IID may be associated with a single member. This is very useful when you are implementing only a "most derived" interface and wish to provide all intermediate interfaces as well. A good example of this is the IOleInPlaceFrameWindow interface. Its hierarchy looks like this:

IUnknown
    IOleWindow
        IOleUIWindow
            IOleInPlaceFrameWindow

If an object implements IOleInPlaceFrameWindow, a client may QueryInterface on any of these interfaces: IOleUIWindow, IOleWindow, or IUnknown, besides the "most derived" interface IOleInPlaceFrameWindow (the one you are actually implementing). To handle this you can use more than one INTERFACE_PART macro to map each and every base interface to the IOleInPlaceFrameWindow interface:

in the class definition file:

BEGIN_INTERFACE_PART(CMyFrameWindow, IOleInPlaceFrameWindow)

in the class implementation file:

BEGIN_INTERFACE_MAP(CMyWnd, CFrameWnd)
    INTERFACE_PART(CMyWnd, IID_IOleWindow, MyFrameWindow)
    INTERFACE_PART(CMyWnd, IID_IOleUIWindow, MyFrameWindow)
    INTERFACE_PART(CMyWnd, IID_IOleInPlaceFrameWindow, MyFrameWindow)
END_INTERFACE_MAP

The framework takes care of IUnknown because it is always required.

INTERFACE_PART — Macro Description

INTERFACE_AGGREGATE(theClass, theAggr)

Parameters

theClass
The name of the class that contains the interface map,

theAggr
The name of the member variable that is to be aggregated.

Remarks

This macro is used to tell the framework that the class is using an aggregate object. It must appear between the BEGIN_INTERFACE_PART and END_INTERFACE_PART macros. An aggregate object is a separate object, derived from IUnknown. By using an aggregate and the INTERFACE_AGGREGATE macro, you can make all the interfaces that the aggregate supports appear to be directly supported by the object. The theAggr argument is simply the name of a member variable of your class which is derived from IUnknown (either directly or indirectly). All INTERFACE_AGGREGATE macros must follow the INTERFACE_PART macros when placed in an interface map.

See also

Technical Notes by Number
Technical Notes by Category