patent
EP0669020
filed
1993-11-12
published
1994-05-26
granted
1997-04-02
score
-
bayes
-
votes
-

A method and system for marshalling interface pointers for remote procedure calls.

applicantMicrosoft corp (US)
inventorsAtkinson, Robert, G. (US) Corbett, Tom (US) Jung, Edward, K. (US) Williams, Antony, S. (US)
eclaNone
ipcG06F9/44; G06F9/46
designatedDE,FR,GB
priorityWO1993US10965 19931112; US19920975775 19921113
applicationEP19940901451 19931112
state (internal)68
claims0

abstract

A computer method and system for passing a pointer to an interface from a server process to a client process. In a preferred embodiment, the server process instantiates an object that has multiple interfaces. The server process identifies an interface to pass to the client process and creates a stub object for receiving a request to invoke a function member of the interface and for invoking the requested function member upon receiving the request. The server process then sends an identifier of the stub to the client process. When the client process receives the identifier of the stub, it instantiates a proxy object for receiving requests to invoke a function member of the interface and for sending the request to the identified stub. The client process can then invoke the function members of the interface by invoking function members of the proxy object. The proxy object sends a request to the identified stub. The identified stub then invokes the corresponding function member of the interface.

description

Description of correspondent: WO9411810


Description
A METHOD AND SYSTEM FORMARSHALLING INTERFACE
POINTERS FOR REMOTE PROCEDURE CALLS
Technical Field
This invention relates generally to a computer method and system for passing data and, more specifically, to a method and system for passing pointers to objects between processes.

Background of the Invention
Computer systems typically have operating systems that support multitasking. A multitasking operating system allows multiple tasks (processes) to be executing concurrently. For example, a database server process may execute concurrently with many client processes, which request services of the database server process. A client process (client) may request services by issuing a remote procedure call (RPC). A remote procedure call allows a server process (server) to invoke a server procedure on behalf of the client. To issue a remote procedure call, the client packages the procedure name and the actual in-parameters for the procedure into an interprocess communications message and sends the message to the server.The server receives the message, unpackages the procedure name and any actual in-parameters, and invokes the named procedure, passing it the unpackaged in-parameters.Men the procedure completes, the server packages any out-parameters into a message and sends the message to the client. The client receives the message and unpackages the out-parameters. The process of packaging parameters is known as marshalling, and the process of unpackaging parameters is known as unmarshalling.

Parameters may be marshalled by storing a copy of the value of the actual parameter in a message. For certain types of parameters, the marshalling may involve more than simply storing a copy of the value. For example, a floating point value may need to be converted from the format of one computer system to the format of another computer system when the processes reside on different computer systems.

The copying of the value of an actual parameter has a couple disadvantages. First, when a copy is passed changes to the original value are not reflected in the copy. For example, if a parameter representing a time of day value is passed from a client to a server by copying, then the copy that the server receives is not updated as the client updates its time of day value. Second, with certain types of parameters, it may be impractical to make a copy of the value.

For example, the overhead of copying a large array may be unacceptable. As discussed in the following, it may also be impractical to make a copy of an object when marshalling the object because the object may be large and include various functions.

The use of object-oriented programming techniques have many advantages over prior techniques. Thus, the use of object-oriented techniques is increasing. However, the inability to efficiently marshal and unmarshal objects when invoking remote procedures limit these advantages. A description of how objects are typically marshalled and unmarshalled will help to explain these limits.

Figure 1 is a block diagram illustrating typical data structures used to represent an object. An object is composed of instance data (data members) and member functions, which implement the behavior of the object. The data structures used to represent an object comprise instance data structure 101, virtual function table 102, and the function members 103, 104, 105. The instance data structure 102 contains a pointer to the virtual function table 102 and contains data members. The virtual function table 102 contains an entry for each virtual function member defined for the object. Each entry contains a reference to the code that implements the corresponding function member. The layout of this sample object conforms to the model defined in U.S.Patent Application Serial
No. 07/682,537, entitled "A Method for Implementing Virtual Functions and
Virtual Bases in a Compiler for an Object Oriented Programming Language," which is hereby incorporated by reference. In the following, an object will be described as an instance of a class as defined by the C+ + programming language.

One skilled in the art would appreciate that objects can be defined using other programming languages.

If an object in a server process is to be copied and passed to a client process during a remote procedure call, then not only the data members must be copied, but also the function members must be accessible to the client process. To access the copied object, the client process would need to load each function member into its own process space. This loading can be time consuming
Moreover, the copying of an object may be intractable because a function member loaded in the server process space may need to access data or other functions in the server process space.

An advantage of using object-oriented techniques is that these techniques can be used to facilitate the creation of compound documents. A compound document is a document that contains objects generated by various computer programs. (Typically, only the data members of the object and the class type are stored in a compound document.) For example, a word processing document that contains a spreadsheet object generated by a spreadsheet program is a compound document. A word processing program allows a user to embed a spreadsheet object (e.g., a cell) within a word processing document. To allow this embedding, the word processing program would be compiled using the class definition of the object to be embedded to access function members of the embedded object.Thus, the word processing program would need to be compiled using the class definition of each class of objects that can be embedded in a word processing document. To embed an object of a new class into a word processing document, the word processing program would need to be recompiled with the new class definition. Thus, only objects of classes selected by the developer of the word processing program can be embedded. Furthermore, new classes can only be supported with a new release of the word processing program.

To allow objects of an arbitrary class to be embedded into compound documents, interfaces (abstract classes) are defined through which an object can be accessed without the need for the word processing program to have access to the class definitions at compile time. An abstract class is a class in which a virtual function member has no implementation (pure). An interface is an abstract class with no data members and whose virtual functions are all pure.

The following class definition is an example definition of an interface. In this example, for simplicity of explanation, rather than allowing any class of object to be embedded in its documents, a word processing program allows spreadsheet objects to be embedded. Any spreadsheet object that provides this interface can be embedded, regardless of how the object is implemented.

Moreover, any spreadsheet object, whether implemented before or after the word processing program is compiled, can be embedded.

class ISpreadSheet
{ virtual void File() = 0;
virtual void Edit() = 0;
virtual void Formula() = 0;
virtual void Format() = 0;
virtual void GetCell (string RC, cell *pCell) = 0;
virtual void Data() = 0;
.

The developer of a spreadsheet program would need to provide an implementation of the interface to allow the spreadsheet objects to be embedded in a word processing document. When the word processing program embeds a spreadsheet object, the program needs access to the code that implements the interface for the spreadsheet object. To access the code, each implementation is given a unique class identifier. For example, a spreadsheet object developed by
Microsoft Corporation may have a class identifier of "MSSpreadsheet," while a spreadsheet object developed by another corporation may have a class identifier of "LTSSpreadsheet." A persistent registry in each computer system is maintained that maps each class identifier to the code that implements the class.Typically, when a spreadsheet program is installed on a computer system, the persistent registry is updated to reflect the availability of that class of spreadsheet objects.

So long as a spreadsheet developer implements each function member defined by the interface and the persistent registry is maintained, the word processing program can embed the developer's spreadsheet objects into a word processing document.

Various spreadsheet developers may wish, however, to implement only certain function members. For example, a spreadsheet developer may not want to implement database support, but may want to support all other function members. To allow a spreadsheet~developer to support only some of the function members, while still allowing the objects to be embedded, multiple interfaces for spreadsheet objects are defined. For example, the interfaces IDatabase and
IBasic may be defined for a spreadsheet object as follows.

class IDatabase
{ virtual void Data() = 0;
} class IBasic
{ virtual void File() = 0;
virtual void Edit() = 0;
virtual void Formula() = 0;
virtual void Format() = 0;
virtual void GetCell (string RC, cell *pCell) = 0;
.

Each spreadsheet developer would implement the IBasic interface and, optionally, the IDatabase interface.

At run time, the word processing program would need to determine whether a spreadsheet object to be embedded supports the IDatabase interface.

Tomake this determination, another interface is defined (that every spreadsheet object implements) with a function member that indicates which interfaces are implemented for the object. This interface is known as IUnknown and is defined by the following.

class IUnknown
{ virtual booleanQuerylnterface (iidInterface, pInterface) = 0;
. . .

The IUnknown interface defines the function member (method)Querylnterface.

The methodQueryinterface is passed an interface identifier (e.g., "IDatabase") and returns a pointer to the implementation of the identified interface for the object for which the method is invoked. If the object does not support the interface, then the method returns a false.

The IDatabase interface and IBasic interface inherit the IUnknown interface. Inheritance is well known in object-oriented techniques by which a class definition can incorporate the data and function members of previously-defined classes. The following definitions illustrate the use of the IUnknown interface.

class IDatabase :IUnknown
{ virtual void Data() = 0;
} class IBasic: IUnknown
{ virtual void File() = 0;
virtual void Edit() = 0;
virtual void Formula() = 0;
virtual void Format() = 0;
virtual void GetCell (string RC, cell *pCell) = 0;
Figure 2 is a block diagram illustrating a sample data structure of a spreadsheet object. The spreadsheet object comprises interface data structure 201, IBasic interface data structure 202, IDatabase interface data structure 205, and methods 208 through 212. The interface data structure 201 contains a pointer to each interface implemented and may contain data members of the implementation. The IBasic interface data structure 202 contains instance data structure 203 and virtual function table 204. Each entry in the virtual function table 204 points to a method defined for the IBasic interface.The IDatabase interface data structure 205 contains instance data structure 206 and virtual function table 207. Each entry in the virtual function table 207 contains a pointer to a method defined in the IDatabase interface. Since the IBasic and IDatabase interfaces inherit the IUnknown interface, each virtual function table 204 and 207contains a pointer to the methodQuerylnterface 208. In the following, an object data structure is represented by the shape 213 labelled with an interface through which the object may be accessed.

The following pseudocode illustrates how a word processing program determines whether a spreadsheet object supports the IDatabase interface.

if(pIBasic- > Querylnterface("IDatabase", & IDatabase))
\* IDatabase supported
else
\* IDatabase not supported
The pointer pIBasic is a pointer to the IBasic interface of the object. If the object supports the IDatabase interface, the methodQuerylnterface sets the pointerpIDatabase to point to the IDatabase data structure and returns true as its value.

Normally, an object can be instantiated (an instance of the object created in memory) by a variable declaration or by the "new" operator. However, both techniques of instantiation need the class definition at compile time. A different technique is needed to allow a word processing program to instantiate a spreadsheet object at run time. One technique provides an interface called
IClassFactory, which is defined in the following.

class IClassFactory: IUnknown
{
virtual voidCreatelnstance (iidInterface, & lnterface) =0;
}
The methodCreatefnstance instantiates an object and returns a pointerinterface to the interface of the object designated by argumentiidInterface.

Although the use of the above described interfaces can be used to facilitate embedding objects in a compound document, an efficient technique is needed for allowing pointers to objects (interfaces) to be passed as parameters in a remote procedure call. The passing of pointers avoids the overhead of copying objects and allows the receiving process to see changes that the sending process.

makes to the object.

Surnmarv of the Invention
It is an object of the present invention to provide a method and system for allowing a client process to access an interface of an object instantiated in a server process.

It is another object of the present invention to provide a method and system for allowing an object to implement methods for class-specific (custom) marshalling and unmarshalling of pointers to the object.

It is another object of the present invention to provide a method and system for passing pointers to interfaces of object between processes.

These and other objects, which will become apparent as the invention is more fully described below, are obtained by a method and system for passing a pointer to an interface from a server process to a client process. In a preferred embodiment, the server process instantiates an object that has multiple interfaces. The server process identifies an interface to pass to the client process and creates a stub object for receiving a request to invoke a function member of the interface and for invoking the requested function member upon receiving the request. The server process then sends an identifier of the stub to the client process. When the client process receives the identifier of the stub, it instantiates a proxy object for receiving requests to invoke a function member of the interface and for sending the request to the identified stub. In another embodiment, the server process sends to the client process the class of the proxy object and the client dynamically loads code to instantiate the proxy object. In another embodiment, objects implement an interface with methods for custom marshalling and unmarshalling pointers to interfaces of the object. These methods are invoked to marshal and unmarshal pointers to interfaces.

Brief Description of theDrawings
Figure 1 is a block diagram illustrating typical data structures used to represent an object.

Figure 2 is a block diagram illustrating a sample data structure of a spreadsheet object.

Figure 3 is a block diagram illustrating the data structures generated and code loaded during marshalling and unmarshalling.

Figures 4A through 4C are block diagrams illustrating the marshalling of an interface pointer.

Figure 5 is a flow diagram illustrating the method GetCell.

Figure 6 is a flow diagram of the function MarshalInterface.

Figure 7 is a flow diagram of a sample implementation of the method MarshalInterface.

Figure 8 is a flow diagram of the function MakeStub.

Figure 9 is a flow diagram of the function UnMarshalInterface.

Figure 10 is a flow diagram of a sample implementation of the method UnMarshalInterface of the IMarshal interface.

Figure 11 is a flow diagram of the function GetClassObject.

Figures 12A and 12B are block diagrams illustrating custom marshalling to avoid proxy-to-proxy messages.

Figure 13 is a block diagram illustrating custom marshalling with shared memory.

Detailed Description of the Invention
The present invention provides a method and system for passing pointers to objects as parameters in a remote procedure call. In a preferred embodiment, a server process passes a pointer to an interface of an object to a client process. The server marshals a pointer to the interface and sends the marshalled pointer to the client. The client process unmarshals the pointer and accesses the passed object using the unmarshalled pointer. The marshalling and unmarshalling techniques of the present invention load code and generate data structures to support the accessing of the object by the client.

Figure 3 is a block diagram illustrating the data structures generated and code loaded during marshalling and unmarshalling. The data structures and code include the object 301 and stub object 302 within the server, and proxy object 303 within the client. The proxy 303 and stub 302 are created when a pointer to object 301 is passed to the client. The marshalling process of the server marshals the pointer to an interface of object 301 by loading the code for stub 302, assigning an interprocess communications message address for the stub, storing a pointer to an interface of object 301 within the stub, and packaging the message address of the stub and a class identifier of the proxy into a message. The server then sends the message to the client.When the client receives the message, the client unmarshals the pointer to an interface of object 301 by retrieving the class identifier of the proxy and the stub message address, dynamically loading code to create an instance of the class identifier class identified by the retrieved instantiating proxy 303, and storing the stub message address with the proxy 303.

The client then accesses the interface of object 301 through proxy 303.

Proxy 303 is an object that implements the same interface as the interface of object 301, but with a differentimplementation. Each method of proxy 303 marshals its name and its actual parameters into a message, sends the message to stub 302, waits for a return message from stub 302, and unmarshals any returned parameters.Table 1 lists sample pseudocode for a proxy method named "File."
EMI9.1


<tb> <SEP> Table <SEP> 1
<tb> void <SEP> File <SEP> (string <SEP> Name, <SEP> int <SEP> OpenMode, <SEP> int <SEP> Status)
<tb> <SEP> {package <SEP> "File" <SEP> into <SEP> message
<tb> <SEP> package <SEP> Name <SEP> into <SEP> message
<tb> <SEP> package <SEP> OpenMode <SEP> into <SEP> message
<tb> <SEP> send <SEP> message <SEP> to <SEP> stub
<tb> <SEP> wait <SEP> for <SEP> message <SEP> from <SEP> stub
<tb> <SEP> unpackage <SEP> Status <SEP> from <SEP> message
<tb> <SEP> }
<tb>
Stub 302 is an object that implements an interface that receives messages from the client, unmarshals the method name and any in-parameters, invokes the named method of object 301, marshals any out-parameters into a message, and sends the message to the proxy 303.Table 2 lists sample pseudocode for a stub.
EMI10.1


<tb> <SEP> Table <SEP> 2
<tb> void <SEP> Stub
<tb> <SEP> (while <SEP> (true) <SEP>
<tb> <SEP> (wait <SEP> for <SEP> message <SEP> from <SEP> a <SEP> client
<tb> <SEP> unpackage <SEP> method <SEP> name
<tb> <SEP> unpackage <SEP> any <SEP> parameters
<tb> <SEP> invoke <SEP> named <SEP> method <SEP> for <SEP> the <SEP> object
<tb> <SEP> package <SEP> return <SEP> parameters
<tb> <SEP> send <SEP> message <SEP> to <SEP> the <SEP> client
<tb> <SEP> }
<tb>

Figures 4A through 4C are block diagrams illustrating the marshalling of an interface pointer. Server 401 contains spreadsheet object 403, cell object 404, and cell object 405. Spreadsheet object 403 has an implementation of the IBasic interface. The method GetCell of the IBasic interface is defined by the following prototype.

void GetCell (string RC, cell *pCell)
The method GetCell retrieves a pointer to the cell object representing the cell specified by the passed row/column designator (RC) and returns a pointer to that cell (pCell). The cell objects 404 and 405 have an implementation of the interface
IData, which is defined by the following.

class IData:IUnknown
{ virtual string GetFormula();
virtual void SetFormula(string formula);
..

The method GetFormula returns the formula value of the cell object as a string, and the method SetFormula sets the formula in the cell object to the passed string.

Figure 4B is a block diagram illustrating the marshalling of a cell object to the client. When a client wants to retrieve the formula of cellAl represented as cell object 404, the client executes the following statements.

plBasic- > GetCell("Al", pCell);
formula = pCell- > GetFormula();
The spreadsheet proxy 407 is pointed to by pointer pIBasic 408. The client first invokes the spreadsheet proxy method GetCell. The method GetCell packages the method name "GetCell" and the string "Al" into a message and sends the message to spreadsheet stub 406. The spreadsheet stub 406 unpackages the method name and string. The spreadsheet stub 406 then invokes the GetCell method of the spreadsheet object 403. The method GetCell returns to the spreadsheet stub 406 a pointer to cell object 404. The spreadsheet stub 406 then marshals the cell pointer by creating cell stub 409 for cell object 404, assigning a message address to cell stub 409, packaging the message address and an unmarshal class identifier (described below) into a message, and sending the message to the spreadsheet proxy 407.When the spreadsheet proxy 407 receives the message, method GetCell then unmarshals the pointer to the cell object 404.

Figure 4C is a block diagram illustrating the data structures used during the unmarshalling of a cell pointer. The method GetCellunmarshals the pointer to cell Al by first retrieving the message address and the unmarshal class identifier ("CellProxy") from the received message. In a preferred embodiment, the persistent registry 413 contains the name of a dynamic link library for each class. Dynamic link libraries are libraries of routines (methods, functions, etc.) that are loaded at run time of a program. Dynamic link libraries are described in the reference "Programming Windows," by Charles Petzold and published by
Microsoft Press. Each dynamic link library for a class contains a functionGetiClassFactory which returns a pointer to an IClassFactory interface for the class. The method GetCell loads the dynamic link library 410 for the retrieved unmarshal class identifier (the unmarshal class) and invokes the function
GetIClassFactory which returns a pointer to the IClassFactory interface 411. The method GetCell then invokes the methodCreatelnstance of the IClassFactory interface 411 to create cell proxy 412. The cell proxy 412 is then initialized with the retrieved message address for cell stub 409. The method GetCell then returns with the pointer to the cell proxy 412. The client can then access the cell object 404 through the cell proxy 412.

Figure 5 is a flow diagram illustrating the method GetCell. Steps 501 through 505 compose a flow diagram for the method GetCell of the IBasic interface of a spreadsheet proxy. Steps 506 through 512 compose a flow diagram for a spreadsheet stub. The method GetCell marshals the in-parameter, sends the marshalled parameters to the spreadsheet stub, and receives a marshalled pointer to the cell object indicated by the in-parameter. In step 501, the method invokes the function Marshal to marshal the method name (i.e. "GetCell") into the message 501A. The functions Marshal and UnMarshal package and unpackage data into a message. In step 502, the method invokes the function Marshal to marshal the in-parameter RC into the message 502A.In step 503, the method sends the message to the spreadsheetstub In step 504, the method receives a message from the spreadsheet stub. In step 505, the method invokes function
UnMarshalInterface to unmarshal the interface pointer received in message 505A.

The method then returns with the pointer pCell pointing to a cell proxy.

In step 506, the stub receives a message sent from a client and the message address of the cell proxy that sent the message. In step 507, the stub invokes function UnMarshal to unmarshal the method name from message507at
In step 508, if the method name is "GetCell", then the stub continues at step 509, else the stub continues to determine if the method name corresponds to another method of the IBasic interface (as indicated by the ellipsis). In step 509, the stub invokes function UnMarshal to unmarshal the in-parameter designating the cell location from message 509A. In step 510, the stub invokes the method GetCell for the spreadsheet object passing the cell location. The method GetCell returns a pointer to the cell object for the passed location. In step 511, the stub invokes the function MarshalInterface passing it the pointer to the cell object.In step 512, the stub sends the message 511A to the cell proxy that sent the message received in step 506. The stub then loops to step 506 to wait for the next message.

The above-described marshalling techniques are referred to as standard marshalling. In a preferred embodiment, the present invention allows an object to specify how pointers to it are to be marshalled in a process referred to a custom marshalling. Each object that implements custom marshalling provides an implementation of a custom marshalling interface called IMarshal. The IMarshal interface provides function members to marshal and unmarshal an interface pointer. When marshalling and unmarshalling a pointer to an interface, the methodQueryinterface of the object is invoked to determine whether the object implements the IMarshal interface. If the object implements the IMarshal interface, then the function members of that interface are invoked to marshal and unmarshal the interface pointer.In Figure 5, the functionsUnMarshalInterface (step 505) andMarshallnterface (step 511) preferably determine if the object implements the IMarshal interface and invokes the function members of the
IMarshal interface as appropriate.

Figures 6 through 10 are flow diagrams of the methods and functions invoked during the marshalling and unmarshalling of a pointer to a cell object. As described below, these methods and functions implement standard marshalling.

Figure 6 is a flow diagram of the function MarshalInterface. The function has the following prototype.

void MarshalInterface (pstm, iid, pInterface, DestContext)
This function marshals the designated pointer (pInterface) to an interface for an object into the designated message (pstm). In step 601, the function determines whether the object implements custom marshalling. The function invokes the methodQueryinterface of the interface to retrieve a pointer to an IMarshal interface. If the object implements custom marshalling, then a pointer (pMarshal) to the IMarshal interface for the object is returned and the function continues at step 603, else the function continues at step 602. In step 602, the function invokes the function GetStandardMarshal to retrieve a pointer (pMarshal) to an IMarshal interface with default marshalling methods. In step 603, the function invokes the method IMarshal::GetUnmarshalClass pointed to by the retrieved pointer.The method GetUnmarshalClass returns the class identifier of the class that should be used in the unmarshalling process to instantiate a proxy for the designated interface (iid). In step 604, the function invokes the function Marshal to marshal the unmarshal class identifier to the designated message. In step 605, the function invokes the methodIMarshal::Marshalinterface pointed to by the retrieved pointer (pMarshal). The method MarshalInterface marshals the designated interface pointer (pInterface) to the designated message. The method then returns.

Figure 7 is a flow diagram of a sample standard implementation of the method MarshalInterface. The method has the following prototype.

void IMarshal::MarshalInterface (pstm, iid, pInterface, DestContext)
The method MarshalInterface marshals the designated pointer to an interface (pInterface) to the designated message (pstm). In step 701, the method invokes function MakeStub. The functionMakeStub makes a stub for the designated pointer to the interface and returns the message address of the stub. In step 702, the method invokes the function Marshal to marshal the returned message address in the designated message. The method then returns.

Figure 8 is a flow diagram of the function MakeStub. The function has the following prototype.

void MakeStub (cid, iid, pInterface, MessageAddress)
The function MakeStub makes a stub for the designated pointer (pInterface) to the designated interface (iid) and returns the message address (MessageAddress) of the stub. In step 801, the function loads a copy of the stub code for the designated interface. In step 802, the function stores the designated pointer. to the interface so that it is accessible to the stub code. The storing of the pointer associates the stub code with the object pointed to by the designated pointer. In step 803, the function registers the stub code with the messaging system. The messaging system returns the message address of the stub. In step 804, the function stores the message address to return to the caller. The function
MakeStub then returns.

Figure 9 is a flow diagram of the functionUniMarshalinterface. The function has the following prototype.

void UnMarshalInterface (pstm, iid, pInterface)
The function UnMarshalInterface unmarshals a pointer to an interface that was previously marshalled in the designated message (pstm) and returns a pointer (pInterface) to the designated interface (iid) of the object. In step 901, the function invokes function UnMarshal to unmarshal the unmarshal class identifier from the designated message. In step 902, the function invokes function
GetClassObject to retrieve an IClassFactory interface for the class indicated by the unmarshal class identifier (the unmarshal class). In step 903, the function invokes the methodCreatelnstance of the retrieved IClassFactory interface to create an instance of the proxy and returns a pointer to its IMarshal interface. In step 904, the function invokes method UnMarshalInterface of the proxy. The function UnMarshal then returns.

Figure 10 is a flow diagram of a sample implementation of the method UnMarshalInterface. The method has the following prototype.

voidiMarshal::UnMarshallnterface (pstm, iid, pInterface)
The method UnMarshalInterface initializes the newly created proxy and returns a pointer (pInterface) to the designated interface (iid). In step 1001, the method invokes function UnMarshal to unmarshal the stub message address from the designated message. In step 1002, the method stores the stub message address. In step 1003, the method retrieves a pointer to the designated interface. The method
UnMarshalInterface returns.

Figure 11 is a flow diagram of the function GetClassObject. The function has the following prototype.

void GetClassObject (cid, iid, pInterface)
The function GetClassObject returns a pointer (pInterface) to the designated interface (iid) of the designated class (cid). In step 1101, the function retrieves the name of the dynamic link library for the designated class from the persistent registry. In step 1102, the function loads the dynamic link library. In step 1103, if the designated interface is the IClassFactory, then the function continues at step 1104, else the function returns. In step 1104, the function invokes the function
GetIClassFactory to get the pointer to the IClassFactory interface. The function
GetIClassFactory is preferably provided by the developer of the object. The function GetClassObject then returns.

A developer of an object can provide an implementation of the
IMarshal interface to provide custom marshalling and unmarshalling for the object. The IMarshal interface is defined in the following.

class IMarshal: IUnknown
{ virtual void GetUnmarshalClass (iid, pInterface, DestContext, cid) = 0;
virtual void MarshalInterface(pstm, iid, pInterface, DestContext) = 0;
virtual void UnMarshalInterface(pstm, iid, pInterface) = 0;
)
A developer may implement custom marshalling to optimize the marshalling process. Figures 12A and 12B are block diagrams illustrating custom marshalling of interface pointers. Figure 12A shows the effects of standard marshalling. In Figure 12A, cell object 1201 in processP1 has been marshalled to process P2 as represented by cell stub 1202 and cell proxy 1203. Cell proxy 1203 has been marshalled to process P3 as represented by cell stub 1204 and cell proxy 1205. Cell proxy 1203 was marshalled using the standard marshalling of the type shown in Figure 7.When process P3 invokes a proxy method of cell proxy 1205, the proxy method marshals the parameters and sends a message to cell stub 1204.

Cell stub 1204 receives the message, unmarshals the parameters, and invokes the appropriate proxy method of cell proxy 1203. The proxy method of cell proxy 1203 marshals the parameters and sends a message to cell stub 1202. Cell stub 1202 receives the message, unmarshals the parameters, and invokes the appropriate method of cell object 1201. When the method returns, cell stub 1202 marshals any out-parameters and sends a message to cell proxy 1203. When cell proxy 1203 receives the message, the proxy method unmarshals the parameters and returns to its caller, cell stub 1204. Cell stub 1204 marshals and out-parameters and sends a message to cell proxy 1205. When cell proxy 1205 receives the message, the proxy method unmarshals the parameters and returns to its caller. Thus, whenever process P3 accesses the cell object 1201, the access is routed through process P2.

Figure 12B shows an optimization resulting from custom marshalling. In Figure 12B, when process P3 accesses cell object 1201, the access is not routed through process P2 but rather is routed directly to process P1. To achieve this optimization, the IMarshal interface for a cell proxy is implemented with custom marshalling. The methodIMarshal::Marshallnterface is implemented as shown by the following pseudocode.

void IMarshal::MarshalInterface (pstm, iid, pInterface, DestContext)
{ Marshal (pstm,this- > MessageAddress);
}
As discussed above, the standard IMarshal::Marshallnterface creates a stub and sends the message address of the stub. However, the custom
IMarshal::MarshalInterface simply sends the message address of the stub to which it communicates. Thus, when the custom IMarshal::MarshalInterface of cell proxy 1203 is invoked, the message address of cell stub 1202 is sent to process P3, along with the unmarshal class identifier (CellProxy). When process P3 initializes cell proxy 1205 using the standard unmarshalling method, it stores the message address of cell stub 1202. Thus, when process P3 access cell object 1201, it bypasses process P2.

Figure 13 is a block diagram illustrating custom marshalling with shared memory. In certain situations, an object may store its data members in memory that is shared across processes to avoid the overhead of remote procedure calling to access the data members. In Figure 13, the data 1302 for cell object 1301 is stored in sharedmemoir. Cell object 1303 is created when a pointer to cell object 1301 is marshalled and sent from process P1 to process P2.

The custom methods of the IMarshal interface for a cell object are implemented by the following pseudocode.

void IMarshal::MarshalInterface (pstm, iid, pInterface, DestContext)
{ if (DestContext = = Shared) then
{ Marshal (pstm, address ofdata);}
else
{ MakeStub ("Cell", iid, pInterface, & essageAddress);
Marshal (pstm, MessageAddress)),
) void IMarshal::GetUnmarshalClass (iid,pInterface, DestContext, cid)
{ if (DestContext = = Shared)
{cid = "Cell"}
else
{cid = "CellProxy"}
} void IMarshal::UnMarshalInterface (pstm, iid, pInterface);
{ UnMarshal (pstm, address of data);
initialize object to point to shared data
)
The parameter DestContext of the method IMarshal::Marshallnterface indicates whether the data of the cell object is stored in shared memory. If the data is not stored in shared memory, then the equivalent of the standard marshalling is performed. If, however, the data is stored in shared memory, then the address of the shared data is marshalled into the message. The method
IMarshal::GetUnmarshalClass determines the unmarshal class based on whether the data is in shared memory. If the data is not in shared memory, then theCellProxy class is the unmarshal class. If the data is in shared memory, then the
Cell class is the unmarshal class. The unmarshal class and address of the shared data are sent to process P2. After process P2 instantiates the cell object 1303, the process P2 invokes the custom method IMarshal::UnMarshalInterface, which unmarshals the address of the shared data and initializes the object to point to the data.

Custom marshalling can be used to provide more efficient access to immutable objects. An immutable object is an object whose data members (state) cannot be changed. When an interface pointer to an immutable object is passed from one process to another, custom marshalling can be used to make a copy of the object in the other process. Thus, when the other process accesses the object, it accesses the local copy and no interprocess communication is needed. To allow copying of immutable objects, the methods of the IMarshal interface of the immutable objects are implemented in the following way. In a preferred embodiment, the method IMarshal::GetUnMarshalClass specifies that the unmarshal class is the same class as the class of the immutable object. This ensures that the same type of object is instantiated by the other process. The method IMarshal::MarshalInterface stores the data members of the immutable object into a message for passing to the other process. The method
IMarshal::MarshalInterface creates no stub. When the other process receives the message, the other process creates an object of the class of the immutable object.

The methodIMarshal::UnMarshallnterface then retrieves the data members from the message and initializes the newly-created object. The other process can then access the immutable object.

Although the present invention has been described in terms of preferred embodiments, it is not intended that the invention be limited to these embodiments. Modifications within the spirit of the invention will be apparent to those skilled in the art. The scope of the present invention is defined by the claims that follow.

claims

Claims of correspondent: WO9411810


Claims

1. A method in a computer system for passing a pointer to an interface of an object from a server process to a client process, the method comprising the steps of:
within the server process,
instantiating the object, the object having a plurality of interfaces, each interface having a function member;
identifying an interface of the object to pass to the client process;
creating a stub for receiving a request to invoke a function member of the interface and for invoking the requested function member upon receiving the request, the stub having an identifier; and
sending the identifier of the stub to the client process; and
within the client process,
receiving the identifier of the stub from the server process; and
instantiating a proxy object for receiving requests to invoke a function member of the interface and for sending the request to the identified stub.

2. The method of claim 1 including the steps of:
within the server process,
sending an unmarshal class identifier identifying the class of the proxy object to instantiate in the instantiating step to the client process; and
within the client process,
receiving the unmarshal class identifier from the server process; and
dynamically loading code to instantiate an object of the class identified by the unmarshal class identifier wherein the step of instantiating a proxy object executes the dynamically-loaded code.

3. The method of claim 2 including the step of:
within the client process,
retrieving a persistent registry entry indicator of the code to dynamically load from a persistent registry, the persistent registry having an entry associating the unmarshal class identifier with the persistent registry entry indicator.

4. A method in a computer system for passing a pointer to an interface of an object from a server process to a client process, the method comprising the steps of:
within the server process,
instantiating the object, the object having a plurality of interfaces, each interface having a function member, the object further having a marshalling function member for performing custom marshalling for a pointer to one of the plurality of interfaces;
identifying an interface of the object to pass to the client process;
invoking the marshalling function member of the object; and
sending an unmarshal class identifier identifying the class of the proxy object to be instantiated by the client process; and
within the client process,
receiving the unmarshal class identifier from the server process;
instantiating an object of the class identified by the unmarshal class identifier, the object of the class identified by the unmarshal class identifier having an unmarshalling function member; and
invoking the unmarshalling function member of the object of the class identified by the unmarshal class identifier, the unmarshalling function member for performing custom unmarshalling for the pointer to the identified interface.

5. The method of claim 4 wherein the step of instantiating an object of the class identified by the unmarshal class identifier includes the step of dynamically loading code to instantiate the object.

6. The method of claim 5 including the step of:
within the client process,
retrieving a persistent registry entry indicator of the code to dynamically load from a persistent registry, the persistent registry having an entry associating the unmarshal class identifier with the persistent registry entry indicator.

7. A method in a computer system of passing a pointer to an object from a server process to a client process, the method comprising the steps of:
within the server process,
instantiating the object;
creating a stub for receiving a request to invoke a function member of the object and for invoking the requested function member upon receiving the request, the stub having an identifier; and
sending an unmarshal class identifier identifying the class of a proxy object to instantiate within the client process and the identifier of the stub to the client process; and
within the client process,
receiving the unmarshal class identifier and the identifier of the stub from the server process;
dynamically loading code to instantiate the proxy object of the class identified by the unmarshal class identifier; and
executing the loaded code to instantiate a proxy object of the class identified by the unmarshal class identifier and to initialize the proxy object so that when the proxy object receives requests to invoke a function member of the object, the proxy object sends the request to the identified stub.

8. The method of claim 7 including the step of:
within the client process,
retrieving a persistent registry entry indicator of the code to dynamically load from a persistent registry, the persistent registry having an entry associating the unmarshal class identifier with the persistent registry entry indicator.

9. A method in a computer system of passing a pointer to an object from a server process to a client process, the method comprising the steps of:
within the server process,
instantiating the object having a marshalling function member for performing custom marshalling for a pointer to the object;
invoking the marshalling function member of the object; and
sending to the client process an unmarshal class identifier identifying the class of a proxy object to instantiate in the client process; and
within the client process,
receiving the unmarshal class identifier from the server process;
dynamically loading code to instantiate the object of the class identified by the unmarshal class identifier;
executing the loaded code to instantiate an object of the class identified by the unmarshal class identifier, the object of the class identified by the unmarshal class identifier having an unmarshalling function member; and
invoking the unmarshalling function member of the object of the class identified by the unmarshal class identifier, the unmarshalling function member for performing custom unmarshalling for the pointer to the object.

10. The method of claim 9 including the step of:
within the client process,
retrieving a persistent registry entry indicator of the code to dynamically load from a persistent registry, the persistent registry having an entry associating the unmarshal class identifier with the persist registry entry indicator.

11. A method in a computer system for passing a pointer to an object from a server process to a first client process and from the first client process to a second client process, the method comprising the steps of:
within the server process,
instantiating the object;
creating a stub for receiving a request to invoke a function member of the object and for invoking the requested function member upon receiving the request, the stub having an identifier; and
sending an unmarshal class identifier identifying the class of an object to be instantiated in the first and second client processes and the identifier of the stub to the first client process;
within the first client process,
receiving the unmarshal class identifier and the identifier of the stub from the server process;
dynamically loading code to instantiate a first proxy object of the class identified by the unmarshal class identifier; ;
executing the loaded code to instantiate a first proxy object of the class identified by the unmarshal class identifier and having a marshalling function member to specify that the identifier of the stub is to be sent to the second client process and to initialize the first proxy object so that when the first proxy object receives requests to invoke a function member of the object, the first proxy object sends the request to the identified stub;
invoking the marshalling function member of the proxy object; and
sending the unmarshal class identifier and the identifier of the stub to the second client process; and
within the second client process,
receiving the unmarshal class identifier and the identifier of the stub from the first client process;
dynamically loading code to instantiate a second proxy object of the class identified by the unmarshal class identifier; and
executing the loaded code to instantiate a second a proxy object of the class identified by the unmarshal class identifier and to initialize the second proxy object so that when the second proxy object receives requests to invoke a function member of the object, the second proxy object sends the request to the identified stub.

12. The method of claim 11 wherein the object has a plurality of interfaces and the pointer to the object is a pointer to one of the interfaces of the object.

13. A method in a computer system of passing a pointer to an object from a server process to a client process, the method comprising the steps of:
within the server process,
instantiating the object in shared memory that is accessible by the server process and the client process and that has a marshalling function member to store a pointer to data members of the object in a message;
storing an unmarshal class identifier in a message;
invoking the marshalling function member of the object; and
sending the message to the client process; and
within the client process,
receiving the message from the server process;
retrieving the unmarshal class identifier from the message;
dynamically loading code to instantiate an object of the class identified by the unmarshal class identifier;
executing the loaded code to instantiate an object of the class identified by the unmarshal class identifier, the object of the class identified by the unmarshal class identifier having an unmarshalling function member; and
invoking the unmarshalling function member of the object of the class identified by the unmarshal class identifier wherein the unmarshalling function member stores the pointer to the data members in the object so that function members of the object can access the data members in shared memory.

14. The method of claim 13 wherein the unmarshal class identifier identifies the same class as the object instantiated by the server process.

15. The method of claim 13 wherein the object instantiated by the server has a plurality of interfaces and the pointer to the object is a pointer to one of the interfaces of the object.

16. A method in a computer system of passing a pointer to an immutable object from a server process to a client process, the method comprising the steps of:
within the server process,
instantiating the immutable object, the immutable object having a marshalling function member to store the data members of the object in the message;
storing an unmarshal class identifier in a message;
invoking the marshalling function member of the object; and
sending the message to the client process; and
within the client process,
receiving the message from the server process;
retrieving the unmarshal class identifier from the message;
dynamically loading code to instantiate an object of the class identified by the unmarshal class identifier;
executing the loaded code to instantiate an object of the class identified by the unmarshal class identifier, the object of the class identified by the unmarshal class identifier having an unmarshalling function member; and
invoking the unmarshalling function member of the object of the class identified by the unmarshal class identifier wherein the unmarshalling function member retrieves the data members from the message and stores the data members in the object so that the function members of the object can access the data members of the immutable object.

17. The method of claim 16 wherein the unmarshal class identifier identifies the same class as the immutable object.

18. The method of claim 16 wherein the immutable object has a plurality of interfaces and the pointer to the immutable object is a pointer to one of the interfaces of the immutable object.

19. An object in a computer system the object having a plurality of data members and function members, the object having: a method for retrieving an unmarshal class identifier; a marshalling method for marshalling a pointer to an object; and an unmarshalling method for unmarshalling the pointer to the object.