Steps for Creating a WebLogic Enterprise Server Application

This chapter describes the basic steps involved in creating a WebLogic Enterprise server application. The steps shown in this chapter are not definitive; there may be other steps you may need to take for your particular server application, and you may want to change the order in which you follow some of these steps. However, the development process for every WebLogic Enterprise server application has each of these steps in common.

This chapter presents the following topics:

• Summary of the WebLogic Enterprise Server Application Development Process

• Development and Debugging Tips

• Servant Pooling

• Delegation-based Interface Implementation

This chapter begins with a summary of the steps, and also lists the development tools and commands used throughout this book. Your particular deployment environment might use additional software development tools, so the tools and commands listed and described in this chapter are also not definitive.

The chapter uses examples from the Basic University sample application, which is provided with the WebLogic Enterprise software. For complete details about the Basic University sample application, see the Guide to the University Sample Applications. For complete information about the tools and commands used throughout this book, see the C++ Programming Reference.

Summary of the WebLogic Enterprise Server Application Development Process

The basic steps to create a server application are:

Step 1: Compile the OMG IDL file for the server application.
Step 2: Write the methods that implement each interface's operations.
Step 3: Create the Server object.
Step 4: Define the in-memory behavior of objects.
Step 5: Compile and link the server application.
Step 6: Deploy the server application.

The WebLogic Enterprise software also provides the following development tools and commands:

Step 1: Compile the OMG IDL file for the server application.

The basic structure of the client and server portions of the application that runs in the WebLogic Enterprise domain are determined by statements in the application's OMG IDL file. When you compile your application's OMG IDL file, the IDL compiler generates some or all of the files shown in the following diagram, depending upon which options you specify in the idl command. The shaded components are the generated files that you modify to create a server application.

The files produced by the IDL compiler are described in Table 2-1.

Table 2-1 Files Produced by the IDL Compiler

Using the IDL Compiler

To generate the files listed in Table 2-1, enter the following command:

idl [options] idl-filename [icf-filename]

In the idl command syntax:

• options represents one or more command-line options to the IDL compiler. The command-line options are described in the C++ Programming Reference. If you want to generate implementation files, you need to specify the -i option.

• idl-filename represents the name of your application's OMG IDL file.

• icf-filename is an optional parameter that represents the name of your application's Implementation Configuration File (ICF file), which you use to specify object activation policies or to limit the number of interfaces for which you want skeleton and implementation files generated. Using the ICF file is described in the section "Step 4: Define the in-memory behavior of objects." on page 2-14.

For more information about the IDL compiler, including details on the idl command, see the C++ Programming Reference.

Generating the Skeleton and Implementation Files

The following command line generates client stub, skeleton, and initial implementation files, along with skeleton and implementation header files, for the OMG IDL file univb.idl:

idl -i univb.idl

For more information about the idl command, see the C++ Programming Reference. For more information about generating these files for the WebLogic Enterprise University sample applications, see the Guide to the University Sample Applications.

Note: If you plan to specify nondefault object activation or transaction policies, or if you plan to limit the number of interfaces for which you want skeleton and implementation files generated, you need to generate and modify an Implementation Configuration File (ICF) and pass the ICF file to the IDL compiler. For more information, see "Specifying Object Activation and Transaction Policies in the ICF File" on page 2-14.

Generating Tie Classes

The IDL compiler also provides the -T command-line option, which you can use for generating tie class templates for your interfaces. For more information about implementing tie classes in a WLE application, see the section "Delegation-based Interface Implementation" on page 2-28.

Step 2: Write the methods that implement each interface's operations.

As the server application programmer, your task is to write the methods that implement the operations for each interface you have defined in your application's OMG IDL file.

The implementation file contains:

• Method declarations for each operation specified in the OMG IDL file

• Your application's business logic, include files, and other data you want the application to use

• Constructors and destructors for each interface implementation (implementing these is optional)

• Optionally, the Tobj_ServantBase::activate_object() and Tobj_ServantBase::deactivate_object() operations

  Within the Tobj_ServantBase::activate_object() and Tobj_ServantBase::deactivate_object()operations, you write code that performs any particular steps related to activating or deactivating an object. This includes reading and writing the object's durable state from and to disk, respectively. If you implement these operations in your object, you must also edit the implementation header file and add the definitions for these operations in each implementation that uses them.

The Implementation File Generated by the IDL Compiler

Although you can create your server application's implementation file entirely by hand, the IDL compiler generates an implementation file that you can use as a starting place for writing your implementation file. The implementation file generated by the IDL compiler contains signatures for the methods that implement each of the operations defined for your application's interfaces.

You typically generate this implementation file only once, using the -i option with the command that invokes the IDL compiler. As you iteratively refine your application's interfaces, and modify the operations for those interfaces, including operation signatures, you add all the required changes to the implementation file to reflect those changes.

Implementing a Factory

As mentioned in the section "How Client Applications Access and Manipulate Your Application's CORBA Objects" on page 1-4, you need to create factories so that client applications can easily locate the objects managed by your server application. A factory is like any other CORBA object that you implement, with the exception that you register it with the FactoryFinder object. Registering a factory is described in the section "Writing the Code That Creates and Registers a Factory" on page 2-10.

The primary function of a factory is to create object references, which it does by invoking the TP::create_object_reference() operation. The TP::create_object_reference() operation requires the following input parameters:

• The Interface Repository ID of the object's OMG IDL interface

• The object ID (OID) in string format

• Optionally, routing criteria

For example, in the Basic University sample application, the RegistrarFactory interface specifies only one operation, as follows:

University::Registrar_ptr RegistrarFactory_i::find_registrar()

The find_registrar() operation on the RegistrarFactory object contains the following invocation to the TP::create_object_reference() operation to create a reference to a Registrar object:

CORBA::Object_var v_reg_oref =
            TP::create_object_reference(
                University::_tc_Registrar->id(),
                object_id,
                CORBA::NVlist::_nil()
            );

In the previous code example, notice the following:

• The following parameter specifies the Registrar object's Interface Repository ID by extracting it from its typecode:

  University::_tc_Registrar->id()

• The following parameter specifies that no routing criteria are used, with the result that an object reference created for the Registrar object is routed to the same group as the RegistrarFactory object that created the object reference:

  CORBA::NVlist::_nil()

  For information about specifying routing criteria that affect the group to which object references are routed, see Chapter 7, "Scaling a WebLogic Enterprise Server Application"

Step 3: Create the Server object.

Implementing the Server object is not like implementing other language objects. The header class for the Server object has already been created, and the Server object class has already been instantiated for you. Creating the Server object involves implementing a specific set of methods in the prepackaged Server object class. The methods you implement are described in this section.

To create the Server object, create a new file using a common text editor and implement the following operations:

There is only one instance of the Server object in any server application. If your server application is managing multiple CORBA object implementations, the Server::initialize(), Server::create_servant(), and Server::release() operations you write must include code that applies to all those implementations.

The code that you write for most of these tasks involves interaction with the TP Framework. The sections that follow explain the code required for each of these Server object operations and shows sample code from the Basic University sample application.

Initializing the Server Application

The first operation that you implement in your Server object is the operation that initializes the server application. This operation is invoked when the WebLogic Enterprise system starts the server application. The TP Framework invokes the following operation in the Server object during the startup sequence of the server application:

CORBA::Boolean Server::initialize(int argc, char** argv) 

Any command-line options specified in the CLOPT parameter for your specific server application in the SERVERS section of the WebLogic Enterprise domain's UBBCONFIG file are passed to the Server::initialize() operation as argc and argv. For more information about passing arguments to the server application, see Administration Guide. For examples of passing arguments to the server application, see the Guide to the University Sample Applications.

Within the Server::initialize() operation, you include code that does the following, if applicable:

• Creates and registers factories

• Allocates any machine resources

• Initializes any global variables needed by the server application

• Opens the databases used by the server application

• Opens the XA resource manager

Writing the Code That Creates and Registers a Factory

If your server application manages a factory that you want client applications to be able to locate easily, you need to write the code that registers that factory with the FactoryFinder object, which is invoked typically as the final step of the server application initialization process.

To write the code that registers a factory managed by your server application, you do the following:

1. Create an object reference to the factory.

   This step involves creating an object reference as described in the section "Implementing a Factory" on page 2-7. In this step, you include an invocation to the TP::create_object_reference() operation, specifying the Interface Repository ID of the factory's OMG IDL interface.The following example creates an object reference, represented by the variable s_v_fact_ref, to the RegistrarFactory factory:

   University::RegistrarFactory s_v_fact_ref =
        TP::create_object_reference(
            University::_tc_RegistrarFactory->id(),
            object_id,
            CORBA::NVList::_nil()
        );

2. Register the factory with the WebLogic Enterprise domain.

   This step involves invoking the following operation for each of the factories managed by the server application:

   TP::register_factory (CORBA::Object_ptr factory_or,
                         const char* factory_id);

   The TP::register_factory() operation registers the server application's factories with the FactoryFinder object. This operation requires the following input parameters:

   • The object reference for the factory, created in step 1 above

   • A string identifier, based on the factory object's interface typecode, used to identify the Interface Repository ID of the factory's OMG IDL interface

     The following example registers the RegistrarFactory factory with the WebLogic Enterprise domain:

     TP::register_factory(s_v_fact_ref.in(),
                          University::_tc_RegistrarFactory->id());

     Notice the parameter University::_tc_RegistrarFactory->id(). This is the same parameter specified in the TP::create_object_reference() operation. This parameter extracts the Interface Repository ID of the object's OMG IDL interface from its typecode.

After the server application initialization process is complete, the server application is ready to begin processing client requests. If a request arrives for an operation on a CORBA object for which there is no servant available in memory, the TP Framework invokes the following operation in the Server object:

Tobj_Servant Server::create_servant(const char* interfaceName)

The Server::create_servant() operation contains code that instantiates a servant for the object required by the client request. For example, in C++, this code includes a new statement on the interface class for the object.

The Server::create_servant() operation does not associate the servant with an OID. The association of a servant with an OID takes place when the TP Framework invokes the Tobj_ServantBase::activate_object() operation on the servant, which completes the instantiation of the object. (You cannot associate an OID with an object in the object's constructor.) Likewise, the disassociation of a servant with an OID takes place when the TP Framework invokes the deactivate_object() operation on the servant.

This behavior of a servant in the WebLogic Enterprise system makes it possible, after an object has been deactivated, for the TP Framework to make a servant available for another object instantiation. Therefore, do not assume that an invocation of an object's Tobj_ServantBase::deactivate_object() operation results in an invocation of that object's destructor. If you use the servant pooling feature in your server application, you can implement the TP::application_responsibility() operation in an object's Tobj_ServantBase::deactivate_object() operation to pass a pointer to the servant to a servant pool for later reuse. Servant pooling is discussed in the section "Servant Pooling" on page 2-27.

The Server::create_servant() operation requires a single input argument. The argument specifies a character string containing the Interface Repository ID of the OMG IDL interface of the object for which you are creating a servant.

In the code you write for this operation, you specify the Interface Repository IDs of the OMG IDL interfaces for the objects managed by your server application. During run time, the Server::create_servant()operation returns the servant needed for the object specified by the request.

The following code implements the Server::create_servant() operation in the University server application from the Basic University sample application:

Tobj_Servant Server::create_servant(const char* intf_repos_id)
{
  if (!strcmp(intf_repos_id, University::_tc_RegistrarFactory->id())) {
      return new RegistrarFactory_i();
  }
  if (!strcmp(intf_repos_id, University::_tc_Registrar->id())) {
      return new Registrar_i();
  }
  if (!strcmp(intf_repos_id, University::_tc_CourseSynopsisEnumerator->id())) {
      return new CourseSynopsisEnumerator_i();
  }
  return 0; // unknown interface
}

When the WebLogic Enterprise system administrator enters the tmshutdown command, the TP Framework invokes the following operation in the Server object of each running server application in the WebLogic Enterprise domain:

void Server::release()

Within the Server::release() operation, you may perform any application-specific cleanup tasks that are specific to the server application, such as:

• Unregistering object factories managed by the server application

• Deallocating resources

• Closing any databases

• Closing an XA resource manager

Once a server application receives a request to shut down, the server application can no longer receive requests from other remote objects. This has implications on the order in which server applications should be shut down, which is an administrative task. For example, do not shut down one server process if a second server process contains an invocation in its Server::release() operation to the first server process.

During server shutdown, you may want to include the following invocation to unregister each of the server application's factories:

TP::unregister_factory (CORBA::Object_ptr factory_or,
				const char* factory_id)

The invocation of the TP::unregister_factory() operation should be one of the first actions in the Server::release() implementation. The TP::unregister_factory() operation unregisters the server application's factories. This operation requires the following input arguments:

• The object reference for the factory

• A string identifier, based on the factory object's interface typecode, used to identify Interface Repository ID of the object's OMG IDL interface

The following example unregisters the RegistrarFactory factory used in the Basic sample application:

TP::unregister_factory(s_v_fact_ref.in(), UnivB::_tc_RegistrarFactory->id());

In the preceding code example, notice the use of the global variable s_v_fact_ref. This variable was set in the Server::initialize() operation that registered the RegistrarFactory object, which is used again here.

Notice also the parameter UnivB::_tc_RegistrarFactory->id(). This is also the same as the interface name used to register the factory.

Step 4: Define the in-memory behavior of objects.

As stated in the section "Managing Object State" on page 1-10, you determine what events cause an object to be deactivated by assigning object activation policies, transaction policies, and, optionally, using the application-controlled deactivation feature.

You specify object activation and transaction policies in the ICF file, and you implement application-controlled deactivation via the TP::deactivateEnable() operation. This section explains how you implement both mechanisms, using the Basic University sample application as an example.

The sections that follow describe the following:

• How to specify object activation and transaction policies in the ICF file

• How to implement application-controlled deactivation

Specifying Object Activation and Transaction Policies in the ICF File

The WebLogic Enterprise software supports the following activation policies, described in "Object Activation Policies" on page 1-12:

The WebLogic Enterprise software also supports the following transaction policies, described in Chapter 5, "Integrating Transactions into a WebLogic Enterprise Server Application":

To assign these policies to the objects in your application:

1. Generate the ICF file by entering the genicf command, specifying your application's OMG IDL file as input, as in the following example:

   # genicf university.idl

   The preceding command generates the file university.icf.

2. Edit the ICF file and specify the activation policies for each of your application's interfaces. The following example shows the ICF file generated for the Basic University sample application. Notice that the default object activation policy is method, and that the default transaction activation policy is optional.

   module POA_UniversityB
       {
       implementation CourseSynopsisEnumerator_i
           {
           activation_policy ( method );
           transaction_policy ( optional );
           implements ( UniversityB::CourseSynopsisEnumerator );
           };
       };
   module POA_UniversityB
       {
       implementation Registrar_i
           {
           activation_policy ( method );
           transaction_policy ( optional );
           implements ( UniversityB::Registrar );
           };
       };
   module POA_UniversityB
       {
       implementation RegistrarFactory_i
           {
           activation_policy ( method );
           transaction_policy ( optional );
           implements ( UniversityB::RegistrarFactory );
           };
       };

3. If you want to limit the number of interfaces for which you want skeleton and implementation files generated, you can remove from the ICF file the implementation blocks that implement those interfaces. Using the preceding ICF code as an example, to prevent skeleton and implementation files from being generated for the RegistrarFactory interface, remove the following lines:

   implementation RegistrarFactory_i
           {
           activation_policy ( method );
           transaction_policy ( optional );
           implements ( UniversityB::RegistrarFactory );
           };

4. Pass the ICF file to the IDL compiler to generate the skeleton and implementation files that correspond to the specified policies. For more information, see the section "Generating the Skeleton and Implementation Files" on page 2-5.

After you have finished writing the code for the Server object and the object implementations, you compile and link the server application.

You use the buildobjserver command to compile and link WebLogic Enterprise server applications. The buildobjserver command has the following format:

buildobjserver [-o servername] [options]

In the buildobjserver command syntax:

• -o servername represents the name of the server application to be generated by this command.

• options represents the command line options to the buildobjserver command.

For complete information about compiling and linking the University sample applications, see the Guide to the University Sample Applications. For complete details about the buildobjserver command, see the C++ Programming Reference.

Step 6: Deploy the server application.

You or the system administrator deploy the WebLogic Enterprise server application by using the procedure summarized in this section. For complete details on building and deploying the University sample applications, see the Guide to the University Sample Applications.

To deploy the server application:

1. Place the server application executable file in an appropriate directory on a machine that is part of the intended WebLogic Enterprise domain.

2. Create the application's configuration file, also known as the UBBCONFIG file, in a common text editor.

3. Set the following environment variables on the machine from which you are booting the WebLogic Enterprise server application:

   • TUXCONFIG, which needs to match exactly the TUXCONFIG entry in the UBBCONFIG file. This variable represents the location or path of the application's UBBCONFIG file.

   • APPDIR, which represents the directory in which the application's executable file exists.

4. Set the TUXDIR environment variable on all machines that are running in the WebLogic Enterprise domain or that are connected to the WebLogic Enterprise domain. This environment variable points to the location where the WebLogic Enterprise software is installed.

5. Enter the following command to create the TUXCONFIG file:

   tmloadcf -y application-ubbconfig-file

   The command-line argument application-ubbconfig-file represents the name of your application's UBBCONFIG file. Note that you may need to remove any old TUXCONFIG files to execute this command.

6. Enter the following command to start the WebLogic Enterprise server application:

   tmboot -y

   You can reboot a server application without reloading the UBBCONFIG file.

For complete details about configuring the University sample applications, see the Guide to the University Sample Applications. For complete details on creating the UBBCONFIG file for WebLogic Enterprise applications, see the Administration Guide.

Development and Debugging Tips

The following topics are discussed in this section:

• Use of CORBA and M3 exceptions and the user log

• Detecting error conditions in the callback methods

• Common pitfalls of OMG IDL interface versioning and modification

• Caveat for state handling in the Tobj_ServantBase::deactivate_object() operation

Use of CORBA and M3 Exceptions and the User Log

This section discusses the following topics:

• The client application view of exceptions

• The server application view of exceptions

Client Application View of Exceptions

When a client application invokes an operation on a CORBA object, an exception may be returned as a result of the invocation. The only valid exceptions that can be returned to a client application are the following:

• Standard CORBA-defined exceptions that are known to every CORBA-compliant ORB

• Exceptions that are defined in OMG IDL and known to the client application via either its stub or the Interface Repository

The WebLogic Enterprise system works to ensure that these CORBA-defined restrictions are not violated, which is described in the section "Server Application View of Exceptions" on page 2-20.

Because the set of exceptions exposed to the client application is limited, client applications may occasionally catch exceptions for which the cause is ambiguous. Whenever possible, the WebLogic Enterprise system supplements such exceptions with descriptive messages in the user log, which serves as an aid in detecting and debugging error conditions. These cases are described in the following section.

Server Application View of Exceptions

This section presents the following topics:

• Exceptions raised by the WebLogic Enterprise system that can be caught by application code

• The M3 system's handling of exceptions raised by application code during the invocation of operations on CORBA objects

Exceptions Raised by the WebLogic Enterprise System that Can Be Caught by Application Code

The WebLogic Enterprise system may return the following types of exceptions to an application when operations on the TP object are invoked:

• CORBA-defined system exceptions

• CORBA UserExceptions defined in the file TobjS_c.h. The OMG IDL for the exceptions defined in this file is the following:

  interface TobjS { 
            exception AlreadyRegistered { };
            exception ActivateObjectFailed { string reason; };
            exception ApplicationProblem { };
            exception CannotProceed { };
            exception CreateServantFailed { string reason; };
            exception DeactivateObjectFailed { string reason; };
            exception IllegalInterface { };
            exception IllegalOperation { };
            exception InitializeFailed { string reason; };
            exception InvalidDomain { };
            exception InvalidInterface { };
            exception InvalidName { };
            exception InvalidObject { };
            exception InvalidObjectId { };
            exception InvalidServant { };
            exception NilObject { string reason; };
            exception NoSuchElement { };
            exception NotFound { };
            exception OrbProblem { };
            exception OutOfMemory { };
            exception OverFlow { };
            exception RegistrarNotAvailable { };
            exception ReleaseFailed { string reason; };
            exception TpfProblem { };
            exception UnknownInterface { };
  }

The M3 System's Handling of Exceptions Raised by Application Code during the Invocation of Operations on CORBA Objects

A server application can raise exceptions in the following places in the course of servicing a client invocation:

• In the Server::create_servant, Tobj_ServantBase::activate_object(), and Tobj_ServantBase::deactivate_object() callback methods.

• In the implementation code for the invoked operation.

It is possible for the server application to raise any of the following types of exceptions:

• A CORBA-defined system exception

• A CORBA user-defined exception defined in OMG IDL

• A CORBA user-defined exception defined in the file TobjS_c.h. The following exceptions are intended to be used in server applications to help the WebLogic Enterprise system send messages to the user log, which can help with troubleshooting:

  interface TobjS { 
            exception ActivateObjectFailed { string reason; };
            exception CreateServantFailed { string reason; };
            exception DeactivateObjectFailed { string reason; };
            exception InitializeFailed { string reason; };
            exception ReleaseFailed { string reason; };
  }

• Any other C++ exception type

All exceptions raised by server application code that are not caught by the server application are caught by the WebLogic Enterprise system. When these exceptions are caught, one of the following occurs:

• The exception is returned to the client application without alteration.

• The exception is converted to a standard CORBA exception, which is then returned to the client application.

• The exception is converted to a standard CORBA exception, and the following actions occur:

  • The exception is returned to the client application

  • One or more messages containing descriptive information about the error are sent to the user log. The descriptive information may originate from either the server application code or from the WebLogic Enterprise system.

The following sections show how the M3 system handles exceptions raised by the server application during the course of a client invocation on a CORBA object.

Exceptions raised in the Server::create_servant() operation

If any exception is raised in the Server::create_servant() operation, then:

• The CORBA::OBJECT_NOT_EXIST exception is returned to the client application.

• If the exception raised is TobjS::CreateServantFailed, then a message is sent to the user log. If a reason string is supplied in the constructor for the exception, then the reason string is also written as part of the message.

• Neither the Tobj_ServantBase::activate_object() or Tobj_ServantBase::deactivate_object() operations are invoked. The operation requested by the client is not invoked.

Exceptions raised in the Tobj_ServantBase::activate_object() operation

If any exception is raised in the Tobj_ServantBase::activate_object() operation, then:

• The CORBA::OBJECT_NOT_EXIST exception is returned to the client application.

• If the exception raised is TobjS::ActivateObjectFailed, a message is sent to the user log. If a reason string is supplied in the constructor for the exception, the reason string is also written as part of the message.

• Neither the operation requested by the client nor the Tobj_ServantBase::deactivate_object() operation is invoked.

Exceptions raised in operation implementations

The WebLogic Enterprise system requires operation implementations to throw either CORBA system exceptions, or user-defined exceptions defined in OMG IDL that are known to the client application. If these types of exceptions are thrown by operation implementations, then the WebLogic Enterprise system returns them to the client application, unless one of the following conditions exists:

• The object has the always transaction policy, and the WebLogic Enterprise system automatically started a transaction when the object was invoked. In this case, the transaction is automatically rolled back by the WebLogic Enterprise system. Because the client application is unaware of the transaction, the WebLogic Enterprise system then raises the CORBA::OBJ_ADAPTER CORBA system exception, and not the CORBA:: TRANSACTION_ROLLEDBACK exception, which would have been the case had the client initiated the transaction.

• The exception is defined in the file TobjS_c.h. In this case, the exception is converted to the CORBA::BAD_OPERATION exception and is returned to the client application. In addition, the following message is sent to the user log:

  "WARN: Application didn't catch TobjS exception. TP Framework throwing CORBA::BAD_OPERATION." 

  If the exception is TobjS::IllegalOperation, the following supplementary message is written to warn the developer of a possible coding error in the application:

  "WARN: Application called TP::deactivateEnable() illegally and didn't catch TobjS exception." 

  This can occur if the TP::deactivateEnable() operation is invoked inside an object that has the transaction activation policy. (Application-controlled deactivation is not supported for transaction-bound objects)

• The WebLogic Enterprise system raised an internal system exception following the client invocation. In this case, the CORBA::INTERNAL exception is returned to the client. This usually indicates serious system problems with the process in which the object is active.

Exceptions raised in the Tobj_ServantBase::deactivate_object() operation

If any exception is raised in the Tobj_ServantBase::deactivate_object() operation, the following occurs:

• The exception is not returned to the client application.

• If the exception raised is TobjS::DectivateObjectFailed, a message is sent to the user log. If a reason string is supplied in the constructor for the exception, the reason string is also written as part of the message.

• A message is sent to the user log for exceptions other than the TobjS::DeactivateObjectFailed exception, indicating the type of exception caught by the WebLogic Enterprise system.

Detecting Error Conditions in the Callback Methods

The WebLogic Enterprise system provides a set of predefined exceptions that allow you to specify message strings that the TP Framework writes to the user log if application code gets an error in any of the following callback methods:

• Tobj_ServantBase::activate_object()

• Tobj_ServantBase::deactivate_object()

• Server::create_servant()

• Server::initialize()

• Server::release()

You can use these exceptions as a useful debugging aid that allows you to send unambiguous information about why an exception is being raised. Note that the TP Framework writes these messages to the user log only. They are not returned to the client application.

You specify these messages with the following exceptions, which have an optional reason string:

To send a message string to the user log, specify the string in the exception, as in the following example:

throw CreateServantFailed("Unknown interface");

Note that when you throw these exceptions, the reason string parameter is required. If you do not want to specify a string with one of these exceptions, you must use the double quote characters, as in the following example:

throw ActivateObjectFailed("");

The Server object's implementation of the Server::create_servant() operation instantiates an object based on its interface ID. It is crucial that this interface ID is the same as the one supplied in the factory when the factory invokes the TP::create_object_reference() operation. If the interface IDs do not match, the Server::create_servant() operation usually raises an exception or returns a NULL servant. The WebLogic Enterprise system then returns a CORBA::OBJECT_NOT_EXIST exception to the client application.(The WebLogic Enterprise system does not perform any validation of interface IDs in the TP::create_object_reference() operation.)

It is possible for this condition to arise if, during the course of development, different versions of the interface are being developed or many modifications are being made to IDL file. Even if you typically specify string constants for interface IDs in OMG IDL and use these in the factory and the Server::create_servant() operation, it is possible for a mismatch to occur if the object implementation and factory are in different executables. This potential problem may be difficult to diagnose.

You may want to consider the following defensive programming strategies during development to avoid this potential problem. This code should be included only in debugging versions of your application, because it introduces performance inefficiencies that may be unacceptable in the production versions of your software.

• Immediately before factory invokes the TP::create_object_reference() operation, include code that checks the Interface Repository to see if the required interface exists. (Make sure that all the application OMG IDL is up-to-date and loaded into the Interface Repository). Should this check fail to find the interface ID, you can assume that there is a mismatch.

• Following the invocation of the TP::create_object_reference() operation in your factories, include code that "pings" the object. That is, the code invokes any operation on the object (typically an operation that does not do anything). If this invocation raises the CORBA::OBJECT_NOT_EXIST exception, an interface ID mismatch exists. Note that "pinging" an object causes the object to be activated, with the overhead associated with the activation.

Caveat for State Handling in Tobj_ServantBase::deactivate_object()

The Tobj_ServantBase::deactivate_object() operation is invoked when the activation boundary for an object is reached. You may, optionally, write durable state to disk in the implementation of this operation. It is important to understand that exceptions raised in this operation are not returned to the client application. The client application will be unaware of any error conditions raised in this operation unless the object is participating in a transaction. Therefore, in cases where it is important that the client application know whether the writing of state via this operation is successful, we recommend that transactions be used.

If you decide to use the Tobj_ServantBase::deactivate_object() operation for writing state, and the client application needs to know the outcome of the write operations, we recommend that you do the following:

• Ensure that each operation that affects object state is invoked within a transaction, and that deactivation occurs within the transaction boundaries. This can be done by using either the method or transaction activation policies, and is possible with the process activation policy if the TP::deactivateEnable() operation is invoked within the transaction boundary.

• If an error occurs during the writing of object state, invoke the COSTransactions::Current::rollback_only() operation to ensure that the transaction is rolled back. This ensures that the client application receives one of the following exceptions:

  • If the client application initiated the transaction, the client application receives the CORBA::TRANSACTION_ROLLEDBACK exception.

  • If the WebLogic Enterprise system initiated the transaction, the client application receives the CORBA::OBJ_ADAPTER exception.

If transactions are not used, we recommend that you write object state within the scope of individual operations on the object, rather than via the Tobj_ServantBase::deactivate_object() operation. This way, if an error occurs, the operation can raise an exception that is returned to the client application.

Servant Pooling

As mentioned in the section "Servant Pooling and Stateless Objects" on page 1-19, servant pooling provides a means to reduce the cost of object instantiation for method-bound or transaction-bound objects.

How Servant Pooling Works

Normally, during object deactivation (that is, when the TP Framework invokes the Tobj_ServantBase::deactivate_object() operation), the TP Framework deletes the object's servant; however, when servant pooling is used, the TP Framework does not delete the servant at object deactivation. Instead, the server application maintains a pointer to the servant in a pool. When a subsequent client request arrives that can be satisfied by a servant in that pool, the server application reuses the servant and assigns a new object ID. When a servant is reused from a pool, the TP Framework does not create a new servant.

How You Implement Servant Pooling

You implement servant pooling by doing the following:

1. In the Server::initialize() operation on the Server object, write the code that sets up the servant pool. The pool consists of a set of pointers to one or more servants, and the code for the pool specifies how many servants for a given class are to be maintained in the pool.

2. In the pooled servant's Tobj_ServantBase::deactive_object() operation, you implement the TP::application_responsibility() operation. In the implementation of the TP::application_responsibility() operation, you provide code that places a pointer to the servant into the servant pool at the time that the TP Framework invokes the Tobj_ServantBase::deactivate_object() operation.

3. In the Server object's implementation of the Server::create_servant() operation, write code that does the following when a client request arrives:

   1. Checks the pool to see if there is a servant that can satisfy the request.

   2. If a servant does not exist, create a servant and invoke the Tobj_ServantBase::activate_object() operation on it.

   3. If a servant exists, invoke the Tobj_ServantBase::activate_object() operation on it, assigning the object ID contained in the client request.

For information about the TP::application_responsibility() operation, see the C++ Programming Reference.

Delegation-based Interface Implementation

There are two primary ways in which an object can be implemented in a WebLogic Enterprise application: by inheritance, or by delegation. When an object inherits from the POA skeleton class, and is thus a CORBA object, that object is said to be implemented by inheritance.

However, there may be instances in which you want to use a C++ object in a WebLogic Enterprise application in which inheriting from the POA skeleton class is difficult or impractical. For example, you might have a C++ object that would require a major rewrite to inherit from the POA skeleton class. You can bring this non-CORBA object into a WLE application by creating a tie class for the object. The tie class inherits from the POA skeleton class, and the tie class contains one or more operations that delegate to the legacy class for the implementation of those operations. The legacy class is thereby implemented in the WebLogic Enterprise application by delegation.

About Tie Classes in the WebLogic Enterprise System

To create a delegation-based interface implementation, use the -T command-line option of the IDL compiler to generate tie class templates for each interface defined in the OMG IDL file.

Using tie classes in a WebLogic Enterprise application also affects how you implement the Server::create_servant() operation in the Server object. The following sections explain the use of tie classes in WebLogic Enterprise in more detail, and also explains how to implement the Server::create_servant() operation to instantiate those classes.

In WebLogic Enterprise, the tie class is the servant, and, therefore, serves basically as a wrapper object for the legacy class.

The following figure shows the inheritance characteristics of the interface Account, which serves as a wrapper for a legacy object. The legacy object contains the implementation of the operation op1. The tie class delegates op1 to the legacy class.

Tie classes are transparent to the client application. To the client application, the tie class appears to be a complete implementation of the object that the client application invokes. The tie class delegates all operations to the legacy class, which you provide. In addition, the tie class contains the following:

• Constructor and destructor code, which handles startup and shutdown procedures for the tie class and the legacy class

• Housekeeping code, which implements operations such as accessors

When to Use Tie Classes

Tie classes are not unique to WebLogic Enterprise, and they are not the only way to implement delegation in a WebLogic Enterprise application. However, the WebLogic Enterprise convenience features for tie classes can greatly reduce the amount of coding you need to do for the basic constructor, destructor, and housekeeping operations for those tie classes.

Using tie classes might be recommended in one of the following situations:

• You want to implement an object in a WebLogic Enterprise application in which inheriting from the POA skeleton class is difficult or impractical.

• All the invocations on a legacy class instance can be accomplished from a single servant.

• You are using a legacy class in your WebLogic Enterprise application, and you want to tie the lifetime of an instance of that legacy class to a servant class.

• Delegation is the only purpose of a particular servant; therefore, nearly all the code in that servant is dedicated to legacy object startup, shutdown, access, and delegation.

Tie classes are not recommended when:

• The operations on an object instance delegate to more than one legacy object instance.

• Delegation is only a part of the purpose of an object.

How to Create Tie Classes in a WebLogic Enterprise Application

To create tie classes in an application in a WebLogic Enterprise domain:

1. Create the interface definition for the tie class in an OMG IDL file, as you would for any object in your application.

2. Compile the OMG IDL file using the -T option.

   The IDL compiler generates a C++ template class, which takes the name of the skeleton, with the string _tie appended to it. The IDL compiler adds this template class to the skeleton header file.

   Note that the IDL compiler does not generate the implementation file for the tie class; you need to create this file by hand, as described in the next step.

3. Create an implementation file for the tie class. The implementation file contains the code that delegates its operations to the legacy class.

4. In the Server object's Server::create_servant() operation, write the code that instantiates the legacy object.

   In the following example, the servant for tie class POA_Account_tie is created, and the legacy class LegacyAccount is instantiated.

   Account * Account_ptr = new LegacyAccount();
   AccountFactoryServant = new POA_Account_tie<LegacyAccount> (Account_ptr)

   Note: When compiling tie classes with the Digital C++ V6.0 compiler for Tru64 UNIX, you must include the -noimplicit_include option in the definition of the CFLAGS or CPPFLAGS environment variables used by the buildobjserver command. This option prevents the Digital C++ compiler from automatically including the server skeleton definition file (_s.cpp) everywhere the server skeleton header file (_s.h) is included, which is necessary to avoid multiply-defined symbol errors. See the publication Using DIGITAL C++ for Digital Unix Systems for additional information about using class templates, such as the tie classes, with Digital C++.