Designing Efficient Interfaces

There are some special considerations to keep in mind when designing a custom interface. If you follow these guidelines, your marshalling code can be used across the network when support for remotable objects is provided in a future release.

First of all, because data is shipped across address spaces, using architecture-dependent types such as int may prohibit the data from being re-created correctly in the object application's address space. The size of an integer varies from architecture to architecture. Thus, the data type int should not be used for structure data members or interface method parameters. Specifying data types unambiguously to the MIDL compiler will allow your interface to be more easily transferred to remote machines with different architectures and addressing schemes. The MIDL compiler requires all integer variables that may be remoted to be explicitly declared as short, long, or their unsigned equivalents.

A frequently used data type for pointers, the void * construct, allows the callee of a method to interpret the data pointed to according to the need. While local interfaces can use this construct, distributed applications cannot. This is because the MIDL compiler must know exactly what types of data are being transmitted so that the data can be accurately re-created on the receiving side. The void * construct is too vague.

Pointers to data must be used carefully. To re-create the data in the callee's address space, RPC must know the exact size of the data. If, for example, a char * parameter points to a buffer of characters rather than to a single character (as is implied), the data cannot be correctly re-created. Use the syntax available with MIDL to accurately describe the data structures represented by your type definitions.

Initialization is essential for pointers that are embedded in arrays and structures and passed across process boundaries. Uninitialized pointers may work when passed to a program in the same process space, but proxies and stubs assume that all pointers are initialized with valid addresses or are null.

Be careful when aliasing pointers (allowing pointers to point to the same piece of memory). If the aliasing is intentional, these pointers should be declared aliased in the IDL file. Pointers declared as non-aliased should never alias each other.

The MIDL has directional attributes that let you specify the data flow direction. The proxies and stubs use these attributes to determine whether to send the data from the client to the object or from the object to the client, respectively. Data labeled as out only is uninitialized on the way to the interface stub; data labeled as in only does not affect the data structures upon return. The in, out attribute indicates that data is sent to the object initialized and the object will change it before sending it back.

It is also imperative to realize the importance of defining status codes. This must be coordinated by the definer of the interface to avoid conflicting error codes.

Finally, the designer of an interface must develop an understanding of how the interface will be used by client applications. In particular, the frequency of method calls across the interface boundary and the amount of data to be transferred to complete a given method call together determine whether the interface will be efficient across process and machine boundaries. Although OLE makes cross-process and, eventually, cross-network calls transparent to programs, it cannot make high-frequency and high-bandwidth calls efficient across address spaces. In some cases, it is more appropriate to design interfaces that will normally be implemented only as in-process servers while other interfaces are more appropriate for remote use.

Software for developers
Delphi Components
.Net Components
Software for Android Developers
More information resources
MegaDetailed.Net
Unix Manual Pages
Delphi Examples