Key Properties for Open Implementations

The principled separation between base- and meta-interfaces is the primary principle of the reflective modules we propose for open implementations. But there is a great deal more than this to properly opening implementations in this way.

In some sense, traditional black-box abstractions have always had a crude meta-interface --- if client programmers did not like mapping decisions made inside a module implementation, they could write a new one entirely from scratch!

To capture the key properties of a well-designed open implementation, the following additional concepts are important:

• Incrementality - is the degree to which a client's desire to change a small part of a module's implementation requires writing a correspondingly small meta-program.

• Scope Control - Is the degree to which the effect of changes made using the meta-interface can be restricted to affect only some entities in (or parts of) the base program.

• Interoperability - Is the degree to which parts of the base-program that use non-standard mapping decisions can freely interact with other parts of the base program.

• Reliability - Is the probability of failure-free operation of the running system - including the client base- and meta-programs - for a specified time in a specified environment.

  Note that a client of a properly designed OI can often be more reliable than the client of a black-box module approach, because the small meta-programs reduce the occurrence of large failure-prone hematomas that may otherwise be present.

• Efficiency - Is the performance of the entire running system --- including the client base- and meta-programs --- as compared to the entire running system using a black-box module.

Return to the Open Implementation Site Guide Return to the Open Implementation Home Page