Extensibility

  An important characteristic of a language is extensibility. COP offers two forms of extensibility; more details on both are provided in Appendix A.

The simplest form of extensibility is for a user (or HLL compiler) to use Lisp primitives to define a new operator in terms of existing operators. The transpose operator is a (pre-defined) example of this; the operator is defined in Lisp to return a parallel group of appropriate streams.

A more substantive way to extend the language is also needed, however. Consider the following scenarios:

• An operator might require more sophisticated use of the routers than can be expressed by the existing implementations.
• An application might have an operator subset with a particular structure that an optimized implementation could take advantage of.
• An application might use a particular set of runtime values for an operator which an optimized schedule-generator could handle.
• Or, more generally, an operator may just not be as efficient if mapped down to streams.

To handle this, COP allows users to provide operator implementations which can take effect under the appropriate circumstances. Such operators are simply written in C (or any other language with a compatible application binary interface), then dynamically linked by the COP compiler at compile-time and chosen for use in exactly the same way that pre-defined implementations are chosen.

It is also possible to define entirely new fundamental operators, but it requires modification of the compiler to characterize the new operator type appropriately. Having done so once, implementations of the operator can be written as needed.