Areas of work


Study groups



Domain-dependent interpretation


A domain in Ptolemy II implements a model of computation, which effectively defines the interaction between a collection of actors. It does so be controlling the exchange of tokens among actors as well as their scheduling, i.e. it determines when each actors fires. In order to do their job, domains often employ special "contracts" in addition to the generic Ptolemy II actor/director/receiver contracts. These may range from special attributes (such as the token rates for SDF actors) describing specific actor properties of interest to the domain to the use of sophisticated specialized infrastructure (such as the one used for reading and writing tokens in CSP).

Actors that are to be used in such a domain therefore need to (a) be aware of its existence and (b) need to be coded to its particular assumption. This tends to tie in actors to a specific domain, and also introduces dependencies between domain code and actor code, contributing to the brittleness of the system.

Domain-dependent interpretation of CAL actors tries to address these issues. The basic idea is that there is a variety of domain-specific CAL interpreters, which interpret the same CAL description slightly differently, making use of domain-specific interaction mechanisms, while keeping the CAL description itself agnostic about the domain or the way it is implemented. This provides the following benefits:

  1. Actor descriptions become more domain-polymorphic, and thus reusable, while still making use of domain-specific mechanisms and features.
  2. Actor descriptions are decoupled from domain implementations---changes in the way domains are implemented do not affect actors, once the domain-dependent interpreter has been updated.
  3. Actor writers do not need any specific knowledge about the implementation of domains. All they need to know is how actor features are interpreted in the context of a domain.

Domain-dependent interpreters

A domain-dependent interpreter (DDI) has to perform the following tasks:

  1. When given an actor, it has to determine whether that actor is well-formed, i.e. whether the actor can be legally used inside the domain. The well-formedness of an actor might depend on the context it is used in, in particular it might depend on the values given to its parameters, but also, in principle, on the way it is related to other actors in the same model.
  2. For any well-formed actor, the DDI publishes the static properties that are relevant to the domain in a way that they can be seen and used by the director. This usually involves analyzing the actor and annotating it with attributes that are understood by the director and that reflect the analysis results.
  3. For any well-formed actor, the DDI produces the dynamic behavior of the actor by interacting with the domain according to the generic Actor/Director API as well as any relevant domain-specific constructions.

The current implementation inside Ptolemy uses a generic typed atomic actor (ptolemy.caltrop.actors.CalInterpreter). It detects the model of computation it is running inside of and then instantiates a domain-dependent plugin that contains the DDI. This architecture allows CAL actors inside a model to adapt when the designer changes the domain of the model without further user intervention.


You are not logged in 
©2002-2017 U.C. Regents