Copyright Tristan Aubrey-Jones May 2008.
Abstract: A project investigating and developing an implicitly concurrent programming language, based on a metaphor taken from the physical world is reported. Uses a programming paradigm where programs consist of systems of autonomous agents, or active objects which communicate via message passing. A language enhancing Java with actors and linear types is presented. Example programs are written, compiled, and executed to evaluate the usefulness of the language. The language found to provide a familiar notation for implicit parallelism, and a compelling new model for concurrency, combining the performance of shared variables with the elegance of message passing.
Introductory Slides (PDF),
Report (PDF),
ActiveJava compiler prototype (ajavac),
ActiveJava runtime library (ajava_lang).
Examples:
calc - pocket calculator actor program dining - dining philosophers actor program (never deadlocks) sort - parallel quicksort implementation ("SortBenchmark" sorts 10,000 random integers using actors, java threads, and sequentially and compares)To compile examples use:
compile.bat ./calc compile.bat ./sort compile.bat ./diningTo run examples use:
run ./calc Main run ./dining Main run ./dining Main fast run ./sort Main run ./sort SortingBenchmark
import org.taj.ajava.lang.*;
public class ALU extends Actor
{
private double register;
private char operator = '=';
public final Event OnResult = new Event();
public ALU()
{
register = 0.0;
}
private void react_0(Calculator.Operation op)
{
System.out.print(op.operand);
System.out.print(op.operator);
switch (operator) {
case '+':
{
register += op.operand;
break;
}
case '-':
{
register -= op.operand;
break;
}
case '*':
{
register *= op.operand;
break;
}
case '/':
{
register /= op.operand;
break;
}
case '^':
{
register = Math.pow(register, op.operand);
break;
}
case '=':
{
register = op.operand;
break;
}
}
operator = op.operator;
System.out.println(" returns " + Double.toString(register));
OnResult.deliver(register);
}
public void deliver(Calculator.Operation op)
{
bufferMessage(new org.taj.ajava.runtime.ActorMessage(op, 0));
}
protected void react(Calculator.Operation op)
{
react_0(op);
}
protected void processMessage(org.taj.ajava.runtime.ActorMessage msg)
{
switch (msg.reactorId) {
case 0:
{
react_0(((Calculator.Operation)msg.payload));
return;
}
default:
{
super.processMessage(msg);
return;
}
}
}
}