Friday 24 August 2001

1
Translation of Java to Real-Time Java Using
Aspects
Nicholas Leidenfrost
Morgan Deters
Ron K. Cytron
Department of Computer Science Washington
University in St. Louis
Presented at the Workshop on Aspect-Oriented Progr
amming and Separation of Concerns Lancaster
University, UK
Friday 24 August 2001
Research sponsored by National Science
Foundation (0081214) DARPA (F33615-00-C-1697)
2
Overview

• Briefing on the Real-Time Java Memory Model
• Dynamic Analysis with Aspects
• RT-Java Transformation with Aspects
• Concluding Remarks

3
Real-Time Java

• Real-Time Specification for Java (RTSJ) brings
  infrastructure for real-time accountability to
  Java
• Real-time schedulability
• High-precision timers
• Asynchronous transfer of control
• Physical memory address access
• Choice of memory allocators and garbage
  collectors
• Memory reclamation without garbage collection

4
Real-Time Java Memory Model
MemoryArea
ImmortalMemory
ScopeMemory
HeapMemory
LTMemory
VTMemory
5
ScopeMemory

• A ScopeMemory area is tied to a particular scope
  of execution at runtime
• Programmer specification of scoped memory can
  introduce application-specific code into
  application classes, defeating their reuse

class C void foo() / call bar() /
bar() void bar() / generate
garbage /
class C implements Runnable void foo()
/ call bar() / ScopeMemory lt new
LTMemory(1024,1024) lt.enter(this)
void bar() / generate garbage /
void run() bar()
6
Nested ScopeMemory References
Y
X
Z
A
G
D
B
F
E
C
ScopeMemory areas may be nested
Object references must point outward
A thread entering nested memory scopes
7
High-Level View
AspectJ
Real-Time Java
Java
P


AspectJ
8
Reference Probe

• Determine references between objects
• Track objects' references to other objects
• target(a) args(x) set( .)
• Build a doesReference graph
• Discover legal scoping hierarchies
• Take the reverse of the doesReference graph
• Collapse strongly connected components
• The resulting DAG is a representation of legal
  scoping hierarchies

9
Liveness Probe(Death Probe)

• Determine the points at which objects become
  collectible
• Track object birth times
• After each execution join point, run a garbage
  collection cycle and determine which objects were
  collected
• Express object liveness behavior in terms of
• Enclosing execution join points
• Distance from birth to death

10
Object Recognition

• How to map objects observed in previous analyses
  to objects observed in current program execution?
• Class-based behavior
• All instances of a class behave similarly
• "New site"-based behavior
• All instances of a class instantiated at the same
  source code location behave similarly
• Instance-based behavior
• Instances of a class behave differently

11
Adding an Enforcement Aspect

• We have
• A set of all legal ScopeMemory hierarchies
• Object lifetime behavior (join points)
• From this information, we
• Select a desirable, legal scope memory hierarchy
• Direct a dispatch unit to implement this hierarchy

12
Concluding Remarks

• We have demonstrated the use of aspects in
  dynamic analysis
• We are working toward
• A software analysis tool that can provide crucial
  memory trace information to the programmer
• Automation of scope detection and selection
• A Real-Time Java translation system, implemented
  in AspectJ, that requires no program annotation
  and transforms Java programs into
  ScopeMemory-aware RT-Java programs modularly

13
Thanks !
Morgan Deters mdeters_at_cs.wustl.edu Nicholas
Leidenfrost naleiden_at_cs.wustl.edu Ron K.
Cytron cytron_at_cs.wustl.edu www.cs.wustl.edu/
doc Department of Computer Science Washington
University Box 1045 St. Louis, MO 63130 USA