1. Master’s Thesis Defense: Aspectual Concepts John J. Sung

2. AspectJ from PARC AJ J EMETERJ (Demeter AspectJ) Aspect-oriented tools referred to in this talk

3. Static Scattering and Tangling aspect i is scattered across many classes (i = 1,2,3) class X tangles aspects 1, 2 and 3 aspect1aspect2aspect3 class A consisting of three aspects classes for aspect1 classes for aspect2 classes for aspect3 Class X class diagram Adding to classes

4. Dynamic Scattering and Tangling program execution involving three aspects (colors r b g) program call tree (objects executing method calls) Each aspect (colors) is scattered across many classes (shapes) Class tangles all three aspects At those calls the aspect enhances the behavior classes this(s) target(t) f(..) Enhancing calls t.f(..);

5. subcomputation = join points related to traversing through the objects guided by traversal specification and class graph.

6. Pointcut –set of execution points of any method, … –rich set of primitive pointcuts: this, target, call, … + set operations –where to enhance Advice –how to enhance Visitor method sig. –set of execution points of traversals –specialized for traversals (nodes, edges) –where to enhance Visitor method bodies –how to enhance DJAspectJ From DJ to AspectJ

7. Outline Motivation JPM and Demeter Concepts Demeter integrated with AspectJ (DAJ) DAJ Semantics/Syntax DAJ System Architecture Four Graph Model of Programs (4GMP) Conclusion

8. Motivation Aspect Oriented Software Development Gaining Momentum –1 st International Conference on AOSD –AspectJ, DemeterJ, ConcernJ, HyperJ, etc. AOSD is Relatively Young –DemeterJ ~1989 –AspectJ ~1997 Karl J. Lieberherr: Demeter C++ 1989 DemeterJ 1996 Karl J. Lieberherr: Demeter C++ 1989 DemeterJ 1996

9. Questions What are the fundamental concepts behind AOSD? How are these concepts related? Can these concepts be mixed? What are the fundamental problems that software language features attempt to solve? Karl J. Lieberherr: How can these concepts be combined? Karl J. Lieberherr: How can these concepts be combined?

10. Joinpoint Model (JPM) Fundamental Concepts –Joinpoint, Pointcut, Advice, Aspect, Introduction –Concepts applied to call graph in AspectJ –Building or Construction Metaphor Karl J. Lieberherr: Also to class graph : introductions Karl J. Lieberherr: Also to class graph : introductions

11. Construction Metaphor Concepts in JPM create an image of constructed structures from building blocks Karl J. Lieberherr: Of structures constructed from building … Karl J. Lieberherr: Of structures constructed from building …

12. AspectJ Abstraction of Programs Call GraphPointcuts ++ Advices => Program

13. AspectJ Abstraction of Programs foo(){ x = 0; if (x < 0) throw KException; } bar int x; b1.foo(); c1 within(bar) aspect before() : within(bar) { }

14. Another View Java Program AspectJ Advice pointcut advice body join point

15. Demeter Fundamental Concepts –Class Graph, Strategy, Visitor, Advice –Concepts applied to data structures, i.e. class graph –Journey Metaphor

16. Journey Metaphor Concepts in Demeter create an image of a person on a journey

17. Demeter Abstraction of Programs Class Graph Strategy Graph ++ Visitor Advices => Program

18. Demeter Abstraction of Programs => before(..) after(..) before(..) Class Graph Strategy Graph Traversal Graph Visitor Advices

19. Mixing Concepts Application of JPM on Demeter –Class graph is built of classes and edges –Joinpoints become the points along traversals –Advices are executed at these joinpoints

20. Mixing Concepts Application of Demeter on JPM –CPU is the global visitor –Advice with a pointcut are Demeter advices for the global visitor –Visitor is traversing the Dynamic Call Graph Karl J. Lieberherr: Advices with a … Karl J. Lieberherr: Advices with a … Karl J. Lieberherr: This continues to be non-convincing. Karl J. Lieberherr: This continues to be non-convincing.

21. Demeter AspectJ (DAJ) Uses AspectJ as the backend compiler Uses DJ to generate Class Graph, Traversal Graph Generates the AspectJ implementation of the traversal Defines a traversal specification language Implemented using DemeterJ Application of JPM on Demeter

22. Traversal Specification Language Designed to have similar syntax to DJ and AspectJ Allows users to specify Class Graph, Traversal, and Visitor Generates AspectJ implementation of traversals

23. Class Graph Specification Default Class Graph –ClassGraph cgvar; Class Graph Slice –ClassGraph cgvar = new ClassGraph(cg, “strategy”);

24. Visitor Specification Uses Java Reflection to obtain method signatures Recognized methods are –around, before, after, start, finish, returnValue Visitor Declaration –Visitor visitorClass;

25. Traversal Specification Default Traversal Specification –declare traversal tvar : “strategy”; Traversal with Class Graph –declare traversal tvar(cgvar) : “strategy”; Traversal with Visitor –declare traversal tvar(cgvar, visitorvar) : “strategy”;

26. Aspect Specification aspect aspectName { class graph declarations; traversal declarations; visitor declarations; }

27. What DAJ Generates For each default and class graph slice traversal –method void tvar() for the source node of the traversal strategy For each traversal with visitor –method void tvar() for the source node of the traversal strategy –method void tvar(visitorClass) for the source node of the traversal strategy –Appropriate AspectJ Advices for each advice in visitorClass

28. A Simple Basket Example class Basket { Basket(Fruit _f, Pencil _p) { f = _f; p = _p; } Basket(Fruit _f, Fruit _f2, Pencil _p) { f = _f; f2 = _f2; p = _p; } Fruit f, f2; Pencil p; } class Fruit { Fruit(Weight _w) { w = _w; } Weight w; } class Orange extends Fruit { Orange(Color _c) { super(null); c=_c;} Orange(Color _c, Weight _w) { super(_w); c = _c;} Color c; } class Pencil {} class Color { Color(String _s) { s = _s;} String s; } class Weight{ Weight(int _i) { i = _i;} int i; int get_i() { return i; } }

29. A Simple Basket Example Basket FruitPencil p f, f2 Orange Weight Color wc String s int i

30. BasketVisitor class BasketVisitor { int total; public void start() { total = 0; } public int returnValue() { return total; } void before(Weight w) { total += w.get_i(); }

31. Basket Traversal aspect BasketTraversal { ClassGraph default; ClassGraph myClassGraph = new ClassGraph(default, "from Basket to *"); Visitor BasketVisitor; declare traversal t1(myClassGraph,BasketVisitor) : "from Basket to Weight"; declare traversal t2(myClassGraph,BasketVisitor) : "from Basket via Orange to Weight"; }

32. Basket Main class BasketMain { static public void main(String args[]) throws Exception { Basket b = new Basket(new Orange(new Color("orange"), new Weight(5)), new Fruit( new Weight(10)), new Pencil() ); BasketVisitor bv = new BasketVisitor(); b.t1(bv); int totalWeight = bv.returnValue(); System.out.println("Total weight of basket = " + totalWeight); b.t2(bv); totalWeight = bv.returnValue(); System.out.println("Total weight2 of basket = " + totalWeight); } Karl J. Lieberherr: Visitor object generation not automatic. More flexibility this way. Karl J. Lieberherr: Visitor object generation not automatic. More flexibility this way.

33. Generated Code for Visitor public aspect BasketTraversal { static BasketVisitor t1_visitor; public void Basket.t1(BasketVisitor v) { t1_visitor=v; t1_visitor.start(); t1(); } before(Weight host) : call(public void t1*()) && target(host) { t1_visitor.before(host); } void Basket.t1() { t1_copy0(); } Karl J. Lieberherr: What is this? Karl J. Lieberherr: What is this?

34. Basket Class Graph Basket FruitPencil p f, f2 Orange Weight Color wc String s int i

35. Generated Code for Traversal // traversal t1 : {source: Basket -> target: Weight} with { } public void Basket.t1_copy0(){ if (f != null) t1_copy0_crossing_f(); if (f2 != null) t1_copy0_crossing_f2(); } public void Basket.t1_copy0_crossing_f() { f.t1_copy0();} public void Basket.t1_copy0_crossing_f2() { f2.t1_copy0();} public void Fruit.t1_copy0(){ if (w != null) t1_copy0_crossing_w(); } public void Fruit.t1_copy0_crossing_w() { w.t1_copy0();} public void Weight.t1_copy0(){ } public void Orange.t1_copy0(){ super.t1_copy0(); } pointcut pointcut_t1() : call(public void t1*()); before () : pointcut_t1 () { System.out.println(thisJoinPoint); }

36. System Architecture DAJ Main –Parses command line arguments –Manages the DAJ code generation phases Stub Generation –Generates stubs for traversal methods Traversal Generation Compilation –Compiles using ajc the stubs, user code, and CreateClassGraph.java Karl J. Lieberherr: How are stubs different from traversals? Karl J. Lieberherr: How are stubs different from traversals?

37. System Architecture Traversal Generation –Uses DJ, Java Reflection, and DemeterJ to generate AspectJ traversal code Traversal Compilation –Compiles the generated traversal code with user code

38. System Architecture Stub Generation Traversal Generation Compilation Traversal Generation Traversal Compilation DAJ Main method call shell Karl J. Lieberherr: Do not mention shell scripts Karl J. Lieberherr: Do not mention shell scripts

39. DAJ Process Stub Generation Traversal Files Generated Stubs Traversal Generation Compilation Class Files Traversal Generation Traversal Compilation User Code CreateClassGraph.java Traversal Implementation Karl J. Lieberherr: CreateClassGraph.java is provided by DAJ Very confusing naming What information is generated by each box: input files and output files and who provides them Karl J. Lieberherr: CreateClassGraph.java is provided by DAJ Very confusing naming What information is generated by each box: input files and output files and who provides them

40. Future Improvements http://www.ccs.neu.edu/research/demeter/DAJ Two syntax supported –DJ/Java like style –AspectJ’s declare syntax Handle Collections –Traversal through Java Collections Handle interfaces –Traversal “correctly” for interfaces Karl J. Lieberherr: Correctly?? Karl J. Lieberherr: Correctly??

41. Four Graph Model of Programs Merging of graphs in Demeter and JPM Model for creating relationships between different features AOP feature analysis Only a proposal and needs lots of work

42. 4GMP Compile-TimeRuntime Data Structures Algorithm Program Class Graph Object Graph Static Call Graph Dynamic Call Graph Executable Process

43. Compile-Time vs. Run-Time Compile-time graphs specify all possible run-time graphs The computer system combines the compile-time graph with input to generate the run-time graph In a way, the compile-time graph has factored out the commonalities within the run-time graph Factorizational Concern

44. Input Specification Processor Output

45. Factorizational Concern +

46. Factorizational Concern Examples Functions that are used multiple times Inherited methods and data members Parser generation for DemeterJ Wildcards for pointcut designators in AspectJ Traversal generation in DAJ

47. Data Structure vs. Algorithm One of many ways of slicing up a program –data flow / control flow –Layering –Modules It does not change the amount of code that one has to write Large programs are a fact of life and it needs to be organized in some way Organizational Concern

49. Organizational Concern Examples Functions that are called only once Non-inherited methods and data members Open classes in DemeterJ Introductions in AspectJ

50. Factorizational/Organizational Concerns Most programming language features have both characteristics depending on their uses Both are needed to cope with the problem of programs growing ever larger

51. Concern Relationship Diagram We may approximate relationships between features and the graphs in 4GMP

52. Concern Relationship Diagram Class Graph Object Graph Static Call Graph Dynamic Call Graph FC OC Concern Relationship FC = Factorizational Concern OC = Organizational Concern

53. AspectJ CRD Class Graph Object Graph Static Call Graph Dynamic Call Graph FC OC FC = Factorizational Concern OC = Organizational Concern Introduction Pointcut Join Point Aspect Advice SP = Specification Feature SP FC OC SP FC

54. DemeterJ CRD Class Graph Object Graph Static Call Graph Dynamic Call Graph FC OC FC = Factorizational Concern OC = Organizational Concern FC Strategy Graph Class Dictionary Traversal Graph Visitor Advice OC FC OC FC

55. Ideal CRD Feature FC Feature FC Feature OC Feature OC

56. Ideal CRD Feature FC Feature FC Feature OC

57. What Good Are CRDs? CRD is a higher level feature map Allows designers to see imbalance of concerns Designers can fix these imbalances with appropriate features that address appropriate concerns Programming Feature Analysis!

58. DemeterJ Feature Analysis CRD has many FC relations Possibly add features with OC relations to Class Dictionary, Strategy Graph, Traversal Graph

59. DemeterJ CRD Class Graph Object Graph Static Call Graph Dynamic Call Graph FC OC FC = Factorizational Concern OC = Organizational Concern FC Strategy Graph Class Dictionary Traversal Graph Visitor Advice OC FC OC FC OC

60. AspectJ Feature Analysis CRD has many OC relations Possibly add features with FC relations to Aspect and Introductions

61. AspectJ CRD Class Graph Object Graph Static Call Graph Dynamic Call Graph FC OC FC = Factorizational Concern OC = Organizational Concern Introduction Pointcut Join Point Aspect Advice SP = Specification Feature SP FC OC SP FC

62. Problems With CRDs In what scope should one create CRDs? –Should AspectJ and DemeterJ CRDs account for features in Java? Most features have FC and OC –How do I deal with these features?

63. Other Types of Concerns Interface Concerns –When two entities within a program need to communicate –Abstract classes in OOP –Interfaces in Java Specification Concerns –Things that programmers want to specify or reference –Java reflections –AspectJ joinpoints

64. Questions What are the fundamental concepts behind AOSD in Demeter and AspectJ? How are these concepts related? Can these concepts be mixed? What are the fundamental problems that software language features attempt to solve?

65. Conclusions Metaphors used in Demeter and JPM –Construction and Journey Metaphors –Usage of one or the other depends on the users and application –Can describe each other –Mixable

66. Conclusions DAJ –Mixing Demeter with AspectJ concepts –Uses DJ, DemeterJ and AspectJ –www.ccs.neu.edu/research/demeter/DAJwww.ccs.neu.edu/research/demeter/DAJ

67. Conclusions Four Graph Model of Programs –Merges graphs in Demeter and AspectJ –Factorizational/Organizational Concerns –Concern Relationship Diagram –Programming Feature Analysis –Needs lots of work!

68. Future Direction Try mixing other tools –ComposeJ, HyperJ, etc. Feature analysis of other tools and concepts Investigate other types of concerns Relationship between concerns relationships in features with HCI and SE

69. References www.ccs.neu.edu/research/demeter/DAJ www.ccs.neu.edu/research/demeter/DemeterJava www.ccs.neu.edu/research/demeter/DJ www.aspectj.org