Tom Tourwé, ohan Brichau, Andy Kellens and Kris Gybels: Induced Intentional Software Views (ESUG 2003, Bled)

1. ESUG 2003
Induced Intentional Software Views
Tom Tourwé Johan Brichau
Andy Kellens and Kris Gybels
Programming Technology Lab
Department of Computer Science
Vrije Universiteit Brussel
1

2. Problems with Software Documentation
ÿ Software becomes very large, more complex and constantly
evolves
ÿ Software documentation is extremely important to cope
with these issues
ÿ avoid design degradation
ÿ understand inner workings
ÿ implement correct behaviour
ÿ Documentation is often non existent or outdated
ÿ not active part of development process
ÿ documentation and implementation are separated
ÿ not robust w.r.t. evolution
August 2003 2
© PROG, VUB

3. Software Views
Software View
ÿ Documentation technique used to highlight important
design structures
ÿ design patterns, framework hotspots, collaborations, ...
ÿ Collection of source code artifacts
ÿ classes, methods, variables, ...
ÿ Two different kinds of software views
ÿ Extensional views
ÿ Intentional views
August 2003 3
© PROG, VUB

4. Extensional Views
View name
Drag & Drop
View artifacts
August 2003 4
© PROG, VUB

5. Extensional Views
ÿ Manual enumeration of source code artifacts
ÿ Advantages
ÿ easy to define (drag & drop)
ÿ Disadvantages
ÿ not robust w.r.t. evolution
ÿ not scalable
ÿ not intention revealing
Reduces interest of using software views
August 2003 5
© PROG, VUB

6. Intentional Views
August 2003 6
© PROG, VUB

7. Intentional Views
ÿ Defined by means of an intentional description
ÿ executable expression in a programming language
ÿ view’s content is computed from the source code
ÿ Advantages
ÿ robust w.r.t. evolution
ÿ scalable
ÿ intention revealing Complicates use of Intentional Views
ÿ Disadvantages
ÿ hard to define (requires meta-programming skills)
ÿ risk to be overly general
ÿ requires detailed knowledge of the application’s internal structure
August 2003 7
© PROG, VUB

8. Induced Intentional Views
ÿ Combines advantages of extensional and intentional views
ÿ ease of use of extensional views
ÿ robustness and scalability of intentional views
ÿ Inducing views
ÿ manually classify source code artifacts
ÿ automatically derive intention behind it
ÿ Techniques
ÿ Logic Metaprogramming
ÿ to connect views to implementation
ÿ Inductive Logic Programming (Machine Learning)
ÿ to derive intention automatically
August 2003 8
© PROG, VUB

9. Logic Meta Programming (LMP)
ÿ Using a logic programming language (Prolog) at the meta
level to reason about and manipulate programs at the base
level (in Smalltalk)
ÿ Allows to define intentional views in a concise and
declarative manner
ÿ SOUL
ÿ Interpreter integrated in VW
ÿ Contains extensive library of logic predicates that consult source
code
August 2003 9
© PROG, VUB

10. Inductive Logic Programming
ÿ Machine learning technique
ÿ Discovers a general pattern underlying a number of examples
ÿ Requires a set of examples and a background theory
Background
Examples Induced Logic Rules
Theory
grandFather(tom,bob). father(tom,peter). grandFather(?grandfather,?person) if
grandFather(tom,jim). father(tom,marie). father(?grandfather,?father),
grandFather(tom,ellen). father(peter,bob). father(?father,?person).
grandFather(tom,bart). father(peter,jim). grandFather(?grandfather,?person) if
mother(marie,ellen). father(?grandfather,?mother),
mother(marie,bart). mother(?mother,?person).
August 2003 10
© PROG, VUB

11. Software Views with LMP
ÿ Extensional
ÿ Logic facts (enumeration)
ÿ class(ScExpression).
class(ScConsExpression).
ÿ methodInClass(analyse,ScConsExpression).
...
ÿ
ÿ Intentional
ÿ Logic rules (program)
Induction algorithm
class(?method) if
methodInClass(analyse,?method).
...
August 2003 11
© PROG, VUB

12. Proof of concept experiment
ScExpression
newClosure
analyse
ScSequence ScCons
Expression Expression SpecialForm
Handler
newClosure newClosure
analyse analyse newClosure
handle:
DefineHandler IfHandler
newClosure newClosure
handle: handle:
August 2003 12
© PROG, VUB

13. Classified Items
analyser(classImplementsMethodNamed(ScExpression,analyse)).
analyser(classImplementsMethodNamed(ScConsExpression,analyse)).
analyser(classImplementsMethodNamed(ScSequenceExpression,analyse)).
...
analyser(classImplementsMethodNamed(SpecialFormHandler,handle:)).
analyser(classImplementsMethodNamed(DefineHandler,handle:)).
analyser(classImplementsMethodNamed(IfHandler,handle:)).
August 2003 13
© PROG, VUB

14. Derived Rules
defines intention
intention(analyser,<?class,?selector>) if
in terms of
analyser(classImplementsMethodNamed(?class,?selector)).
derived rules
analyser(classImplementsMethodNamed(?class, handle:)) if
methodSendsMessage(?class, handle:, newConverterFor:),
methodSendsMessage(?class, handle:, newClosure),
methodSendsMessage(?class, handle:, analyse),
classInHierarchyOf(?class,Scheme.SpecialFormHandler),
classInHierarchyOf(?class,Scheme.SpecialFormHandlerWithSuccessor),
classInHierarchyOf(?class, ?class).
Redundant
August 2003 14
© PROG, VUB

15. Derived Rules
analyser(classImplementsMethodNamed(?class, analyse)) if
methodSendsMessage(?class, analyse, newClosure),
classInHierarchyOf(?class,Scheme.ScExpression),
classInHierarchyOf(?class, ?class).
Redundant
analyser(classImplementsMethodNamed(Scheme.DefineRelHandler,handle:)).
August 2003 15
© PROG, VUB

16. Discussion
ÿ Results show that intentions are discovered
ÿ Problems encountered
ÿ algorithm is sensitive to order of examples presented
ÿ sufficient number of examples is needed
ÿ rules are either too restrictive or too general
ÿ performance issues
ÿ Scalability
ÿ only two small experiments, no large-scale study yet
August 2003 16
© PROG, VUB

17. Prototype Tool Support
August 2003 17
© PROG, VUB

18. Conclusion
ÿ Induced intentional views combine advantages of
extensional and intentional views, while removing their
respective disadvantages
ÿ Can be used to tackle software documentation problems
ÿ explicit link between source code and documentation by means of
LMP
ÿ robust w.r.t. evolution
ÿ Can be integrated easily into already existing development
tools
August 2003 18
© PROG, VUB