This document discusses using an explanation-based constraint programming (CP) approach to identify the use or misuse of design patterns in software code. It describes how design pattern solutions can be modeled as constraints over classes and relationships in code. The explanation-based CP system called PaLM is used to generate justifications for modifications to code that would transform a misuse into a proper use of a design pattern. Experiments applying this approach to various codebases, including JEdit and JHotDraw, showed promising results. Future work aims to improve the scalability and usability of the system.

1. Ptidej in my PaLM
Using explanations for
design patterns identification

2. Us
Æ Yann-Gaël Guéhéneuc
n 2nd year PhD student
n Object Models and Components Team
n Partly funded by Object Technology International, Inc.
(Canada)
Æ Narendra Jussien
n Assistant Professor
n Constraint Programming Team
Æ Research hosted by the
École des Mines de Nantes

3. Long-term goal
n Toimprove the quality of software
architecture
– To improve flexibility, understandability, …
– Maintenance = 75% of the life of a software
n Design patterns
« Describe simple and elegant solutions to specific
architectural problems in OO software design »
– Experience of skillful developers
– Recurrent architectural problems

4. sic Composite
A Ba xample :
E pattern solution
Composite composes objects into tree structures to
represent part-whole hierarchies. It lets clients treat
individual objects and compositions of objects uniformly.

5. Main ideas
n Design patterns are good micro-
architectural solutions
n Micro-architectures
close to a design
pattern solution may be improved
n Theseare transformed into the
corresponding design pattern solution

6. Problem
n To identify
– The Use of a design pattern solution
• Well written code
• Notion of complete version
– The Misuse of a design pattern solution
• Code improvement
• Notion of distorted version
+ No up-front descriptions of the distortions

7. ic
A Bas xample : Problem
E
Complete and
distorted versions?

8. Complete and distorted versions
n Related works
– The search of sub-graphs in a graph
– The phase of adaptation in case-based
reasoning
n But we need complete and distorted
versions
Æ Explanation-based CP (eCP)

9. Solution
n Explanation-based CP enables
– Justifications
• Programming = « artistic » activity
• Explain why / how the code is modified
– Complete and distorted versions
– Strong interactivity with the user
• Dynamic description of the distortions

10. Explanation-based CP 1/3
n Setof constraints justifying an
action performed by the solver
– Value removal
– Contradiction
n Trace of the solver behavior

11. Explanation-based CP 2/3
n Application
– Assistance upon failure
– Interactive solvers
– New search algorithms
• Path-repair (AAAI 2000)
• Mac-DBT (CP 2000)
•…

12. Explanation-based CP 3/3
n An implementation: PaLM
– Developed at the EMN
– On top of Choco
– Written in Claire
– Used in Path-Repair, Mac-DBT
– Reliable and efficient

13. Our application
n The design pattern solution
– A class defines a variable
– Relationships among classes define
constraints
n The source code of the application
– The classes define the domain of the
variables
– Relationships among classes give the
semantics of the constraints

14. ic
A Bas xample : Constraints
E
n The Composite design pattern solution
n Its constraints
composite < component
leaf < component
composite ⊃ component

15. ic
A Bas xample : Interaction
E
composite < component
leaf < component
composite ⊃ component

16. ic
A Bas xample : Results
E

17. Our results 1/3
n Implementation
– Constraints
• Inheritance, creation, composition, …
– Design pattern solutions
• Composite, Façade, Factory Method, …
n Experiments
– Small test-cases
– JEdit v3.1 (~250 classes), JHotDraw v5.1
(~155 classes), Ptidej itself (!), …

18. Our results 2/3

19. Our results 3/3

20. Conclusion and future
n An application of explanation-based CP
– Non-trivial problem
n Short term
– To add more design pattern solutions
– To develop the constraints system
• New algorithm CSP-NN [Morning session]
– Scalability and weights (noise)
n Long term
– automation and interaction (usability)