Detecting Implicit Collaboration Patterns

1
Detecting Implicit Collaboration Patterns

• Gabriela Arévalo, Frank Buchli, Oscar Nierstrasz
• Software Composition Group
• University of Bern Switzerland
• http//www.iam.unibe.ch/scg

2
Roadmap

• Motivation
• General Approach
• Formal Concept Analysis in a Nutshell
• FCA in detecting Collaboration Patterns
• Validation Case Studies Results
• Conclusions Future Work

3
Motivation Pattern Mining

• Pattern is a recurring set of properties
  (structural relationships) that characterize a
  set of classes.
• design patterns, architectural constraints,
  idioms and conventions adopted in the system.
• Detecting implicit contracts is not trivial
• The documentation is out-of-date
• The information is not explicit in the code

4
Our Approach
5
Formal Concept Analysis in a Nutshell
Aspect Patterns FCA FanIn Static Dynamic
Mariano True True True
Arie True True True True
Gaby True True True
Top

• Concept Maximal group of elements based on
  their common properties, and maximal group of
  properties based on their elements.
• Lattice partial order over the concepts

Bottom
6
FCA Mapping Elements Properties
A

• Elements
• Tuples of classes
• Number of classes order
• Properties
• Unary Relations isAbstract,
• isRoot, isSingleton, hasLocalMethodDefined
• Binary Relations isSubclass, accesses,
  hasAsAttribute, invokes, usesLocally

C
P
Element (C A P) C isSubclass A, P isSubclass
A, P invokes A, A isAbstract
7
FCA Mapping Incidence Table
2
A
(1,2)Sub (3,2)Sub (1,2)Acc (3,2)Inv (2)Abs
(C A P) True True True True
(C A B) True True True True
(Z X Y) True True True
(Z X P) True True True True
C
P
Tuples of classes
1
3
Element (C A P) 1 isSubclass 2, 3 isSubclass
2, 3 invokes 2, 2 isAbstract (1,2)Sub
(3,2)Sub (3,2)Inv (2)Abs
A B C P X Y Z all classes in a system
Properties Set of indexes to show relationships
in the tuples. (i,j)relation and (i)feature
8
Post-Filtering (I)
2
(1,2)Sub (3,2)Sub (1,2)Acc (3,2)Inv (2)Abs
(C A P) True True True True
(C A B) True True True True
(Z X Y) True True True
(Z X P) True True True True
A
C
P
Tuples of classes
1
3
(1,2)Subclass (3,2)Subclass (2)Abstract (3,2)Acces
ses

• Remove unconnected patterns

9
Post-Filtering (II)

• Merging Equivalent Patterns

S
(T U S) (1,3)Sub (2,1)Acc
D
3
(D E F) (2,1)Sub (3,2)Acc
1
E
F
T
U
2
1
3
2
Permutation of indexes ? 1 ? 2, 2 ? 3, 3 ?
1 ?((1,3)Sub (2,1)Acc)) ((2,1)Sub (3,2)Acc))
(T,U,S) is transformed in (S,T,U)
10
Validation Case Studies

• ADvance multidimensional OOAD-tool
  (www.io.com/icc)
• CodeCrawler language independent software
  visualization tool (www.iam.unibe.ch/scg)

11
Selected Collaboration Patterns
Attribute Star
Attribute Chain
A
Order 3
A
Order 3
Order 4
Order 4
Composite
Bridge
A
A
Order 3
Order 4
Order 3
Order 4
12
Statistics about Detected Patterns
13
Evaluation of Results

• Classes as Center of Systems
• CodeCrawler CCTool is in none of patterns
• CCNodePlugin is in 47 patterns up to order 4
• Identify candidates of design patterns
• Façade, Adapter and Bridge
• Coding Styles frequency of some patterns
• CodeCrawler heavy use of Subclass Star and
  Façade
• Advance Composite

14
Conclusions

• Development of an approach that detect unknown
  recurring patterns
• Collaboration Patterns cover more than Design
  Patterns
• Incremental pattern library
• FCA used as a metatool
• FCA is an exciting tool but
• Critical issue Choice of Properties
• Performance of the algorithm
• Filters

15
Future Work

• Incremental Concept Analysis with combination of
  Iceberg idea.
• Use higher abstraction level information
  (isLeaf, isComponent, isFacade)
• Solve the scalability problem
• Better name guessing less restrictive
  interpretation
• Use different orders of tuples in the same lattice

16
The End

• Questions ?

P
Abstract Z
B
C
D
Thank you very much !!!
17
(No Transcript)
18
Problem Example
Abstract X
P
I
J
K
Z
Y
Abstract A
T
F
C
B
H
G
D
19
Comparison

• Refinement and extension of Tonella and Antoniol
  approach to detect classical design patterns
• Similarities
• Source code of an object oriented application
• Extract structural relationships between classes
• Differences
• Improvements to avoid redundancy in the
  information
• Improvements the time performance
• Generalize the technique to a language
  independent approach
• Introduction of a filtering phase to narrow the
  scope of candidate patterns

20
Performance of the Algorithm

• Based on a small experiment of 7 classes

21
Pattern Neighbourhood (1/2)
Almost Pattern
A
X
B
C
Y
Z
A
O3
B
C
A
B
C
Overloaded Pattern
22
Pattern Neighbourhood (2/2)
Cover Pattern
A
D
O4
B
C
A
O3
B
C
O2
A
A
B
C
Sub Pattern
23
Lessons Learned

• Lack of a general methodology
• Modelling software entities as FCA components is
  not a trivial task
• Performance of the algorithms
• Supporting End-Users ConAn PaDi
• Interpretation of the concepts
• Use of Complete Lattice
• Use of the Partial Order

24
Goals

• Detection of classical and non-classical patterns
• DEFINIR QUE ES UN PATTERN
• Establish relationships over detected patterns
  pattern neighbourhoods
• Missing relationships between classes to complete
  a pattern
• Excess relationships between classes that extend
  a pattern
• Incremental construction of a pattern library
• Analysis of advantages and drawbacks of using FCA
  in this approach

25
Formal Concept Analysis in a Nutshell
Rock Jazz Pop Blues Rap
Kim True True
Andy True True
Stef True True
Markus True True True
Orla True True
Gaby True True
Top

• Incidence Table relationships between elements
  and properties.
• Concept Maximal groups of elements based on
  their common properties, and maximal groups of
  properties based on their elements.
• Lattice partial order over the concepts

Bottom