Lightweight Extraction of Object Models from Bytecode

1
Lightweight Extraction of Object Models from
Bytecode

• Daniel Jackson, Allison Waingold
• ICSE 99 , TSE 00
• 2002. 5. 14
• Presented by Jeon Sanguk

2
Contents

• Introduction
• Extraction Mechanism
• Inferring Association
• Inferring Multiplicity
• Inferring Mutability
• Experiment
• Conclusion

3
Introduction(1/3)

• Object model
• Is a representation of the abstract state of a
  program
• An architectural view of a program
• Takes the form of a graph
• Node object
• Edge subset relationship or association

4
Introduction(2/3)

• Usage of object model
• Requirement
• Capture the structure of the problem domain
• Design
• Articulate overall structure
• Guide implementation
• Maintenance
• Understand a program
• Rework a program

5
Introduction(3/3)

• Womble
• Is a tool that extracts object models from Java
  bytecode
• Applies a simple, lightweight analysis to the
  code
• Handles container classes by inferring the types
  of elements stored in a container

class Company Vector employees
employees

Company
Person
employees
Company
Vector
6
Notations

• Association
• extends or implements
• Multiplicity
• (zero or more), ?(zero or one), !(exactly one)
• Mutability

class A B r
r
B
A
class A implements B class A extends B
B
A
r
B cannot change during its lifetime
B
A
7
Extraction Mechanism

• Use of most of information in the classfile
• Name of the class
• Name of the class it extends or implements
• The declarations of fields
• The signature of methods
• The bytecode instructions within methods
• Extraction of basic structure

class A B p C q
p
B
A
q
C
8
Inferring Associations(1/2)

• Ignores fields of primitive type
• Identification of container class
• Its methods and fields reference only built in
  classes(from the java.lang package)
• There is at least one method that takes or
  returns an object of class Object

class A void A_Method(Object o,)
class B Object A_Method()
9
Inferring Associations(2/2)

• Inference of the type of elements stored in it
• Classes appearing in downcasts that are applied
  to objects returned by container methods
• Classes passed as arguments to container methods

r
Vector
Certain_class
v (A)container.a_method()
A a container.a_method(a,)
r

Certain_class
A
10
Inferring Multiplicity(1/2)

• Zero or more
• Default in case of array and container class
• Zero or one / exactly one
• If the field is not an array and does not
  reference a container class
• If the field is ever set to null, then zero or
  one
• Otherwise exactly one

class A B p public A() p
null
p
B
A
11
Inferring Multiplicity(2/2)

• Zero or more
• Default
• Zero or one
• The B object is created in A
• The field p is private
• No method in A returns an object of class B
• Exactly one
• No attempt to identify

p
B
A
12
Inferring Mutability

• Mutability of a field
• If a value is assigned to it in any method of the
  class that is not a constructor
• Mutability of a container
• If it has a method that takes arguments of class
  Object and returns void
• The most unsound analysis in their work

class A void A_Method(Object o,)
13
Example
14
Using Extracted Object Models

• Sketching gross structure
• Shows the essential elements of a programs
  structure
• Comparing to design
• Identifying design patterns
• Identifying data structures
• Looking for design anomalies
• E.g) Useless redundancy
• Looking for bugs

15
Experiment(1/5)

• Suite of programs
• Blob A graph drawing tool
• Alcoa An object constraint solver
• Fusion An environment for teaching graphics
  programming
• Grappa A graph drawing toolkit
• Haystack An information retrieval system
• Rivet An open Java virtual machine for dynamic
  analysis
• Womble itself

16
Experiment(2/5)

• Basic statistics for the suite of programs

17
Experiment(3/5)

• Inferring multiplicity and immutability

18
Experiment(4/5)

• Inferring container classes

17
19
Experiment(5/5)

• Comparison of association annotations

20
Flaws in Womble

• Containers
• Cannot infer actual type of element stored in
  nested container
• Cannot distinguish keys and values in a hashtable
• Cannot product qualified association

21
Superwomble

• New version of Womble
• Derives specification from each class
• Gives the types of methods and fields
• Can handle nested container hashtable
• Elision of containers is optional

22
Conclusion

• Womble
• Is a tool that extracts object models from Java
  bytecode
• Uses a collection of simple heuristics
• The analysis is linear, mostly monotonic, and
  reasonably accurate