A%20Slicing%20Method%20for%20Object-Oriented%20Programs%20Using%20Lightweight%20Dynamic%20Information

1
A Slicing Method forObject-Oriented
ProgramsUsingLightweight Dynamic Information

• Fumiaki OHATA, Kouya HIROSE,
• Masato FUJII and Katsuro INOUE
• Osaka University, JAPAN

2
Contents

• Program Slice
• Dependence-Cache (DC) Slice
• Object-Oriented Dependence-Cache (OODC) Slice
• Implementation
• Evaluation
• Summary and Future Work

3
Contents (1/6)

• Program Slice
• Dependence-Cache (DC) Slice
• Object-Oriented Dependence-Cache (OODC) Slice
• Implementation
• Evaluation
• Summary and Future Work

4
Program Slice (Slice)

• Subprogram which affects the value of slicing
  criterion lts, vgt in p
• s Statement
• v Variable
• p Program
• Applications
• Program understanding
• Program debugging

5
Application Example Program Debugging

• Slice is effective on fault localization process.
• The slice for lt9, maxgt indicates the statements
  that might cause the unexpected value of max at
  line 9, so that we have only to focus on them.

6
Computation Process

• Phase 1 Defined and referred variables
  extraction
• Phase 2 Dependence analysis
• Data dependence (DD) analysis
• Control dependence (CD) analysis
• Phase 3 Program dependence graph (PDG)
  construction
• Phase 4 Slice extraction using PDG traversal

7
Data Dependence (DD) Analysis

• Extract DD relations between two statements
• DD relation represents
• data-flow
• through variable.

1 a 3 2 b 2 3 scanf("d", c)
4 if ( c 0 ) 5 d a 6 else 7 d
a 1 8 e a b 9 printf("d",
d) 10 printf("d", e)
d
d
8
Control Dependence (CD) Analysis

• Extract CD relations between two statements
• CD relation represents
• control-flow
• from conditional expression
• to conditional predicate
• or
• from method invocation
• to method definition.

1 a 3 2 b 2 3 scanf("d", c)
4 if ( c 0 ) 5 d a 6 else 7 d
a 1 8 e a b 9 printf("d",
d) 10 printf("d", e)
9
Static Slice

• Scope all possible execution paths
• Target source code
• Execution not required
• Dependence analysis
• DD static
• CD static
• Advantage
• Small analysis cost
• Disadvantage
• Imprecise analysis results

1 a 3 2 b 2 3 scanf("d", c)
4 if ( c 0 ) 5 d a 6 else 7 d
a 1 8 e a b 9 printf("d",
d) 10 printf("d", e)
d
d
Slice for lt9, dgt
10
Dynamic Slice

• Scope single execution path
• Target execution trace
• Execution required
• Dependence analysis
• DD dynamic
• CD dynamic
• Advantage
• Precise analysis results
• Disadvantage
• Large analysis cost

- input 0 for c -
1 a 3 2 b 2 3 scanf("d", c)
4 if ( c 0 ) 5 d a 6 else 7 d
a 1 8 e a b 9 printf("d",
d) 10 printf("d", e)
d
d
11
Contents (2/6)

• Program Slice
• Dependence-Cache (DC) Slice
• Object-Oriented Dependence-Cache (OODC) Slice
• Implementation
• Evaluation
• Summary and Future Work

12
Problems

• Static slice has two problems,
• array indexes problem it is difficult for us to
  determine the values of array indices, and
• pointer alias problem it is difficult for us to
  determine the destination of pointer variables,
• so that extracted DD relations are imprecise.
• Dynamic slice can resolve these problems
  however, it requires large analysis cost.

13
Dependence-Cache (DC) Slice

• Scope single execution path
• Target source code
• Execution required
• Dependence analysis
• DD dynamic
• CD static
• Advantage
• More precise analysis
• results than static slice
• Smaller analysis cost than dynamic slice

- input 0 for c -
1 a 3 2 b 2 3 scanf("d", c)
4 if ( c 0 ) 5 d a 6 else 7 d
a 1 8 e a b 9 printf("d",
d) 10 printf("d", e)
14
Dynamic DD Analysis

• DD relation DD(s, t, v) exists when the following
  conditions are all satisfied
• statement s defines variable v, and
• statement t refers v, and
• at least one execution path from s to t without
  re-defining v exists.
• Dynamic analysis
• On program execution, we have only to trace the
  most-recently defined statement for each variable
  using cache.

15
Cache

• Cache(v) statement that defined variable v
  most-recently.
• Operations for caches
• Before program execution,
• For each variable v, Cache(v) ? ?.
• On program execution,
• For each statement s,
• - when v is defined, Cache(v) ? s.
• - when v is referred, we extract DD relation
  DD(Cache(v), s, v).

16
Comparison withStatic Slice Dynamic Slice

• Analysis precision (slice size)
• Static slice ? DC slice ? Dynamic slice
• Analysis cost (memory space computation time)
  
• Static slice lt DC slice Dynamic slice

Static slice DC slice Dynamic slice
DD analysis Static Dynamic Dynamic
CD analysis Static Static Dynamic
Target Source code Source code Execution trace
Ashida, Y., Ohata, F. and Inoue, K. Slicing
Methods Using Static and Dynamic Information,
Proceedings of the 6th Asia Pacific Software
Engineering Conference, 344-350, 1999.
17
Contents (3/6)

• Program Slice
• Dependence-Cache (DC) Slice
• Object-Oriented Dependence-Cache (OODC) Slice
• Implementation
• Evaluation
• Summary and Future Work

18
Object-Oriented DC (OODC) Slice

• Extended DC slice for Object-Oriented (OO)
  programs
• OO languages have concepts which procedural
  languages do not have.
• Class, Object (Instance)
• Inheritance, Class hierarchy, Method overriding
• Dynamic binding based on the reference-type of
  the object, an appropriate overriding method is
  selected and invoked.

19
Analysis Policy

• Character 1 object is a collection of
  attributes and methods that operate them.
• Character 2 dynamic binding feature exists
  however, static analysis can not handle it
  sufficiently.
• Policy 1 when a variable is created, the
  corresponding cache is also created.
• Policy 2 we use dynamic CD analysis for method
  invocation.

20
Dynamic CD analysis for Method Invocation

• CD relation CD(s, t) about method invocation
  exists when the following conditions are all
  satisfied
• statement t is a method definition, and
• statement s calls t.
• Dynamic analysis
• On program execution, we have only to watch the
  execution trace from a method invocation to a
  method definition.

21
Contents (4/6)

• Program Slice
• Dependence-Cache (DC) Slice
• Object-Oriented Dependence-Cache (OODC) Slice
• Implementation
• Evaluation
• Summary and Future Work

22
Implementation

• DC Slicing System for Java
• Analysis libraries
• GUI
• Developing environment
• JDK1.3.0_01
• JavaCC2.0

23
Analysis Libraries

• 10,000 lines
• Analysis method preprocessor style
• Preprocessor loads target program p, and generate
  program p that contains p and the code to
  analyze p dynamically.
• Easily development
• Easily optimization using JIT or JavaVM

24
Example Preprocessed Code
- Preprocessed code -
- Original code -

• For each statement s,
• when variable v is referred,
• we insert ref(v, s) before s.
• when v is defined,
• we insert def(v, s) before s.

0 initPDG() 1 def(a, 1) 1 int a
10 2 def(b, 2) 2 int b 20 3 def(c,
3) 3 int c 4 ref(a, 4) 4 ref(b, 4)
4 def(c, 4) 4 c a b 5 ref(c, 5) 5
printf(d\n, c)
1 int a 10 2 int b 20 3 int c 4
c a b 5 printf(d\n, c)
Developed Preprocessor
25
GUI

• 3,000 lines
• Features
• Program editing
• Slice computation

26
Contents (5/6)

• Program Slice
• Dependence-Cache (DC) Slice
• Object-Oriented Dependence-Cache (OODC) Slice
• Implementation
• Evaluation
• Summary and Future Work

27
Metrics Values (1/2)

• Sample programs
• Analysis precision (slice size) lines

Program Classes Override methods Lines Description
P1 2 0 223 CGI program
P2 3 7 226 Paint program
Slicing criterion Static DC Dynamic
P1(1) 26 15 15
P1(2) 83 27 27
P2(1) 48 14 14
P2(2) 45 12 12
28
Metrics Values (2/2)

• Analysis cost
• Computation cost (execution time) ms
• T1 original code, T2 preprocessed code
  (original code analysis code)
• Space cost (memory use on execution) KByte

Program T1 T2 T2/T1
P1 138 582 4.22
P2 N/A N/A N/A
Program T1 T2 T2/T1
P1 478 645 1.35
P2 836 920 1.10
29
Evaluation

• Analysis precision
• Slice size 30-60 of Static slice
• Static slice ? DC slice ? Dynamic slice
• Analysis cost
• Additional cost for dynamic DD analysis and
  dynamic CD analysis for method invocation is not
  so large.
• - Computation cost ?4.2
• - Space cost ?1.2

Ashida, Y., Ohata, F. and Inoue, K. Slicing
Methods Using Static and Dynamic Information,
Proceedings of the 6th Asia Pacific Software
Engineering Conference, 344-350, 1999.
30
Contents (6/6)

• Program Slice
• Dependence-Cache (DC) Slice
• Object-Oriented Dependence-Cache (OODC) Slice
• Implementation
• Evaluation
• Summary and Future Work

31
Summary

• Classification of slicing methods
• OODC slice
• An intermediate slice between static slice and
  dynamic slice for OO programs
• Dynamic DD analysis
• Dynamic CD analysis for method invocation
• Static CD analysis (except method invocation)
• Implementation (DC Slicing System for Java)
• Evaluation (Experimentation using sample Java
  programs)

32
Future Work

• Application of dynamic CD analysis to other
  dynamically determined elements in Java
• Exception
• Thread
• Experimental comparison with other slicing
  methods on analysis cost
• Application of OODC slice to large programs
• JavaVM-based (interpreter style) DC Slicing
  System for Java (now developing)

33
End.
34
Example Phase 2 (DD Analysis)
1 include ltstdio.hgt 2 3 float
absolute(float x) 4 5 if(x lt 0) 6
x -1 x 7 8 return x 9
10 11 float ave(int a, int b, int c) 12
13 int sum 14 float x 15 16
sum a b c 17 x (float) sum /
3.0 18 x absolute(x) 19 return
x 20 21 22 int main(void) 23 24
int a, b, c 25 float x 26 27
printf("Input a b c ?") 28 scanf("d d
d", a, b, c) 29 30 x ave(a, b,
c) 31 printf("Ave 9.3f\n", x) 32
return 0 33
35
Example Phase 2 (CD Analysis)
1 include ltstdio.hgt 2 3 float
absolute(float x) 4 5 if(x lt 0) 6
x -1 x 7 8 return x 9
10 11 float ave(int a, int b, int c) 12
13 int sum 14 float x 15 16
sum a b c 17 x (float) sum /
3.0 18 x absolute(x) 19 return
x 20 21 22 int main(void) 23 24
int a, b, c 25 float x 26 27
printf("Input a b c ?") 28 scanf("d d
d", a, b, c) 29 30 x ave(a, b,
c) 31 printf("Ave 9.3f\n", x) 32
return 0 33
36
Example Phase 3
1 include ltstdio.hgt 2 3 float
absolute(float x) 4 5 if(x lt 0) 6
x -1 x 7 8 return x 9
10 11 float ave(int a, int b, int c) 12
13 int sum 14 float x 15 16
sum a b c 17 x (float) sum /
3.0 18 x absolute(x) 19 return
x 20 21 22 int main(void) 23 24
int a, b, c 25 float x 26 27
printf("Input a b c ?") 28 scanf("d d
d", a, b, c) 29 30 x ave(a, b,
c) 31 printf("Ave 9.3f\n", x) 32
return 0 33

• Node statement or conditional expression
• Edge dependence relation between two nodes

37
Example Phase 4 (Slice for lt30, xgt)
1 include ltstdio.hgt 2 3 float
absolute(float x) 4 5 if(x lt 0) 6
x -1 x 7 8 return x 9
10 11 float ave(int a, int b, int c) 12
13 int sum 14 float x 15 16
sum a b c 17 x (float) sum /
3.0 18 x absolute(x) 19 return
x 20 21 22 int main(void) 23 24
int a, b, c 25 float x 26 27
printf("Input a b c ?") 28 scanf("d d
d", a, b, c) 29 30 x ave(a, b,
c) 31 printf("Ave 9.3f\n", x) 32
return 0 33

• We start PDG traversal from the slicing criterion
  node in reverse order.
• The corresponding statements to the reachable
  nodes form the slice.