Formal Model of Program Dependences and Its Implications for Software Testing, Debugging, and Mainte

1
Formal Model of Program Dependences and Its
Implications for Software Testing, Debugging, and
Maintenance

• Paper by Podgurski and Clarke
• Presented by Alice Lewis
• 3/15/04

2
September 1990

• IEEE Transactions on Software Engineering what
  a month!

3
Why?

• Create the formal model
• most models to date had been informal
• Present arguments for how to use this for
• software testing
• software debugging
• software maintenance
• What works and what sort of works?
• results support some earlier findings, contradict
  others, and suggest new approaches

4
Four Different Types of Dependences Discussed

• Control dependence
• Data flow dependence
• Weak syntactic dependence
• Strong syntactic dependence
• Introduction of term semantic dependence
• A program statement s is semantically dependent
  on s if the function computed by s potentially
  affects the execution behavior of s

5
Problems

• These dependences are used to find operator
  faults
• The problem of finding all operator faults is
  undecidable but finding program dependences is a
  necessary condition to finding operator faults
  that potentially impact other program statements

6
What to do about undecidable nature of problem?

• Consider a subset/approximation of the problem
• Relate control and data flow dependence to a
  concept called semantic dependence

7
Conclusions

• Determination of syntactic dependences might be
  used to guide more discriminating methods for
  detecting dependence faults
• Dependence analysis can be used to help determine
  what is impacted when code is being debugged or
  changed
• in debug mode, the programmer wants to know what
  statement(s) caused the fault
• in maintenance mode, the programmer wants to know
  what else will be impacted if a statement is
  changed
• in testing we need to know what damage was done
  by any change

8
Three Main Results Presented in the Paper

• Semantic dependence implies weak syntactic
  dependence
• (If semantic dependence does not depend in a
  certain way on a program failing to terminate)
  then semantic dependence implies strong syntactic
  dependence
• Data flow does not imply semantic dependence
• Weak syntactic dependence does not imply semantic
  dependence
• Strong syntactic dependence does not imply
  semantic dependence

9
Control Flow Graph

• Directed graph with the following
• Maximum outdegree of the vertices is at most two
• There are two distinguished vertices index and
  final
• Every vertex occurs on some VI VF walk
• Vertices represent simple program statements
• Arcs represent possible transfers of control

10
Use of Control Flow and def/use Graphs to Depict
Dependences

• Data, Syntactic, and Semantic dependences are
  defined in terms of an annotated control flow
  graph called a def/use graph
• D(v) denotes the set of variables defined at the
  statement represented by v
• U(v) denotes the set of variables used at that
  statement

11
Control Dependence

• Control dependence can be defined strictly in
  terms of the control graph
• Strong control dependence
• u is strongly dependent on v iff there exists a
  v-u walk vWu not containing the immediate forward
  dominator of v
• Weak control dependence
• u is weakly control dependent on v if v has
  successors v and v such that if the branch from
  v to v is executed then u can be bypassed or its
  execution delayed indefinitely
• Weak and strong control dependence can be
  computed in O(IV(G)I3) time

12
Relation between Semantic and Syntactic Dependence

• When can a change in the semantics of a program
  affect the execution behavior of another
  statement?
• the undecidable problem
• So rephrase the question When is one
  statement semantically dependent on another?

13
Some Important Findings/Proofs

• The absence of weak syntactic dependence between
  two statements precludes all relationships
  between them that imply semantic dependence
• The absence of strong syntactic dependence
  between two statements
• does not necessarily preclude all relationships
  between them that imply semantic dependence
• precludes all relationships between them that
  imply finitely demonstrated semantic dependence
• The presence of direct data flow dependence, data
  flow dependence, or weak or strong syntactic
  dependence between two statements does not
  necessarily indicate any relationship between the
  statements that implies semantic dependence

14
Dependence Coverage Criteria

• Test-data selection rules based on covering or
  exercising certain program dependences
• Although its not possible to cover everything
  (complexity is quadratic), you can test high
  risk areas (which walks, which inputs) based on
  syntactic dependences
• Missing (occurs in correct program but not faulty
  one) vs. spurious (reverse is true appears in
  faulty one but not correct one)
• note how often do errors like these occur?

15
Previous Work

• Weiser had flaws in his work
• Weiser proposes that program dependences be used
  to determine the slice in a program that is
  potentially relevant to a given set of statements
• Weiser uses slices to show relevance in debugging
  activities
• the flaw is he assumes semantic dependence
  implies strong syntactic dependence Podgurski
  has proved this is false

16
Analysis of Paper

• Important results were presented with lots of
  substance
• of course much of this was his dissertation
• presented at 1989 ACM SigSoft conference,
  according to Moriconis paper
• Confusing presentation
• definitions first?
• conclusions later?
• central theme throughout?
• had the content of a textbook
• How many bugs come from operator faults or
  dependence faults?
• examples would help