Verification and Validation Overview References: Shach, Object Oriented and Classical Software Engineering Pressman, Software Engineering: a Practitioner’s Approach, McGraw Hill Pfleeger, Software Engineering, Theory and Practice, Prentice Hall Davis, Software Requirements: Objects, Functions, and States, Prentice Hall 1209

Purpose of V&V Groups of 3 2 minutes What is the purpose of verification and validation?

Purpose of V&V Give programmers information they can use to prevent faults Give management information to evaluate risk Provide software that is reasonably defect free Achieve a “testable” design (one that can be easily verified) Validate the software (demonstrate that it works)

Definitions-1 (note: usual definitions, but they do not match all authors or the 1990 IEEE glossary) Validation –Evaluation of an object to demonstrate that it meets expectations. (Did we build the right system?) Verification –Evaluation of an object to demonstrate that it meets its specification. (Did we build the system right?) –Evaluation of the work product of a development phase to determine whether the product satisfies the conditions imposed at the start of the phase. Correct Program –Program matches its specification Correct Specification –Specification matches the client’s intent

Definitions-2 Error (a.k.a. mistake): –A human activity that leads to the creation of a fault –A human error results in a fault which may, at runtime, result in a failure –Kaner: “It’s an error if the software doesn’t do what the user intended” Fault (a.k.a. bug) –May result in a failure –A discrepancy between what something should contain (in order for failure to be impossible) and what it does contain –The physical manifestation of an error Failure (a.k.a. symptom, problem, incident) –Observable misbehavior –Actual output does not match the expected output –Can only happen when a thing is being used

Definitions-3 Fault identification –Process of determining what fault caused a failure Fault correction –Process of changing a system to remove a fault. Debugging –The act of finding and fixing program errors

Definitions-4 Testing –The act of designing, debugging, and executing tests Test –A sample execution to be examined Test case –A particular set of input and the expected output Oracle –Any means used to predict the outcome of a test

Definitions-5 Significant test case –A test case with a high probability of detecting an error –One test case may be more significant than another Significant test set –A test set with a high probability of detecting an error –A test set is more significant than another if the first is a superset of the second –The number of test cases does not determine the significance Regression testing: –rerun a test suite to see if a change fixed a bug a change introduced a new one

Definitions-6 Let S be a relation, a specification of a program. Let P be the implementation of the program. R is the Range. r  R. D is the domain. d  D. S (r,d).The specification. P (r,d).The implementation.

Definitions-7 Failure P (r,d) but not S (r,d) Test case A pair (r,d) such that S (r,d) Test set T A finite set of test cases. P passes T if  t  T, t=(r,d)  S (r,d)  P (r,d) T is ideal if (  d,r | S (r, d)   (P(r, d)))  (  t  T | t=(r’,d’)  S (r’,d’)   P(r’,d’)) R: Range. r  R. D: Domain. d  D. S (r,d).The specification. P (r,d).The implementation.

Definitions-7 Failure P (r,d) but not S (r,d) Test case A pair (r,d) such that S (r,d) Test set T A finite set of test cases. P passes T if  t  T, t=(r,d)  S (r,d)  P (r,d) T is ideal if (  d,r | S (r, d)   (P(r, d)))  (  t  T | t=(r’,d’)  S (r’,d’)   P(r’,d’)) In groups, translate these into English

Parnas: “There are only three engineering techniques for verification” Mathematical analysis Exhaustive case analysis Prolonged realistic testing

Parnas: “There are only three engineering techniques for verification” Mathematical analysis –Works well for continuous functions (software engineering is more difficult than other engineering) –Cannot interpolate reliably for discrete functions Exhaustive case analysis Prolonged realistic testing

Parnas: “There are only three engineering techniques for verification” Mathematical analysis Exhaustive case analysis –Only possible for systems with small state space Prolonged realistic testing

Hierarchy of V&V techniques Static Analysis V&V Dynamic Techniques Model Checking Symbolic Execution Testing Informal Analysis Formal Analysis Static Techniques Proofs Review Inspection Walkthrough

Types of Faults In group 3 minutes List all the types and causes of faults: what can go wrong in the development process?

Some Types of Faults Algorithmic: algorithm or logic does not produce the proper output for the given input Syntax: improper use of language constructs Computation (precision): formula’s implementation wrong or result not to correct degree of accuracy Documentation: documentation does not match what program does Stress (overload): data structures filled past capacity Capacity: system’s performance unacceptable as activity reaches its specified limit Timing: code coordinating events is inadequate Throughput: system does not perform at speed required Recovery: failure encountered and does not behave correctly.

Causes of Faults Requirements System Design Program Design Program Implementation Unit Testing System Testing Incorrect or missing requirements Incorrect translation Incorrect design specification Incorrect design interpretation Incorrect semantics Incorrect documentation Incomplete testing New faults introduced correcting others

Some Verification and Validation Techniques Requirements Operation Design Maintenance Implementation Testing Reviews: walkthroughs/inspections Synthesis Model Checking Correctness Proofs Traditional Runtime Monitoring

Effectiveness of Fault Detection Techniques

Groups: 2 min. What does this slide say?

Error Estimates: 3 errors per 1000 keystrokes for trained typists. 1 bug per 100 lines of code (after publication) 1.5 bugs per line of code (all together, including typing errors). Testing is 30-90% of the cost of a product. probability of correctly changing a program 50% if less than 10 lines, 20% if between 10 and 50 lines