460 likes | 2.02k Views
Test Coverage CS-300 Fall 2005 Supreeth Venkataraman Test Coverage Coverage is a measure of how much testing can be done given a set of test cases. Given a set of test cases, how much of the code in the program being tested is covered?
E N D
Test Coverage CS-300 Fall 2005 Supreeth Venkataraman
Test Coverage • Coverage is a measure of how much testing can be done given a set of test cases. • Given a set of test cases, how much of the code in the program being tested is covered? • Or.. Given a set of test cases how many functionalities as listed in the requirements can be tested? • Coverage in most cases is a term associated with white-box testing or structural testing. Supreeth Venkataraman
Black Box Testing • Black-box testing involves deriving test cases based on the functional requirements of the program being tested. • “Does the software exhibit the intended behavior for each of its functionalities?” • Black-box testing can be used to detect improper functionality, errors of interface, errors in data structures, and initialization and termination errors with little regard for the logical structure of the program. [Pressman 97] Supreeth Venkataraman
Black Box Testing • We have already seen many types of blackbox testing • Fault-Based testing, boundary-value testing, etc. • Functional testing is thought of as being superior to structural testing in most cases • However, it is acknowledged that code based testing does uncover a lot of defects. Supreeth Venkataraman
Equivalence Partitioning • A good test case is one that has a good likelihood of uncovering a fault. • Exhaustive testing is impossible due to the infinite domain of inputs. • We must choose a suitable subset of inputs so that “it covers a large part of other possible test cases” – Myers, p.45 Supreeth Venkataraman
Equivalence Partitioning • Partition the input domain into equivalence classes • The goal here is to partition the inputs in such a way that if one test case in an equivalence class is capable of detecting some fault, then all test cases in that class are capable of detecting the same fault. Supreeth Venkataraman
Equivalence Partitioning • An example of an equivalence class for E-LIB. • A set of strings that obey the rules of structure of the SSN. • While checking for the structure of the SSN, if one test case from our class does not uncover a fault, it is reasonable to assume that none of the strings in the class will reveal a fault. Supreeth Venkataraman
White Box Testing • White-box testing is also known as structural testing or code-based testing. • The test cases to be used to test the program are derived from the program structure, and not from functional specifications. • Many many variations of white-box testing exist. Supreeth Venkataraman
White Box Testing – and criticisms • A lot of research has been undertaken to discover the best kind of code coverage. • “All testing that does not derive from functions is fundamentally misguided” • “People do not use software in order to execute its statements; rather they invoke its functionality” Supreeth Venkataraman
White box or Black box? • Achieving code coverage does not imply that good functional coverage is achieved as well. • A good mix of functional and structural testing is desirable as both functional as well as code coverage are important. • Structural test cases must wait to be derived until the code is written, but functional tests can be derived from requirements. • Weyuker proposed that structural testing be used as a way of evaluating functional test quality. Supreeth Venkataraman
White box or Black Box? • In many cases, structural coverage is done after adequate functional coverage is achieved. • Structural testing is used to back up functional testing • If structural testing is inadequate, create additional functional test cases. Supreeth Venkataraman
Test Adequacy • The term “adequacy” can be thought of as asking oneself the question “How many test cases do I need in order to completely cover the entire program based on the technique that I have chosen for testing? • Or….When do I stop testing? Supreeth Venkataraman
Control Flow Testing • Control-flow testing is based on the flow of control in a program. • There are many variations of control-flow testing. Statement coverage, branch coverage, path coverage etc. • We will examine statement coverage and branch coverage in some detail. Supreeth Venkataraman
The Program Flowgraph • A program flowgraph graphically represents the potential control flow of a program. • Program flowgraphs are made up of nodes and directed edges • Each node represents a basic block, which can be defined as a collection of statements that are always executed together. • Each directed edge between two nodes A and B in a control flow graph represents a potential control path from A to B Supreeth Venkataraman
Example Program void myflow(int x,int y) { int a; if (y == 0) { a = 1; } else { a = y; } y = y – 1; if (a > 0) { a = x; } else { a = x + 1; } printf(“%d\n”, a); } Supreeth Venkataraman
2 3 y == 0 y != 0 4 6 7 8 a > 0 a ≤ 0 9 11 12 • void myflow(int x, int y) • { • 2. int a; • if (y == 0) { • a = 1; • } else { • a = y; • } • y = y – 1; • if (a > 0) { • a = x; • } else { • a = x + 1; • } • printf(“%d\n”, a); • } Supreeth Venkataraman
Statement Coverage • Achieving statement coverage means that all statements in the program have to be executed at least once by a set of test cases • Having complete statement coverage does not necessarily mean that the program cannot fail • Think about the input space.... • Statement coverage is also called all-nodes testing. Supreeth Venkataraman
Statement Coverage Test Cases Test case1: input = {x = 7, y = 0} expected output = {a = 7} Test case2: input = {x = 7, y = -1} expected output = {a = 8} Supreeth Venkataraman
Branch Coverage • Achieving branch coverage means that there must be enough test cases so that every path for a conditional transfer of control is exercised • Or... Every true and false branch have to be exercised by some test case • Complete coverage means that the test cases should be sufficient to traverse all the edges in the program graph. • This form of testing is also known as all-edges testing . Supreeth Venkataraman
Branch Coverage • The fact that there is complete branch coverage does not mean that all errors will be found. Branch testing does not necessarily check all combinations of control transfers. Supreeth Venkataraman
Branch Coverage Test Cases Test case1: input = {x = 7, y = 0} expected output = {a = 7} Test case2: input = {x = 7, y = -1} expected output = {a = 8} Supreeth Venkataraman
Path Coverage • Path coverage means that all possible execution paths in the program must be executed. • This is a very strong coverage criterion, but impractical. • Programs with loops have an infinite number of execution paths, and thus would need an infinite number of test cases Supreeth Venkataraman
Data Flow Testing • Data flow testing is based on the definition and use of variables in a program. • Deals with flow of data rather than flow of control. • There are many data-flow testing methods • All-definitions coverage, all-uses coverage, all-du-paths coverage etc. • We will look at all-uses testing in some detail Supreeth Venkataraman
Terminology (Beizer) • Definition– a variable is said to be defined when it is initialized with a value as part of a declaration, or when it appears on the left hand side of an assignment statement . • eg: x = 4; int y = 0; Supreeth Venkataraman
Terminology • Use– The value of a variable is said to be used either when the variable appears on the right hand side of an assignment statement or when it appears in a predicate statement like if (x > y) (Beizer) • Eg: y = x; Supreeth Venkataraman
Terminology • Definition-clear path - This is a set of connected nodes in the flowgraph where some variable x is defined in the first node, and is not subsequently redefined in any other node in the path. • For example, the path 2-3-4-7-8-9-12 is a definition-clear path with respect to variable “x” in figure 2.2. Supreeth Venkataraman
2 3 y == 0 y != 0 4 6 7 8 a > 0 a ≤ 0 9 11 12 • void myflow(int x, int y) • { • 2. int a; • if (y == 0) { • a = 1; • } else { • a = y; • } • y = y – 1; • if (a > 0) { • a = x; • } else { • a = x + 1; • } • printf(“%d\n”, a); • } Supreeth Venkataraman
All-Uses Testing • The all-uses treatment states that there must be at least one definition-clear path exercised by some test case from every definition of a variable x to all of its uses. • All-uses testing is a powerful test coverage criterion. Supreeth Venkataraman
All-Uses test cases • Test case1: input = {x = 7, y = 0} expected output = {a = 7} Path = {2-3-4-7-8-9} • Test case2: input = {x = 7, y = -1} expected output = {a = 8} Path = {2-3-6-7-8-9} • Test case3: input = {x = 7, y = 4} expected output = {a = 7} Path = {2-3-6-7-8-11} Supreeth Venkataraman
The Subsumes Relationship • The subsumption relationship means that satisfying one test coverage criterion may implicitly force another test coverage criterion to be satisfied as well • Eg: Branch coverage forces statement coverage to be attained as well (This is strict subsumption) • “Subsumes” does not necessarily mean “better” Supreeth Venkataraman
Subsumption Hierarchy (Rapps and Weyuker) Supreeth Venkataraman
Mutation Testing • Mutation testing involves changing (mutating) the program source statement(s) • If the actual output from a test case is different from the expected output (hopefully it will be), the test case is said to kill the mutant. • Mutation coverage requires that all mutants be killed Supreeth Venkataraman
Some Mutation Examples • Operator mutation – change the operator in the mutated version. • Eg: change x + y to x – y or x * y … • Change x < y to x <=y or x >=y … • If all the mutants are killed we are satisfied that no wrong operator has been mistakenly used. Supreeth Venkataraman
Choosing a Testing Method • The process • If the development process is not well defined without a proper test plan one does not really know what type of testing to apply. • If the project starts getting delayed, then testing time is extremely limited. • Resources available • testing time, testing experience, budget ... Supreeth Venkataraman
Choosing a Testing Method • Feasibility Issues • Structural testing may not achieve complete code coverage because of dead code • Dead code is code that never gets executed. • There could be dead statements, dead paths etc. • The tester must devote time to identifying infeasible code. Supreeth Venkataraman
Conclusions • In practice, there is no agreement as to what the payoffs are for each form of structural testing • Functional testing has advantages over structural testing because functional tests capture the expected actions of users. Supreeth Venkataraman