Data Flow Testing

1. Data Flow Testing • Data flow testing(DFT) is NOT directly related to the design diagrams of data-flow-diagrams(DFD). • It is a form of structural testing and a White Box testing technique that focuses on program variablesand the paths: • From the point where a variable, v, is defined or assigned a value • To the point where that variable, v, is used Remember, to generate the path for testing we need to set up the data to drive the path.

2. Static Analysis of Data • Static analysisallows us to check (test or find faults) without running the actual code, and we can apply it to analyzing variables as follows: • A variable that is defined but never used • A variable that is used but never defined • A variable that is defined a multiple times prior to usage. • While these are dangerous signs, they may or may not lead to defects. • A defined, but never used variable may just be extra stuff • Some compilers will assign an initial value of zero or blank to all undefined variable based on the data type. • Multiple definitions prior to usage may just be bad and wasteful logic • We are more interested in “executing” the code than just static analysis, though.

3. Variable Define-Use Testing • In define-use testing, we are interested in testing (executing) certain paths that a variable is defined – to - its usage. • These paths will provide further information that will allow us to decide on choice of test cases beyond just the earlier discussed paths analysis (all statements testing or dd-testing (branch) or linearly independent paths).

4. Data Dependencies and Data Flow Testing(DFT) • In Data Flow Testing (DFT) we are interested in the “dependencies” among data or “relationships” among data ----- Consider a data item, X: • Data Definitions (value assignment) of X: via 1) initialization, 2) input, or 3) some assignment. • Integer X; (compiler initializes X to 0 or it will be “trash”) • X = 3; • Input X; • Data Usage (accessing the value) of X: for 1) computation and assignment (C-Use) or 2) for decision making in a predicate (P-Use) • Z = X + 25; (C-Use) • If ( X > 0 ) then ----- (P-Use)

5. Some Definitions • Defining node, DEF(v,n), is a node, n, in the program graph where the specific variable, v, is defined or given its value (value assignment). • Usage node, USE(v,n), is a node, n, in the program graph where the specific variable, v, is used. • A P-use nodeis a usage node where the variable, v, is used as a predicate (or for a branch-decision-making). • A C-use nodeis any usage node that is not P-used. • A Definition-Use path, du-path, for a specific variable, v, is a path where DEF(v,x) and USE(v,y) are the initial and the end nodes of that path. • A Definition-Clear pathfor a specific variable, v, is a Definition-Use path with DEF(v,x) and USE(v,y) such that there is no other node in the path that is a defining node of v. (e.g. v does not get reassigned in the path. )

6. Simple Example 3 • Pseudo-code Sample • int a, b • input (a, b) • if (a > b) • then Output (a, “ a bigger than b”) • else Output (b, “ b is equal or greater than a”) • end This is type defining not value 4 6 5 7 • The following are examples of the definitions: • DEF(a, 3) – node 3 is a defining node of variable “a” --- a value is assigned to “a” • USE(a, 4) – node 4 is a usage node of variable “a” • USE(a, 5) – node 5 is a usage node of variable “a” • USE (a,4) is a P-use node while • USE(a,5) is C-use node • Path that begins with DEF(a,3) and ends with USE(a,4) is a definition-use path of a • Path that begins with DEF(a,3) and ends with USE(a,5) is a definition-use path of a • Path that begins with DEF(a,3) and ends with USE(a,5) is a definition-clear path of a • Path that begins with DEF(b,3) and ends with USE(b.6) is a definition-use path of b Note that: if we choose the definition-use paths [last two examples above] of both variables a and b, then it is the same as executing the decision-decision (dd) path or branch testing.

7. “Commission Program” Partial Example from Text (pp153-154) :on variable “locks” (using statement number instead of “node”) • we have DEF(locks,13) and DEF(locks,19) • we have USE(locks,14) and USE(locks,16) • We can employ the define/use methodology • and get the following paths: • - path1 = DEF(locks,13) to USE(locks,14) • - path2 = DEF(locks,13) to USE(locks,16) • - path3 = DEF(locks,19) to USE(locks14) • - path4 = DEF(locks,19) to USE(locks,16) other assignments 13. Input (locks) 14. While NOT (locks equals -1) false true • But we have not even tested all the branches! • Modify Path1 to include nodes <13,14,21> and • call it Path1’ • Also modify Path3 to include nodes • <19,’20’,14, 21> and call it Path3’ 15. Input( stocks, barrels) 16. Totallocks= Totallocks + locks 17. Totalstocks =Totalstocks + stocks 18. Totalbarrels = Totalbarrels + barrels 19. Input (locks) 20. EndWhile With Path1’, Path2, Path3’ and Path4, we have covered all 4 cases: 1) Path1’ = by-pass loop, 2) Path2 = drop Into loop, 3) Path3’ = exit loop, and 4) Path4 = repeat the loop 21. Output (“Locks sold “ , Totallocks) continuingother statements

8. Definitions of Definition-Use (DU) testing • All-Defs : contains set of test paths, P, where for every variable v in the program, P includes definition-clear paths from every DEF(v,n) to only oneofits use node. • All-Uses: contains set of test paths, P, where for every variable v in the program, P includes definition-clear paths from every DEF(v,n) to everyuse ofv and to the successor node of that use node. • All-P-Use/Some C-Use: contains set of test paths, P, where for every variable v in the program, P contains definition-clear paths from DEF(v,n) to every predicate –use node of v; and if there is no predicate-use, then the definition-clear path leads to at least one C-use node of v. • All-C-Use/Some P-Use: contains set of test paths, P, where for every variable v in the program, P contains definition-clear paths from DEF(v,n) to every computation-use node of v; and if there is no computation-use, then the definition-clear path leads to at least one predicate-use node of v. • All-DU-paths: contains the set of paths, P, where for every variable v in the program, P includes definition-clear paths from every DEF(v,n) to every USE(v,n) and to the successor node of each of the USE(v,n), and that these paths are either single loop traversals or they are cycle free.

9. Summarizing hierarchy All possible paths All-DU-paths All-Uses All-C-Use/some-P-Use All-P-Use/some-C-Use All-Defs Text page 160 has another chain Under All-P-Use/some-C-Use; take a look at that page.

10. Program Slices • A concept related to dataflow analysis (def-use path) is to look at a slice of program that is related to some “variable of interest” and trace the program statements that are in the program slice. (Program slice was first discussed by Mark Weiser in the 1980’s) • Tracing program statements that contribute or affect the value of the variable of interest at somepoint of the program; { e.g. (v,node) }going “backward” to include all those statements that have affected the variable’s value is considered a slicing of the program with respect to that variable. • The designated program slice then becomes a path that one would consider for testing. • This is also a popular, perhaps subconscious, debugging techniqueused by most of us.

11. A simple example of Program Slice • - Consider that our variable of interest is y • at statement 7. • - But , on second look, we would pick up statements: • 7 • 6 • 4 (because the loop influences statement 6) • 2 • 1 (because limit influences statement 4) • Statements <1,2,4,6,7 > form a program slice • related to variable, y • - We would include this slice as a test case Pseudo code example 1. int limit = 10; 2. int y = 0; 3. int x = 0; 4. for (int i = 0; i < limit ; i++) 5. { x = x + i; 6. y = y + i2 ; } 7. print (“ x = “, x , “y =“, y); Note that: in executing just this slice, the value of y is the same at statement 7 as executing the whole example good for tracing bug but ---- for testing?

12. Definition of Program Slice logical prior, may be physically after • Given a program P and a set of variables, V, in P, a slice on the variable set V at some statement n, denoted at S(V,n), is a set of all statements in P, “priorto” and at n, that contributeto the values of those variables in V at that statement fragment n. (note that V can be one or more variables --- empty set V is not considered ) • “Contribute” is a key word that should be expanded to consider various types of contribution ( usage and definition) : • P- use (in predicate) • C- use (in computation) • O- use (in outputs) • L-use (used as pointers to locations) • I-use (used as part of some iteration counter) • I- def (defined through input) • A-def (defined via assignment)

13. More on Program Slicing • Note that we can’t just blindly apply the usage rules. Note that we have modified statement 5 of the previous example to include y. Now should S(y,7) include statement 5? Even though there is C-use of y, statement 5 does not “contribute” to the value of y in statement 7. So we do not include 5 in the slice, S(y,7). Also, p-use of other variable not in the set V (in this case V = y) may need to be included because it influences the value of y. In this case, the variable, “limit,” in the “for” statement. So, statement 4 is included in S(y.7) as before. And limit is placed into the set V, Variable of interest. Pseudo code example 1. int limit = 10; 2. int y = 0; 3. int x = 0; 4. for (int i = 0; i < limit ; i++) 5. {x = x + i + y; 6. y = y + i2 ; } 7. print (“ x = “, x , “y =“, y); Modified to include y

14. Program Slices with Previous Partial Example other assignments 13. Input (locks) - For variable, locks, we can see that - nodes 13 and 19 are input-definition nodes - node 14 is a predicate – use node - node 16 is a c-use node Now let’s look at some examples - S (locks, 13) = < 13 > - S (locks, 14) = <13, 14, 19, 20> (may split this) - S (locks, 16) = <13, 14, 19, 20> (may split this) - S (locks, 19) = <19> Note that statement 16 is not included in these slices even though it is C-use of locks. 14. While NOT (locks = -1) false true 15. Input( stocks, barrels) 16. totallocks= totallocks + locks 17. totalstocks =totalstocks + stocks 18. Totalbarrels = totalbarrels + barrels 19. Input (locks) 20. EndWhile For variable, stocks: - node 15 is the input-def node - node 14 influencesnode 15 - node 13 influences node 14 - node 19 influencesnode 14 So, Slice(stocks,15) = <13,14,15,19,20> 21. Output (“Locks sold “ , totallocks) continuingother statements

15. Program slices • Although program slice technique is based on the define-use dataflow approach, we do not need to include all the d-u paths that has no real influencing or contribution value to testing ( or debugging) a particular variable. • Program slice requires more analysis, but it allows us to focuson only those parts that are of interest. Remember “testability” from Whitaker’s paper? ---- Perhaps, we can study testability with program slices?