CN104461861A - EFSM model-based path testing data generation method - Google Patents

Abstract

The invention discloses a method for generating path test data based on an EFSM model, comprising the following steps: step 1, performing symbolic execution and data flow dependency analysis on the EFSM model, and obtaining the path constraints of each EFSM model; step 2, using a genetic algorithm to generate The test data of the path constraint, given the initial individual; step 3, evaluate the fitness value of the initial individual according to the fitness function, if the fitness value of the individual is 0, then such an initial individual is the test data that satisfies the path constraint, and the algorithm terminates; Step 4. If the fitness value in step 3 does not exceed the set maximum number of generations, two individuals with the minimum fitness value can be selected as the parent individuals in the current generation to perform crossover or mutation operations to generate a new generation of individuals, and the new generation Repeat steps 3 and 4 for one generation of individuals. For paths with complex constraints or involving multiple variables, our method has a much higher success rate than S.Kalaji's method in generating test data satisfying all constraints.

Description

Path test data generation method based on EFSM model

technical field

The invention relates to the field of software testing, in particular to a method for generating path testing data based on an EFSM model. the

Background technique

At present, there have been many researches on program test data generation, and many researchers have applied search-based methods to this field. However, there are not many articles about the test data generation method of the EFSM model. At present, the main methods are as follows:

J. Zhang et al. proposed a path-oriented test data generation method. First, symbolic execution obtains path constraints, and then uses a constraint solver to solve the path constraints to obtain input values that meet the constraints. But the limitation of symbolic execution and constraint solving is that it cannot solve nonlinear constraints. the

R.Lefticaru defined a fitness estimation method to obtain the input sequence of the path. This method applies Tracey's fitness value function to each migration of the path. The fitness value of the path is regarded as a function of each function in the path. Key nodes are defined. The limitation of this method is that it requires that each function cannot contain internal paths, that is, nested IF statements, otherwise Tracy's method cannot be used. the

S. Kalaji proposed to use genetic algorithm to test EFSM. He combined branch distance and close level as the fitness function of genetic algorithm. The fitness function has a low success rate for generating test data for EFSM paths with complex constraints. For the EFSM path containing equality constraints, the test data conforming to all the constraints cannot be generated, and the average constraint rate of the generated test data violation is high, indicating that the quality of the test data is not high. the

R.Lefticaru et al. improved the fitness function of S.Kalaji. They decomposed the path into multiple independent sub-paths, and calculated the fitness value of each sub-path according to the method of S.Kalaji. The fitness value of the whole path is each sub-path The sum of fitness values. They give better fitness values to those individuals that meet more conditions, which is similar to our method. However, their method still suffers from the same problems as S. Kalaji's method, and we cannot always find independent subpaths of a path. the

Contents of the invention

The technical problem to be solved in the present invention is to provide a path test data generation method based on the EFSM model, take the extended finite state machine (EFSM) model as the research object, use the genetic algorithm to generate the test data facing the EFSM path, and calculate the individual adaptation The degree takes into account both its branch distance and the ratio of uncovered conditions. In the experiment, our method is compared with the method of S. Kalaji in the prior art, and the results show that our method has better effect and can obtain test data with better quality. the

To achieve the above object, the technical scheme of the present invention is as follows:

The path test data generation method based on EFSM model, comprises the following steps:

Step 1. Perform symbolic execution and data flow dependency analysis on the EFSM model to obtain the path constraints of each EFSM model;

Step 2. Use the genetic algorithm to generate test data that satisfies the path constraints, given the initial individual;

Step 3. Evaluate the fitness value of the initial individual according to the fitness function. If the fitness value of the individual is 0, then such an initial individual is the test data that satisfies the path constraint, and the algorithm terminates;

Step 4. If the fitness value in step 3 does not exceed the set maximum number of generations, two individuals with the minimum fitness value can be selected as the parent individuals in the current generation to perform crossover or mutation operations to generate a new generation of individuals, and the new generation Repeat steps 3 and 4 for one generation of individuals. the

In a preferred embodiment of the present invention, if the fitness value in step 4 exceeds the set maximum number of generations, the algorithm is terminated. the

In a preferred embodiment of the present invention, the path constraints in step 1 are expressed by parallel IF statements. the

In a preferred embodiment of the present invention, the path constraints only include input variables. the

In a preferred embodiment of the present invention, the fitness function is the sum of the ratios of the branch distances of the initial individual and the conditions in the path constraints not covered by the individual. the

Through the above technical scheme, the beneficial effects of the present invention are:

For paths with complex constraints or involving multiple variables, our method has a much higher success rate in generating test data that satisfies all constraints than the S.Kalaji method;

For paths that can only generate test data that cannot fully satisfy all conditions, that is, paths that contain equations, we use the average constraint rate of the generated test data to measure the quality of the generated test data. The quality of the test data generated by our method is better than that of The method of S.Kalaji is higher. the

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to these drawings on the premise of not paying creative work. the

Fig. 1 is the work flowchart of the present invention. the

Fig. 2a is the EFSM model of embodiment 1 of the present invention. the

Fig. 2b is the EFSM model of embodiment 2 of the present invention. the

Fig. 2c is the EFSM model of embodiment 3 of the present invention. the

Figure 3a is the average generation of test data generated for the extended state machine model of the simple aircraft safety system of the present invention. the

Fig. 3b is the success rate of generating test data for the extended state machine model of the simple aircraft safety system of the present invention. the

Fig. 3c is the average generation of test data generated by the extended state machine model of the transmission protocol of the present invention. the

Figure 3d is the average generation of test data generated by the extended state machine model of the elevator system of the present invention. the

Fig. 3e is the average constraint rate of test data violations produced by the extended state machine model of the elevator system of the present invention. the

Detailed ways

In order to make the technical means, creative features, goals and effects achieved by the present invention easy to understand, the present invention will be further described below in conjunction with specific illustrations. the

Referring to Fig. 1, the path test data generation method based on EFSM model comprises the following steps:

Step 1: Perform symbolic execution and data flow dependency analysis on the EFSM model to obtain path constraints for each EFSM path. the

Step 2: Use the genetic algorithm to generate test data satisfying the path constraints, and an initial individual is given. the

Step 3. Evaluate the fitness value of the individual according to the fitness function, where the fitness function is the sum of the ratios of the branch distance of the individual to the condition in the path constraint not covered by the individual. If the individual's fitness value is 0, then such an individual is the test data that satisfies the path constraint, and the algorithm terminates. the

Step 4. Otherwise, if the set maximum number of generations is not exceeded, two individuals with the minimum fitness value can be selected in the current generation as the parent individuals for crossover operation to generate a new generation of individuals; or one individual with the minimum fitness value can be selected in the current generation. The individual with the fitness value is used as the parent individual for mutation operation to generate a new generation of individuals. If the set maximum number of generations is exceeded, the algorithm terminates. Repeat steps three to four. the

Concepts relevant to the present invention are as follows:

1.1 Extended Finite State Machine (EFSM)

A finite state machine (FSM) is a Mealy automaton that includes a finite state set, a transition set, and an input-output set. Because it cannot model the data part of the system, this paper adopts the extended finite state machine as the system model. The extended finite state machine extends context variables, predicates and operations on the basis of FSM. the

EFSM is a six-tuple (S; s ₀ ; V; I; O; T), where S is a finite set of states; s ₀ is an initial state; V is a set of context variables; I is a finite set of input messages ; O is a finite output message set; T is a finite migration set.

A transition t ∈ T is a five-tuple (s _s ; i; g; op; s _e ), where s _s is the source state of t; i ∈ I is the input, which may be related to the input parameters; A logical expression of a condition; op is an operation composed of an assignment statement or an output statement, etc.; s _e is the target state of t.

When in the state s _s , if the input i is received and the monitoring condition g is satisfied, then trigger transition t=(s _s ; i; g; op; s _e ), at the same time execute the operation in op and the state transitions to s _e . Both g and op can contain input parameters and context variables. We only consider identified EFSMs here. An EFSM is deterministic if, for multiple transitions from the same state with the same input, only one transition's guard condition is satisfied at a time, i.e., only one transition can be triggered at a time.

In EFSM, transition guard conditions can be connected by logical operators AND and OR. Guardian conditions connected by AND we express as nested IF statements. For the guardianship condition connected by OR, we divide it into the number of OR operators + 1 transition, calculate their fitness values respectively, and finally take the minimum value between them as the fitness value of the whole guardianship condition. the

1.2 Symbolic execution

Symbolic execution is a static program analysis method that uses symbolic values instead of actual values to execute programs, and the result of execution is an expression about symbols. This is useful for analyzing the relationship between inputs and outputs. the

When symbolic execution is applied to the test data generation of paths, it characterizes the test data generation problem as a problem of solving the symbolic expressions obtained after executing paths with symbolic values. For example, considering the transition path t ₁ t ₂ t ₃ , the predicates in each transition are x>0, y<15, z≥10, we use symbolic values a, b, c to substitute variables x, y, z respectively, and use The symbolic value executes the path t ₁ t ₂ t ₃ to obtain the symbolic expression (a>0ANDb<15ANDc≥10), and the problem of generating the test data of the path t ₁ t ₂ t ₃ is transformed into finding the symbolic expression (a>0ANDb< 15ANDc≥10) to solve the problem.

1.3 Data flow dependency

There are some input variables and context variables in EFSM. Usually we only use input variables to represent the symbolic expression to be solved. Then for the context variables in the path, we need to use data flow analysis techniques to replace them with corresponding values and input variables. the

Given a variable v, if v is used as an input parameter in the migration t or is assigned a value in the operation of t, then v is said to be defined in t, denoted as def(t). If v is referenced in the predicate of migration t (p-use) or referenced in the operation (c-use), then v is said to be used in t, denoted as use(t). the

Given a migration path from t _i to t _j , v∈def(t _i ) and v∈use(t _j ), if v is not redefined on the transfer between t _i and t _j , then we call The path from t _i to t _j is a well-defined path of v. (t _i ; t _j ) is called a definition-reference pair of v, and the data of t _j depends on t _i .

After getting the data flow dependencies of context variables on each transition, we use the backward replacement technique to replace these context variables with corresponding values and input variables. The so-called backward replacement means that the last migration of the path is processed from the back to the front, and the variable referenced in the migration is replaced with the definition it depends on. The final path expression obtained is a symbolic expression that only contains input variables. the

1.4 Genetic Algorithm

Search-based testing methods formulate the problem of generating test data as an optimization problem. In search-based testing methods it is necessary to choose a way of representing candidate solutions and a method of evaluating candidate solutions. Candidate solutions are usually expressed in binary, integer and real codes. The method of evaluating candidate solutions is usually called the fitness function. The fitness function is a method used to measure the pros and cons of candidate solutions. It assigns a positive number to each candidate solution, and this positive number is used to evaluate how far the candidate is from an acceptable solution. Since optimization problems are usually minimization problems, candidate solutions with lower fitness values are better, and solutions with a fitness value of 0 are acceptable solutions. the

After we have chosen the encoding method and defined the fitness function, we can use the meta-heuristic search technology. The genetic algorithm is a powerful meta-heuristic technique. In the genetic algorithm, each solution is called a chromosome, which is composed of genes. The main process of the genetic algorithm is to evaluate the individuals in the population using the fitness function, select the solution with a better fitness value as the parent individual, and then generate the next generation of new individuals through crossover and mutation operations. the

The S.Kalaji method is compared with the embodiments of the present invention below:

We analyze it from three aspects. the

Referring to Fig. 2 a, it is embodiment 1 (Simple in-flight safety system EFSM-the extended state machine model of simple aircraft safety system) of the EFSM model of the present invention, is the average algebra that generates test data for the first aspect. the

Referring to Fig. 2b, it is embodiment 2 (Class II transport protocol EFSM—extended state machine model of transmission protocol) of the EFSM model of the present invention, for the second aspect is under the situation that test data cannot be generated 100%, consider two methods success rate. the

Referring to Fig. 2c, it is the embodiment 3 (Lift system EFSM-extended state machine model of elevator system) of the EFSM model of the present invention, for the third aspect is under the situation that the test data that satisfies constraint condition can not be generated at all, consider generating The average rate at which the test data violates the constraints in the path. the

Both our method and S.Kalaji's method are implemented with Genetic Algorithm and Direct Search Toolbox for Mathlab 7.0. We use real coding to represent individuals, use random uniform selection, select scalar crossover, the crossover probability is 0.8, use Gaussian variation, select the population size as 20, the initial value range of each variable is [0…100], the search terminates The condition is that the fitness value is 0 or the maximum number of generations is 1000. Perform the search 10 times for each migration path. the

Refer to Figure 3a, Figure 3b, Figure 3c, and Figure 3d; according to the migration coverage criterion, Simple in-flight safety system EFSM has a total of 20 migration paths, and the average generation of test data generated for these 20 paths is shown in Figure 3(a). the

When the two methods generate test data for paths other than the path p ² ; p ¹⁷ ; p ¹⁸ ; p ¹⁹ ; p ²⁰ ; p ²¹ , their average algebra is roughly the same, about 52. This is possible because in these paths, their constraints are simple and involve only a single variable. For such cases, the performance of both methods is the same. When generating test data for the paths p ² ; p ¹⁷ ; p ¹⁸ ; p ¹⁹ ; p ²⁰ ; p ²¹ , the average algebra of our proposed method is higher than that of S. Kalaji's method.

Neither method can successfully generate test data that satisfies path constraints every time in 10 searches, but our method has a much higher probability of success in generating test data than S.Kalaji's method. The success rate of path generation test data is shown in Fig. 3(b). the

It is possible that the constraints of these 6 paths are more complex and involve multiple variables. The method of S.Kalaji assigns a higher fitness value to individuals when they do not meet the constraints of the outer layer, but some individuals do not satisfy individual Selecting individuals whose outer constraints satisfy multiple inner constraints is likely to generate an optimal solution. Such individuals are discarded in the process of generating test data by S.Kalaji, which leads to the faster convergence of the S.Kalaji method, that is, to obtain a poorer optimal solution faster, so it cannot be generated for paths with complex constraints. Test Data. the

However, the method we propose considers the coverage of constraints by individuals, and the higher the coverage, the lower the fitness value of the individual. Our method selects individuals that are likely to generate optimal solutions, i.e. individuals that do not satisfy the outer constraints but satisfy multiple inner constraints. Therefore, the average algebra when generating test data is higher than that of the S.Kalaji method, but the success rate of generating test data is much higher than that of the S.Kalaji method. the

Similarly, according to the migration coverage criterion, Transport Protocol EFSM has a total of 12 paths, and the average number of generations to generate test data for these 12 paths is shown in Fig. 3(c). Both methods generate test data for paths other than p ⁸ with almost the same average algebra. The reason why test data cannot be generated for p ⁸ is that the path constraints of p ⁸ contain equations, which are difficult to satisfy.

There are 24 paths in Lift System EFSM, of which only two paths p1 and p2 can obtain data through genetic algorithm. For other paths, neither method can generate test data that satisfies the constraints, but the average algebra of the optimal solution generated by our method is higher than that of S.Kalaji's method. The reason for this result is the same as that in In Flight Safety System EFSM explanation of. For these paths that fail to generate test data, in order to analyze the pros and cons of the two methods, we compare them from the average constraint ratio of the generated test data violations. The average constraint ratio of S.Kalaji's method violation is around 80%, while the average constraint ratio of our method is around 10%, that is to say, the test data generated by our method satisfies all constraints other than the violation of the equation . We measure the quality of the test data by the average constraint ratio violated by the test data, and the lower the average constraint ratio violated, the higher the quality of the test data. From Figure 3e, the quality of test data generated by our method is much higher than that of S.Kalaji method. the

The basic principles and main features of the present invention and advantages of the present invention have been shown and described above. Those skilled in the industry should understand that the present invention is not limited by the above-mentioned embodiments. What are described in the above-mentioned embodiments and the description only illustrate the principle of the present invention. Without departing from the spirit and scope of the present invention, the present invention will also have Variations and improvements are possible, which fall within the scope of the claimed invention. The protection scope of the present invention is defined by the appended claims and their equivalents. the

Claims (5)

1. The path test data generation method based on EFSM model is characterized in that, comprises the following steps: Step 1. Perform symbolic execution and data flow dependency analysis on the EFSM model to obtain the path constraints of each EFSM model; Step 2. Use the genetic algorithm to generate test data that satisfies the path constraints, given the initial individual; Step 3. Evaluate the fitness value of the initial individual according to the fitness function. If the fitness value of the individual is 0, then such an initial individual is the test data that satisfies the path constraint, and the algorithm terminates; Step 4. If the fitness value in step 3 does not exceed the set maximum number of generations, two individuals with the minimum fitness value can be selected as the parent individuals in the current generation to perform crossover or mutation operations to generate a new generation of individuals, and the new generation Repeat steps 3 and 4 for one generation of individuals. 2. The path test data generation method based on EFSM model according to claim 1, characterized in that, the adaptation value in the step 4 exceeds the maximum algebra of setting, and the algorithm terminates. 3. The method for generating path test data based on EFSM model according to claim 1, characterized in that, the path constraints in the step 1 are represented by parallel IF statements. 4. The path test data generation method based on EFSM model according to claim 1, characterized in that, only input variables are included in the path constraints. 5. The path test data generation method based on EFSM model according to claim 1, characterized in that, the fitness function is the sum of the ratio of the branch distance of the initial individual and the condition in the path constraint not covered by the individual.