SYSMODIS

A research presentation from International Workshop on Combinatorial Testing (IWCT'21) organized by International Conference on Software Testing (ICST'21)

Transcript

1. SYSMODIS SYSTEMATIC MODEL DISCOVERY APPROACH Omer Korkmaz and Cemal Yilmaz 10th International Workshop on Combinatorial Testing (IWCT 2021)
2. AGENDA 2 01 Introduction 02 Approach 03 Experiments 04 Conclusion & Future Works
3. INTRODUCTION 3 1
4. 4 In today’s technology, mobile devices and applications have increasingly become smarter and more powerful. Started to be more complex In daily life, people are using mobile applications from various categories • education, • health, • economy, • or management Used by millions of people While providing solutions for the demands of the users, the mobile apps started to have huge, complex and interactive logics and functionalities. 2.560.000 Android Apps https://www.statista.com/statistics/276623/number-of-apps-available-in-leading-app-stores/
5. OPS, FAILURES! ▰ The complexity may cause more failures in the applications. ▻ lack of business knowledge, wrong implementation, etc. ▰ Users are exposed to those failures many times using the mobile apps. 5 So, these applications need to be tested thoroughly.
6. 6 The model of an application plays a significant role to represent the test strategies while the system is under test (SUT).
7. 7 SYSMODIS Crawl the screens by interacting with UI, generate test values for the inputs, discover the model and execute the tests systematically and automatically. Discover likely guard-conditions systematically with the UI interactions, leveraging machine learning approaches for predicting guard-conditions MODEL DISCOVERY PREDICTION Achieved %30 more code coverage when compared to other approaches. Achieved high accuracy on predictions when compared to random testing. RESULTS COVERING ARRAYS → SYSTEMATIC SAMPLING → MODEL DISCOVERY CONTRIBUTION
8. 8 WHAT IS MODEL? A state machine S1 S2 Agreement = unchecked ******************** 1010101010101
9. APPROACH 9 2
10. 10 SYSMODIS OVERVIEW
11. 11 SCREEN DETECTION
12. 12 SCREEN DETECTION Screen Detection • Get the XML file of current Android screen. • Catch the Android elements from the XML. • Take the attributes of each element. • Hash the screen elements in an ordered agnostic way • Check the distinctness comparing hash values! • Store all information in the database.
13. 13 INPUT DETECTION
14. 14 INPUT DETECTION Input Detection • Get each element from the current screen. • Get the actionable attributes of each element. • Check the attributes and determine the input type of each element. • For instance; ◦ GLN, Username, Password > Editable ◦ Agreement > Checkable ◦ Register, Login > Clickable
15. 15 DOMAIN DETECTION
16. 16 DOMAIN DETECTION keywords mail e-mail username ... EMAIL DOMAIN EQUIVALENCE CLASSES Valid Email omer@gmail.com Invalid Email qy@1?1.com ATTRIBUTES content-desc “type your email” text “please enter your email” resource-id com.farmazon.id:/emailText Semantic Similarity?
17. 17 CA GENERATION
18. 18 CA GENERATION
19. 19 CA GENERATION Covering Array (t:3) Each row => Test case Each cell => Test action
20. 20 CRAWLING ▰ Opportunistic crawling ▻ Screen is visited → previously untested test case executed ▻ Test suite executed → move nearest state with untested test cases and execute ▰ Continues to execute tests where it left off ▻ In case of any failure, system exception, etc. ▰ Configured to restart the system under test ▻ Preventing crawling from getting stuck as much as possible
21. 21 GC DISCOVERY
22. 22 FOR EACH TARGET STATE ITSELF OTHERS 1 0 BINARY CLASSIFICATION PREDICTED GUARD-CONDITION S2: Agreement Error Screen, S3: Login Screen, S4: Other Errors Screen S2 Agreement: unchecked ... S3 S4 EXAMPLE GC DISCOVERY
23. EXPERIMENTS 23 3
24. 24 EXPERIMENTS Evaluating sensitivity to model parameters Evaluated the sensitivity of the approach to various model parameters, including the number of states, density, the level of determinism, and the complexity of the guard conditions. To this end, used simulations as it was not possible to systematically vary these parameters on real subject apps. Evaluations on Subject Applications Evaluated the proposed approach by conducting comparative studies using real subject applications. Also, applied random-testing with the proposed approach and compared it with other approaches. Study 2 Study 1
25. 25 STUDY 1 • states: the number of states in the model. • density: the density of the model which is used to compute the number of transitions in the model. • parameters: the number of parameters defined in a state, i.e., the number of input fields on a screen. • settings: the number of equivalence classes for a parameter. • guard-complexity: the number of distinct parameters involved in a guard condition associated with a transition. • t: the coverage strength of the covering arrays used for sampling. • determinism: the level of determinism in the model, depicting the probability of taking a transition given that the guard condition of the transition is satisfied. When determinism = 1.0, all the transitions are deterministic given a transition, when the system is currently in the source state and the guard condition of the transition is satisfied, the transition is guaranteed to be taken and the system moves to the target state. FOR EACH CONFIG 100 STATE MACHINES
26. 26 STUDY 1: EVALUATION percentage of the transitions satisfied accuracy of the guard conditions predicted 1 2 3 percentage of the states visited State Coverage Transition Coverage Accuracy The classification models were trained and the classifier was taken from scikit-learn Covering arrays were generated. 4 5 6 I7 6700HQ, 16 GB RAM, Windows 10 Machine Decision Tree Classifier ACTS
27. 27 STUDY 1: ANALYSIS
28. 28 STUDY 2: EVALUATION • screen: page of Android mobile applications. It might be an Android activity or a different page in the same activity (e.g., pop-up, modal). Each page which consists of UI elements is called as screen. • test action: one of the executable tests in test suites. For example, if we have a test suite that includes 3 executable tests, each of them is called as test action. percentage of the source code statements visited 1 2 percentage of the screens visited Screen Coverage Code/Line Coverage covering arrays were generated code coverage was measured 3 4 5 classification models were trained Decision-Tree Classifier ACTS ACVTOOL
29. 29 STUDY 2: ANALYSIS
30. CONCLUSION & FUTURE WORKS 30 4
31. 31 CONCLUSION & FUTURE To sum up briefly... Detects: - screens - inputs - domains and types of inputs Generates convenient test values by using covering arrays - way to build a model - test all states Predicts guard-conditions - Decision-tree classifier Higher code coverage than: - Dynodroid - Monkey - Random-testing Showed the relationship between factors that affects the prediction and coverage - strength (t) - determinism, etc. Comparison Systematic Sampling Prediction Simulations Automated Approach Take into account not only the interactions within a state but also the interactions across the states. Enhance the proposed approach to test iOS-based applications Feedback-Driven Crawling iOS Environment Apply proposed approach to a large number of Android applications. Large Dataset
32. 32 THANKS! Any questions? omerkorkmaz@alumni.sabanciuniv.edu cyilmaz@sabanciuniv.edu