System and method for automated testing of software systems utilizing statistical models

[0002] A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent disclosure, as it appears in the Patent and Trademark Office patent files or records, but otherwise reserves all copyright rights whatsoever.

[0003] The present invention involves a method and apparatus for automated software testing. More specifically the present invention provides the Tester with a method and system to describe the software under test without requiring the Tester to know all the details of the software.

[0004] Automated software testing can be defined as a program that automatically enters a predetermined set of characters or user commands in order to test new or modified versions of software applications.

[0005] Historically it has been observed that there are some serious problems with software automation test case testing tools. One of the key issues is the use of parameters by automation tools. Parameters must be known in advance. As the number of parameters increase, the shear numbers limit what can reasonably be tested.

[0006] If test tools can support differing languages, each have a unique set of parameters. If parameters change from one version to another, each different set of parameters requires modification to the system under test and/or the test tool's parameter handling.

[0007] Another key issue is that automated tests are likely to be static (i.e. they are primarily derived from manual test scripts). Therefore, to test new or changed functions, the test case must be revised. This is also true if a new or changed parameter is added or revised, respectively. This results in high maintenance.

[0008] Yet another key issue is that the function to be tested must exist in some detail for automated test cases to be written. The implication of this is that it is difficult to debug an incomplete function. The coding of the test cases delay the ultimate execution and completion of the testing. In addition, when a new defect is accidentally found outside the testing space, a new test case is required.

[0009] The present invention will be better understood and its numerous objects and advantages will become more apparent to those skilled in the art by reference to the following drawings, in conjunction with the accompanying specification.

[0010] Currently, it is almost an impossible task to test all the various combinations of parameters within a reasonable time frame. There is no way you can test all parameters, any more than you can test all the paths in White Box Testing. It is NP-complete (from a mathematical point of view, it is believed to be a solvable problem; however, it would take an almost infinite length of time to solve), and even if one does not need to test parameters, one must test them because today's tools need them to work.

[0011] Current automated test cases are acceptable for regression testing, but are considered unacceptable for nonregression testing (e.g. development systems). The maintenance as noted is too high. One changed parameter and the test case must be revised. If the vendor changes the tool, it is likely that the test case must be revised. The maintenance of test cases on a developing system is very high. The user must know what all the parameters are and the function they support before they can begin coding the test case.

[0012] It would therefore be a distinct advantage to have a method and system that would eliminate many of the above recited problems and concerns. The present invention provides such a method and system.

[0013] The present invention, Smart Testing Method For Automated Software Testing, resolves the problems outlined above. Essentially this involves Goal oriented testing that queries the environment and takes action based on facts and a model of the system under test. The action can be modified by rules and exceptions, including training testing to focus on certain behaviors. Regarding the key issues outlined above, the use of parameters can be eliminated. Regarding the issue of automated tests being static and primarily derived from manual static test scripts, the instant system is dynamic quasi-random and can be trained to perform ever changing tasks including classic static emulation. Where, as noted in the prior art, functions to be tested must exist in some detail for automated tests to be written, in the instant system, few functional details are necessary. The instant system can run on an incomplete model.

[0014] Test case automation efforts have essentially involved taking a manual script and automating it. To be successful, a Tester must have “up front” knowledge of the details of what they are going to automate. For example, if the Tester wants to automatically select the “Save” menu item, he has to know its in the first drop-down menu for “File” on the main menu bar, and, also, either know its drop-down position or its name. Applicants, in the present invention, have been successful only because their application efforts to date have involved testing existing functions, better known as regression testing.

[0015] Assume, for a moment that the “Save” menu item is a new function. All the Tester knows is that there is a requirement for a file save menu item function and if the function succeeds, a file that previously existed will be rewritten with a new date and/or time. If the function fails a pop up window with the word “error” will appear. In the manual testing paradigm this new function would likely have been tested using “gorilla testing”. It is, as implied, a random “hunt and pick” manual method of testing, but is not repeatable.

[0016] One solution to the test automation problem would be for the Tester to know in advance all the design details. But chances are rare that would happen with any degree of accuracy required to automate the test. Another solution is, thus, proposed in the present invention.

[0017] The present Smart Tester invention gives the Tester a method to describe the system under test without having to know all the details of the system.

[0018] The architecture of the Smart Test is described as follows. Instead of writing a script, the Tester designs a functional model of the system to be tested. The system could be anything from a single application to a LAN environment. Assume for a moment that the system is an application. The Tester would, for example, model a main window, a menu bar, drop-down menus, specialized windows, etc. The model can be incomplete; however, what is complete must be accurate. For example, if there is no menu bar, the model mentioned above is inaccurate and will not work. The more complete the model, the better the chance the testing will cover existing functions.

[0019] The Tester would also define any facts that the model might need; for example, the name of the file to be opened and saved. The Tester then defines any goals or subgoals to be attained. For example, a goal might be saving the file. Then the rules under which the model will operate are defined. One rule might be “if the file (named as a fact) has its date and/or time changed, then the goal of saving the file was reached and the test will end.

[0020] The present invention will be better understood and its numerous objects and advantages will become more apparent to those skilled in the art by reference to the following drawings, in conjunction with the accompanying specification, in which:

[0021] FIG. 1 represents the flow diagram for the Architecture for a Smart Test, the basis for the present invention;

[0022] FIG. 2 shows an empty desktop, or a state of the system when the application is not executing;

[0023] FIG. 3 shows the application after it has been started. Note the window named A and that it has a menu bar with menu items B and C;

[0024] FIG. 4 shows the application when the menu item B is focused. Note the drop-down menu with items D and E;

[0025] FIG. 5 shows the application when the menu item C is focused. Note the absence of a drop-down menu;

[0026] FIG. 6 shows the application when item D is focused. Note the absence of an add on menu;

[0027] FIG. 7 shows the application when menu item E is focused. Note the absence of an add on menu;

[0028] FIG. 8 shows the result of selecting menu item C, window F appears;

[0029] FIG. 9 depicts the selection of the respective “OK” button. When all “OK” buttons have been selected, the application should revert to the state depicted in FIG. 3;

[0030] FIG. 10 shows the result of selecting menu item D, window G appears;

[0031] FIG. 11 depicts the selection of the respective “OK” button. When all “OK” buttons have been selected, the application should revert to the state depicted in FIG. 3;

[0032] FIG. 12 shows the result of selecting menu item D, window H appears. Windows F, G and H will remain on the desktop until their respective “OK” buttons are selected. When the “OK” button is selected, the suitable window will disappear; and

[0033] FIG. 13 depicts the selection of the respective “OK” button. When all “OK” buttons have been selected, the application should revert to the state depicted in FIG. 3.

[0034] There are two ways a Smart Test can be initiated. First is to take some quasi-random action as defined by the model and then query the result. An alternate way is to first query the system under test. In either case, as soon as the query is completed, a rule engine, whose job it is to modify the behavior of the model, applies rules until either another action can be taken or the test can be completed. The model must run forever if rules do not exist or can never be applied.

[0035] This implies that some attainable goal must exist to stop the test. In the case where function is particularly buggy, this should probably be some time limit. Also, goals that are difficult to attain because they represent deep seated or uncommon objectives can be NP-Complete. This is another reason for time limits. However, the problem with deep seated goals can be reduced by the use of sub or interim goals and controlling restart states in the model.

[0036] Failures are logged as required. If an “action is taken”, it is first checked for exceptions. Exceptions are those things that might destroy future testing, for example, rebooting the machine under test.

[0037] Taking quasi-random actions means two things. First, and most important, is that the test can be repeated if necessary, and, second, that the actions can be taken under defined distributions. For example, assume one knows that the average user saves the file ten times for each printing of the file. One can set up a ten to one targeted bin distribution for save versus print; or, one can set up a Gaussian distribution, uniform distribution or whatever is required.

[0038] A trace log is created so that history is preserved. It is also created for future enhancements, such as learning the paths to goals, so that future executions in a regression test mode do not have to “guess” their way toward a goal. They can go there directly.

[0039] The facts, goals, rules and models are application dependant and would be designed by the Tester. Initialization, actions, queries, exception handling, logging and the rules engines are reusable by the Tester.

[0040] The Testing system is defined in the flowchart, FIG. 1. It should be noted in the Functional Model of System under Test 101, if one initially ignores the goals and subgoals 103, 102 the facts, 113 the exceptions and 112 end of test, you have a single rule that always passes, and assumes that a failure will never stop the test. The execution of the functional model System under test 107 begins by initializing the test 100. Initializing is doing such things as placing the date and time in logs. Then, depending on the system under test 110 and the model of the system under test 101, the execution can either take some quasi-random action 105 or by querying the system under test 106. The choice will most likely be based on the state of the system under test 110. If the state is known (for example, an unexecuted software application), then the quasi-random action to execute some application 105 would be taken first. However, if the state of the system under test is unknown (for example, beginning a test on an execution LAN system), then the query of the LAN system would be appropriate before taking any quasi-random action 105, so 106 would occur first.

[0041] Assume that as an example some quasi-random action 105 is taken first. The next action would be logged 109 and taken on the system under test 110. The action would be something like “press the left mouse button on the ‘OK’ button in application XYZ” or “stop execution of task X on the XYZ LAN”. After the action has taken hold (determined by the model), the model of the system under test 101 would query 106, the system under test 110, through an implied path 114 in the model, to determine the effects of the action. Since the rules always pass 107 and failure never stops, a decision will be made 108 to either log a failure 111 and continue another action, or just continue another action where no failure occurs. The test will run endlessly through the path 105, 109, 110-114, 106, 110, 106, 107, 108, 105, 109, 110- . . . or 106, 110, 106, 107, 108, 105, 109, 110-114, 106, 110, 106, . . . in case the query occurred prior to the action.

[0042] If we assume there are actions in a test that the user does not wish to take (for example, actions that are destructive), exception 113 detects such actions as defined by the Tester as a special fact 102 and will purposely fail to execute them each time they occur. Facts 102 are simple declarations such as the name of the button ‘OK’ or the name of a server on a LAN such as ‘ServerOne’. Goals 103 are what the Tester is trying to accomplish (for example, find a button called ‘OK’, or find a message on the LAN with the text “This is a test”). Once goals 103, facts 102 and the result of a query 106 are available, rules 107 can be made to change the behavior of the model 101. For example, one such rule could be “when a goal 103 is reached, then end the test at 112. At this point, the test will stop executing.

[0043] Regarding the dotted line 114 between “Take Quasi-Random Action 105 and “Query System Under Test” 106, the following is noted. Immediately after taking 105, 114 the dotted line begins. Depending on the model 101, the dotted line 114 can take one of three forms. In the first case, immediately query System Under Test 106, in the second case, Query 106 once after some predefined delay, or in the third case, query 106 X number of times with some small delay between queries. The normal expectation is that an action 105, performed on the system under test 110, will cause some change in the state of 110 and the query 106 will detect it. For example, if the left mouse button is down (the action) on the ‘File’ portion of a word processor's menu bar, after a small period of time one would expect to see a drop-down menu with the selection ‘Save’ (the result of the query). Normally, if the state of the system under test 110 is unchanged, it indicates a failure. Therefore, the result of the query does not indicate the presence of a drop-down menu with the selection ‘Save’.

[0044] It cannot be said that it will always be the case that an action is expected to produce some change in state. There are tests where an action is taken and the expectation is that nothing will happen. If something does happen, then it is viewed as a failure. The apply Rules 107 sorts out which query at 106 results in a failure 111 and which does not.

[0045] The following “Example” demonstrates how the present invention can be used to test the system.

[0046] A. Assumptions

[0047] 1. This testing method is used on the application depicted in FIGS. 2 through 13.

[0048] 2. None of the application's parameters will be used by the Tester in designing the test.

[0049] 3. The test is to verify that window H functions are required.

[0050] 4. The Tester is responsible for the following test definition:

[0051] a. the model (FIG. 1 at 101) of the application (FIG. 1 at 110)

[0052] b. the goals or optional subgoals (FIG. 1 at 103)

[0053] c. the facts (FIG. 1 at 102) required, including any exceptions (FIG. 1 at 113)

[0054] d. the rules (FIG. 1 at 107)

[0055] 5. The application defines the limits of the testing space.

[0056] 6. The operating system in which the application executes is outside the testing space.

[0057] 7. The Smart Test software that manipulates the Tester's test is outside the testing space.

[0058] 8. The quasi-random selections are based on a uniform distribution.

[0059] B. Explanation of the Applications Depicted in FIGS. 2 Through 13

[0060] FIG. 2 shows an empty desktop, or a state of the system when the application is not executing.

[0061] FIG. 3 shows the application after it has been started. Note that the window is named A at 300 and that it has a menu bar 301 with menu items B and C.

[0062] FIG. 4 shows the application when the menu item B at 400 is focused. Note a drop-down menu 401 with items D and E.

[0063] FIG. 5 shows window A at 501 and shows the application when the menu item C at 500 is focused. Note the absence of a drop-down menu.

[0064] FIG. 6 shows window A at 601 and shows the application when item D is focused at 600. Note the absence of an add on menu.

[0065] FIG. 7 shows window A at 701 and shows and the application when menu item E at 700 is focused. Note the absence of an add on menu.

[0066] FIG. 8 shows window A at 802 and the result of selecting menu item C at 800, window F appears 801.

[0067] FIG. 10 shows the result of selecting menu item D. window G appears at 100.

[0068] FIG. 12 shows the result of selecting menu item D, window H appears at 120 and the “OK” button at 121. Windows F, G and H will remain on the desktop until their respective “OK” button is selected. When the “OK” button is selected, the suitable window will disappear.

[0069] FIG. 9 at 900, FIG. 11 at 110 and FIG. 13 at 130 depict the selection of the respective “OK” button, window H is at 131. When all “OK” buttons have been selected, the application should revert to the state depicted in FIG. 3.

[0070] A Model of the application in herarchical tops down order would be:

[0071] 1. a main window

[0072] 2. a menu bar

[0073] 3a. a window or

[0074] 3b. drop-down menu

[0075] 4a. a button

[0076] 4b. a window

[0077] 5b. a button

[0078] This model would select menu items until a window appears, then it will close the window by selecting the button.

[0079] How the test will be run without the Tester's use of parameters:

[0080] First, the application to be tested will be placed in a known directory on an operating system. This allows for starting the application without the Tester having to know it's name. Both the operating system and the Smart Test's Initialize (FIG. 1 at 100) function are outside the testing space, so Initialize can look into the directory, get the application's name and execute it. This action would result in FIG. 2 changing to FIG. 3.

[0081] Second, the test if successful or capable of continuing execution with failure *, requires a rule that if met will stop the test. The rule will be constructed statistically. Under uniform distribution, when FIG. 3 is present, menu item B will be selected on average 50% of the time and menu item C will be selected 50% of the time. If menu item B is to be selected (FIG. 4), then on average menu item D and menu item E will be selected 50% of the time, or overall, menu item D and menu item E will be selected 50% times 50% or on average 25% of the time. Menu item E is the one we are interested in in our test because it results in window H (FIG. 12). We can, therefore, assume on average that every fourth window that appears with an “OK” button will be window H. We need only one occurrence to do our test. Assuming a Poisson distribution and running the test for only four occurrences of the “OK” button, then stopping will result in window H being tested only 50% of the time. Assuming a Poisson distribution and running the test for ten occurrences of the “OK” button, then stopping will result in window H being tested 99.8% of the time. We'll assume that 99% or greater is acceptable.

[0082] * designates the case where a failure prohibits “OK” buttons from appearing is ignored, as well as traps, hangs and internal processing errors.

[0083] The facts* would be:

[0084] 1. the failure log=Failure log

[0085] 2. the trace log=Trace log

[0086] 3. the path to the application=d:\test\application

[0087] 4. the number of buttons required=0

[0088] 5. Initialization starts the application

[0089] * there are no exceptions in this example, an exception might exist if one of the menu items or buttons caused the system to shutdown

[0090] The goals* would be:

[0091] 1. find the number of buttons required

[0092] * there are no subgoals in this example, a subgoal might be used to assure an event occurs before the main goal is reached (for example, finding window F before testing for window H).

[0093] The rules, FIG. 1 at 107, hierarchical tops down, would be:

[0094] 1. if the desktop does not contain the main window of the application under test, then log a failure and stop the test

[0095] 2. if a window, then take no action

[0096] 3. if the goal is met, then stop the test

[0097] 4. if the button is pressed and another button exists, then log a failure

[0098] 5. if buttons exist, then select a button and increase the button count by one

[0099] 6. if menu items exist, then select one menu item

[0100] The test would run as follows, assuming no failures:

[0101] Initialization assigns the trace.log and failure.log as defined facts. It starts the application under test. The model, FIG. 1 at 101, takes over control of the test. It queries the desktop for the A window. It tests the rules, FIG. 1 at 101, and rule 2, noted above, applies a null action be taken. The model, FIG. 1 at 101, dictates that the application's menu bar is to be queried for it's content. It returns B & C, tests the rules and 6 applies. It then quasi-randomly selects either B or C. The model then indicates that either a window or a drop-down menu will occur. The application is queried for a window first. If F exists, then test the rules and 5 applies. At this point, on this path the model returns to the beginning and runs again. If F does not exist, then a drop-down menu is queried for menu items. It will return D & E. It will test rules and 6 applies. It will then quasi-randomly select D or E. The model then indicates that a window will appear. The window is queried and it is either G or H. Test the rules and 2 applies. A null action is taken. The model then indicates buttons. It returns the “OK” button. Test the rules and 5 applies. At this point, on this path the model returns to the beginning and runs again. After rule 5 has been applied ten times, the goal is met and rule 3 applies. As we assumed no failure for this run, either rule 1 or rule 4 fire. Rule 1 is to cover the case where the application didn't start and rule 4 for the case where the F, G or H windows failed to close.

[0102] The test is verified by interrogating the trace and failure logs. The trace log should contain at least one entry of the H window being found. The failure log should contain no failures. Failures are produced from both rules and queries. For example, if the model indicates that buttons should exist and none are found then the query will return a failure.

[0103] The following Program Code, written REXX language, demonstrates how the present invention is used for Automated Software Testing.

[0104] Although certain preferred embodiments have been shown and described, it will be understood that many changes and modifications may be made therein without departing from the scope and intent of the appended claims.