GUI-based testing involves exercising a program's graphical user interface to detect errors. There are several approaches to GUI-based testing, including model-based testing using event-flow or state models, capture and replay of user sessions, and manual testing. Choosing a testing approach depends on factors like the GUI technology. Tools like Marathon and Abbot can automate capturing, replaying, and checking GUI test cases to make the process more efficient.

1. GUI-based Testing
Alessandro Marchetto
Fondazione Bruno Kessler - IRST

2. 2
GUI-based Testing
• One of the practical methods commonly used to detect the
presence of errors (failures) in a computer program is to
exercise it by using its Graphical User Interface.
Our program
The output
is correct?
I1, I2, I3,
…, In, …
Expected results
= ?
Obtained results
“Inputs”
GUI

3. GUI-based Testing: again four main questions
 At which level conducting the testing?
3
 Unit
 Integration
 System
 Regression
 How to choose inputs?
 using the specifications/use cases/requirements
 using the code
 only considering the GUI (functionality, structure)
 How to identify the expected output?
 test oracles
 How good test cases are?
 when we can stop the testing activity

4. 4
GUI …. what?
 GUI as a means to use/interact with the software
systems
 GUI are nowadays almost ubiquitous, even in safety
critical systems
 Different types of device (web, pc, tablet, palm, mobile)
 GUI interacts with the underlying code by method calls
or messages
 GUI can exercise remote code
 GUI responds to user events (e.g., mouse clicks)
 GUIs are event-driven systems
 Testing GUI correctness is critical for system usability,
robustness and safety
 The whole system can be executed by means of the
GUI

5. GUI …. more formally
A GUI (Graphical User Interface) is a hierarchical,
5
graphical front end to a software system
A GUI contains graphical objects w, called widgets, each
with a set of properties p, which have discrete values
v at run-time.
At any time during the execution, the values of the
properties of each widget of a GUI define the GUI
state: {… (w, p, v), …}
A graphical event e is a state transducer, which yields the
GUI from a state S to the next state S’.

6. 6
GUI-based Testing
 Testing GUI software systems is different
from testing non-GUI software
 Non-GUI testing: suites are composed of test cases
that invoke methods of the system and catch the
return value/s;
 GUI-based testing: suites are composed of test
cases that are:
 able to recognize/identify the components of a GUI;
 able to exercise GUI events (e.g., mouse clicks);
 able to provide inputs to the GUI components (e.g., filling text
fields);
 able to test the functionality underlying a GUI set of components;
 able to check the GUI representations to see if they are
consistent with the expected ones;
 often, strongly dependent on the used technology;

7. 7
GUI testing difficulties
 GUI test automation is difficult
 Often GUI test automation is technology-dependent
 Observing and trace GUI states is difficult
 UI state explosion problem
 A lot of possible states of the GUI
 Controlling GUI events is difficult
 Explosion of the possible combinations of events to do
the same thing
 GUI test maintenance is hard and costly
 ……

8. 8
GUI testing advantages
 Automation is feasible
 Several frameworks and tools support it
 Easy to conduct for non-expert people
 It is funny to do
 ……

9. Which type of GUI-based testing?
9
 System testing
 Test the whole system
 Acceptance testing
 Accept the system
 Regression testing
 Test the system w.r.t. changes

10. 10
GUI-based Acceptance Testing
Acceptance Tests are specified by the customer and analyst to test that the
overall system is functioning as required (Do developers build the right
system?).
How?
 Manual Acceptance testing. User exercises the system
manually using his creativity.
 Acceptance testing with “GUI Test Drivers” (at the GUI level).
These tools help the developer do functional/acceptance testing
through a user interface such as a native GUI or web interface.
 Table-based acceptance testing. Starting from a user story (or
use case or textual requirement), the customer enters in a table
the expectations of the program’s behavior.
 Black-Box (aka functional) approaches can be used to define
test specification then executed manually, by means of the GUI or
by table-based testing.

11. Approaches for GUI-based testing
11
 Manual based
 Based on the domain and application knowledge of the tester
 Capture and Replay
 Based on capture and replay of user sessions
 Model-based testing
 Based on the execution of user sessions selected from a
model of the GUI
 Which type of model to use?
 Event-based model
 State-based model
 Domain model
 How do obtain the model to be used?
 Specification-based model
 Model recovered from existing software systems
 Log-based model

12. Coverage criteria for GUI-based testing
 Conventional code-based coverage cannot be
12
adequate;
 GUIs are implemented in terms of event-based
system, hence, the abstraction level is different w.r.t.
the conventional system code. So mapping between
GUI events and system code can not be so easy.
 Possible coverage criteria:
 Event-coverage: all events of the GUI need to be executed at
least once
 State-coverage: “all states” of the GUI need to be exercised
at least once
 Functionality-coverage: .. using a functional point of view

13. Event-based Model
Model the space of GUI event interactions as a graph
Given a GUI:
1. create a graph model of all the possible sequences
that a user can execute
2. use the model to generate event sequences
13

14. 14
Event-based Model
“Event-flow graph”
File
New Open Save …
Edit
Cut Copy Paste Replace Undo …
Editbox0 Editbox1 Match case
Find next Replace Replace all Cancel
Top level
TC: <S0, event1, event2, …>
Oracle: <State1, State2, …> & !CRASH

15. 15
Event-based Model
Model Type:
 Complete event-model
 Partial event-model
Event types:
 Structural events (Edit, Replace)
 Termination events (Ok, cancel)
 System interaction events (Editbox0, Find next)
Coverage criteria
 Event coverage
 Event coverage according the exercised functionality
 Coverage of semantically interactive events
 2-way, 3-way coverage
 ….

16. 16
State-based Model
Model the space of GUI event interactions as a state
model, e.g., by using a finite state machine (FSM):
• States are screenshot/representation of the GUI
• Transitions are GUI events that change the GUI state
Given a GUI:
1. create a FSM of the possible sequences that a user
can execute, considering the GUI state
2. use the FSM to generate event sequences
e1
e2
e3
e4 e5

17. 17
State-based Model
<ok>
FSM:
- State: GUI
screenshot
-Transition: event
that changes the GUI
state
<fill + ok>
<click>
<click>
<click>
<fill>
<fill>
<ok>
<fill + ok>

18. 18
..example
Problem: state explosion!
 Use of a state abstraction function
It maps concrete states into abstract states
(i.e., sets of concrete states)

19. 19
Log-based Recovered Model
 How do obtain the model?
 starting from system specification or
requirements
 starting from the system (i.e., reverse
engineering)
1. trace some system executions (at method calls
level)
2. infer a model
3. refine it manually, if needed

20. Test oracles for GUI-based testing
 It could be difficult to detect faults looking the GUI
20
 Crash testing is often used;
 In a GUI test case, an incorrect GUI states can take the
user to an unexpected/wrong interface screen or it
can make the user unable to do a specific action;
 e.g., after the click of a button, we try to click the button again
but we fail since the button no longer exists, after the first click.
 A GUI state can be “represented” by the components
expected to be part of the GUI in a give time and their
state/value
 e.g., window position, GUI objects, GUI title, GUI content, GUI
screenshots,

21. 21
GUI errors: examples
 Incorrect functioning
 Missing commands (e.g., GUI events)
 Incorrect GUI screenshots/states
 The absence of mandatory UI components (e.g., text
fields and buttons)
 Incorrect default values for fields or UI objects
 Data validation errors
 Incorrect messages to the user, after errors
 Wrong UI construction
 ….

22. 22
GUI-based Testing: process
1. Identify the testing objective by defining a coverage
criteria
2. Generate test cases from GUI structure, specification,
model
 Generate sequences of GUI events
 Complete them with inputs and expected oracles
3. Define executable test cases
4. Run them and check the results

23. 23
GUI-based Regression Testing
 GUI-based testing means to execute the GUI of a
system exercising its GUI components;
 A small changes in the GUI layout can make the GUI
test cases old and useless;
 Hence, GUI-based test suite need to be maintained
and often chagned
 supporting tools are welcomed
 Often, GUIs are realized by means of rapid prototyping
or automatic framework. This requires an efficient
approach to generate and maintain GUI test suite
 supporting tools are welcomed

24. Capture and Replay
A capture and replay testing tool captures user
sessions (user inputs and events) and store
them in scripts (one per session) suitable
to be used to replay the user session.
An ad-hoc infrastructure is needed to intercept GUI events,
GUI states, thus storing user sessions and also to be
able to replay them.
- they can work at application or VM level
24

25. 25
Recorded information
 Inputs, outputs, and other information needed
to replay a user session need to be recorded
during the capture process.
 Examples:
 General information: date/time of recording, etc.
 System start-up information
 Events from test tool to system
 Point of control, event
 Events from system to test tool
 Checkpoints / expected outputs
 Time stamps

26. 26
Capture and Replay: the process
1. The tester interacts with the system GUI to run the
system, thus generating sessions of sequence of
mouse clicks, UI and keyboard events;
2. The tool captures and stores the user events and the
GUI screenshots;
 a script is produced per each user session
3. The tester can automatically replay the execution by
running the script
 the script can be also changed by the tester
 the script can be enriched with expected output,
checkpoints
 the script can be replicated to generate many variants (e.g.,
changing the input values)
4. In case of GUI changes, the script must be updated

27. 27
Tools for GUI-based testing
 Marathon
 Abbot
Java
 Guitar
 HtmlUnit, HttpUnit,JWebUnit
 HtmlFixture
 Selenium
 ….
Web

28. 28
Running Example: Calculator
Calc_1 :
 Logic mixed to GUI
 GUI realized by using
Swing
Calc_2 :
 Logic mixed to GUI
 GUI realized by using AWT
Calc_3 :
 Logic separated from the
GUI
 GUI realized by using AWT
+Swing
These apps are from the net hence please refer to the author’s copyright

29. Marathon
 Is a tool that helps writing System/GUI tests
29
for Java/Swing applications
 Consists of a recorder, player, and an editor
 Records tests script in Python
 It is possible to use it without knowing Python …
 Allows to insert assertions in the script easily
 Using a specific GUI
 Shows testing results using JUnit’s control bar
 Red/green
 http://www.marathontesting.com
 download Marathon and its user manual

30. 30
Marathon example (1)
 A calculator application
was developed (Swing)
 We want to write tests
for the calculator
 Ex. add two values
 Steps:
1. create a new Marathon
project
2. create a new empty test
(a test is a Python script
file)
3. build the test (recording)

31. 31
Marathon example (2)
Record the script
• click on the record button in
the toolbar
– the red circle
• enter “12 + 4 =”, the result is
displayed (16)
• press control + right-click in
the text area, to open the
contextual menu
• select Assert Text
• stop recording (the button
with a red square), the script
now looks like:
• save the script

32. 32
Marathon example (3)
Run the script
• Select the “JUnit” tab
• Click the test view, you should
see your new created test
• Click on the “Run all tests”
button, in the JUnit’s control
bar
• and …
• get the green bar …

33. 33
Abbot
 Is a tool that helps writing System/GUI tests for Java
AWT/Swing applications
 Consists of a recorder, player, and an editor (via Costello, built
on top of Abbot)
 Records tests script in Java
 Allows to write test cases directly from Java code (named
programmatic GUI testing)
 Allows to insert assertions in the script easily
 Shows testing results using JUnit’s control bar
 Red/green
 Two main building blocks:
 ComponentReferences to get a handle on a GUI component
 Robot: to perform user-level actions on various GUI
components
 http://abbot.sourceforge.net
 download Marathon and its user manual

34. 34
Abbot example (1)
 A calculator application
was developed
(AWT/Swing)
 We want to write tests
for the calculator
 Ex. add two values
 Steps:
1. create a new empty
Abbot test
2. Initialize the test with the
application under test
information
3. build the test (recording)

35. 35
Abbot example (2)
Recording
1. Start recording of all
actions
2. Execute the software
3. Add assertions
1. select the GUI element
with SHIFT+F1
2. select the property (e.g.,
(“Text”)
3. set the expected value
4. Stop recording

36. 36
Guitar
 A testing framework implementing the “Event-flow
graph” approach
 Four main components:
 The GUIRipper: extract GUI information from a program
 The GUIStructure2Graph: build a traversable graph
representation of the GUI elements
 The TestCaseGenerator: create an extensive set of test
cases based on the graph
 The GUIReplayer: run the program as instructed by these
tests
 It is a research tool…
 http://guitar.cs.umd.edu
 It requires ANT to be executed

37. 37
Guitar (1)
Four main components:
(1) GUIRipper: GUI information extraction
>ant -Dproperties=jfcripper.properties -f jfcripper.xml
(2) GUIStructure2Graph: Event-flow graph inference
>ant -v -f GUIStructure2GraphConvert.xml

38. Guitar (2)
(3) TestCaseGenerator: Test cases generation traversing the graph
38
>ant -v -f TestCaseGenerator.xml
(4) GUIReplayer: Run the program GUI
according to the generated test cases
>ant -Dproperties=jfcreplayer.properties -f jfcreplayer.xml

39. Tools for other GUI-based testing of Java
 HtmlUnit, HttpUnit, JWebUnit
 HtmlFixture
 Selenium
 ….
39

40. 40
XUnit Web tools
HtmlUnit, HTTPUnit, JWebUnit, etc.
- They are inspired by JUnit
- They let us to verify the functionalities of Web sites
- They emulate the relevant portions of browser behavior
form submission, JavaScript, http authentication, cookies, automatic page
redirection, etc.
- They allow Java test code to examine returned pages
in terms of text, DOM, or containers of forms, tables, and links
- They make it easy to write Junit tests that verify HTTP requests
and responses

41. Send the request
41
… HttpUnit example
import junit.framework.TestCase;
import com.meterware.httpunit.WebResponse;
import com.meterware.httpunit.WebConversation;
import com.meterware.httpunit.WebForm;
import com.meterware.httpunit.WebRequest;
public class SearchExample extends TestCase {
public void testSearch() throws Exception {
WebConversation wc = new WebConversation();
WebResponse resp = wc.getResponse("http://www.google.com");
WebForm form = resp.getForms()[0];
form.setParameter("q", "HttpUnit");
WebRequest req = form.getRequest("btnG");
resp = wc.getResponse(req);
assertNotNull(resp.getLinkWith("HttpUnit"));
resp = resp.getLinkWith("HttpUnit").click();
assertEquals(resp.getTitle(), "HttpUnit");
assertNotNull(resp.getLinkWith("User's Manual"));
}
}
Set the google page
Fill the google form
get the google response
verify the response

42. HtmlFixture
http://fitnesse.org/FitNesse.HtmlFixture
- It is used to exercise and test web pages.
- It allows us make assertions about the structure of a page and to
navigate between pages.
- We can also fire java script, submit forms, "click" links, etc.
- It is based on a fixture that lets us navigate to any element of
the
page DOM.
To use this we need some knowledge about the DOM (Document
Object Model).
42

43. 43
… example

44. 44
Capture and Replay for Web
Selenium
http://seleniumhq.org/
Selenium IDE: a Firefox add-on
that will do simple record-and-
playback of interactions
with the browser
Selenium WebDriver: a
collection of language specific
bindings to drive a browser --
the way it is meant to be
driven.

45. Additional references
-Atif M. Memon. An event-flow model of GUI-based applications for testing.
Software Testing, Verification and Reliability, 2007, John Wiley & Sons
http://www.cs.umd.edu/~atif/papers/MemonSTVR2007-abstract.html
-Atif M. Memon, M.E. Pollack, Mary Lou Soffa. A planning-based Approach to GUI
testing. International Software/Internet Quality Week, May 2000
http://www.cs.umd.edu/~atif/papers/MemonSQW2000-abstract.html
-Marlon Vieira, Johanne Leduc, Bill Hasling, Rajesh Subramanyan, Juergen
Kazmeier. Automation of GUI Testing Using a Model-driven Approach AST’06
http://www.irisa.fr/lande/lande/icse-proceedings/ast/p9.pdf
-Jessica Chen and Suganthan Subramaniam. A GUI Environment to Manipulate
FSMs for Testing GUI-based Applications in Java. International Conference on
System Sciences, 2001
http://wenku.baidu.com/view/5f544c5377232f60ddcca145.html?from=related
-Alessandro Marchetto, Paolo Tonella and Filippo Ricca. State-based testing of
Ajax Applications. In Proceedings of the International Conference on Software
Testing, 2008 http://selab.fbk.eu/tonella/papers/icst2008.pdf
45