Unit Testing

As of November 2007 we require all new features going into master to be accompanied with a unit test. Initially we have limited this requirement to qgis_core, and we will extend this requirement to other parts of the code base once people are familiar with the procedures for unit testing explained in the sections that follow.

The QGIS testing framework - an overview

Unit testing is carried out using a combination of QTestLib (the Qt testing library) and CTest (a framework for compiling and running tests as part of the CMake build process). Lets take an overview of the process before I delve into the details:

  • There is some code you want to test, e.g. a class or function. Extreme programming advocates suggest that the code should not even be written yet when you start building your tests, and then as you implement your code you can immediately validate each new functional part you add with your test. In practive you will probably need to write tests for pre-existing code in QGIS since we are starting with a testing framework well after much application logic has already been implemented.
  • You create a unit test. This happens under <QGIS Source Dir>/tests/src/core in the case of the core lib. The test is basically a client that creates an instance of a class and calls some methods on that class. It will check the return from each method to make sure it matches the expected value. If any one of the calls fails, the unit will fail.
  • You include QtTestLib macros in your test class. This macro is processed by the Qt meta object compiler (moc) and expands your test class into a runnable application.
  • You add a section to the CMakeLists.txt in your tests directory that will build your test.
  • You ensure you have ENABLE_TESTING enabled in ccmake / cmakesetup. This will ensure your tests actually get compiled when you type make.
  • You optionally add test data to <QGIS Source Dir>/tests/testdata if your test is data driven (e.g. needs to load a shapefile). These test data should be as small as possible and wherever possible you should use the existing datasets already there. Your tests should never modify this data in situ, but rather may a temporary copy somewhere if needed.
  • You compile your sources and install. Do this using normal make && (sudo)  make install procedure.
  • You run your tests. This is normally done simply by doing make test after the make install step, though I will explain other aproaches that offer more fine grained control over running tests.

Right with that overview in mind, I will delve into a bit of detail. I’ve already done much of the configuration for you in CMake and other places in the source tree so all you need to do are the easy bits - writing unit tests!

Creating a unit test

Creating a unit test is easy - typically you will do this by just creating a single .cpp file (not .h file is used) and implement all your test methods as public methods that return void. I’ll use a simple test class for QgsRasterLayer throughout the section that follows to illustrate. By convention we will name our test with the same name as the class they are testing but prefixed with ‘Test’. So our test implementation goes in a file called testqgsrasterlayer.cpp and the class itself will be TestQgsRasterLayer. First we add our standard copyright banner:

  Date : Friday, Jan 27, 2015
  Copyright: (C) 2015 by Tim Sutton
  Email: [email protected]
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.

Next we use start our includes needed for the tests we plan to run. There is one special include all tests should have:

#include <QtTest/QtTest>

Beyond that you just continue implementing your class as per normal, pulling in whatever headers you may need:

//Qt includes...
#include <QObject>
#include <QString>
#include <QObject>
#include <QApplication>
#include <QFileInfo>
#include <QDir>

//qgis includes...
#include <qgsrasterlayer.h>
#include <qgsrasterbandstats.h>
#include <qgsapplication.h>

Since we are combining both class declaration and implementation in a single file the class declaration comes next. We start with our doxygen documentation. Every test case should be properly documented. We use the doxygen ingroup directive so that all the UnitTests appear as a module in the generated Doxygen documentation. After that comes a short description of the unit test and the class must inherit from QObject and include the Q_OBJECT macro.

/** \ingroup UnitTests
 * This is a unit test for the QgsRasterLayer class.

class TestQgsRasterLayer: public QObject

All our test methods are implemented as private slots. The QtTest framework will sequentially call each private slot method in the test class. There are four ‘special’ methods which if implemented will be called at the start of the unit test (initTestCase), at the end of the unit test (cleanupTestCase). Before each test method is called, the init() method will be called and after each test method is called the cleanup() method is called. These methods are handy in that they allow you to allocate and cleanup resources prior to running each test, and the test unit as a whole.

private slots:
  // will be called before the first testfunction is executed.
  void initTestCase();
  // will be called after the last testfunction was executed.
  void cleanupTestCase(){};
  // will be called before each testfunction is executed.
  void init(){};
  // will be called after every testfunction.
  void cleanup();

Then come your test methods, all of which should take no parameters and should return void. The methods will be called in order of declaration. I am implementing two methods here which illustrates two types of testing. In the first case I want to generally test the various parts of the class are working, I can use a functional testing approach. Once again, extreme programmers would advocate writing these tests before implementing the class. Then as you work your way through your class implementation you iteratively run your unit tests. More and more test functions should complete sucessfully as your class implementation work progresses, and when the whole unit test passes, your new class is done and is now complete with a repeatable way to validate it.

Typically your unit tests would only cover the public API of your class, and normally you do not need to write tests for accessors and mutators. If it should happen that an acccessor or mutator is not working as expected you would normally implement a regression test to check for this (see lower down).

// Functional Testing

/** Check if a raster is valid. */
void isValid();

// more functional tests here ...

Next we implement our regression tests. Regression tests should be implemented to replicate the conditions of a particular bug. For example I recently received a report by email that the cell count by rasters was off by 1, throwing off all the statistics for the raster bands. I opened a bug (ticket #832) and then created a regression test that replicated the bug using a small test dataset (a 10x10 raster). Then I ran the test and ran it, verifying that it did indeed fail (the cell count was 99 instead of 100). Then I went to fix the bug and reran the unit test and the regression test passed. I committed the regression test along with the bug fix. Now if anybody breakes this in the source code again in the future, we can immediatly identify that the code has regressed. Better yet before committing any changes in the future, running our tests will ensure our changes don’t have unexpected side effects - like breaking existing functionality.

There is one more benefit to regression tests - they can save you time. If you ever fixed a bug that involved making changes to the source, and then running the application and performing a series of convoluted steps to replicate the issue, it will be immediately apparent that simply implementing your regression test before fixing the bug will let you automate the testing for bug resolution in an efficient manner.

To implement your regression test, you should follow the naming convention of regression<TicketID> for your test functions. If no redmine ticket exists for the regression, you should create one first. Using this approach allows the person running a failed regression test easily go and find out more information.

// Regression Testing

/** This is our second test case...to check if a raster
 *  reports its dimensions properly. It is a regression test
 *  for ticket #832 which was fixed with change r7650.
void regression832();

// more regression tests go here ...

Finally in our test class declaration you can declare privately any data members and helper methods your unit test may need. In our case I will declare a QgsRasterLayer * which can be used by any of our test methods. The raster layer will be created in the initTestCase() function which is run before any other tests, and then destroyed using cleanupTestCase() which is run after all tests. By declaring helper methods (which may be called by various test functions) privately, you can ensure that they wont be automatically run by the QTest executable that is created when we compile our test.

    // Here we have any data structures that may need to
    // be used in many test cases.
    QgsRasterLayer * mpLayer;

That ends our class declaration. The implementation is simply inlined in the same file lower down. First our init and cleanup functions:

void TestQgsRasterLayer::initTestCase()
  // init QGIS's paths - true means that all path will be inited from prefix
  QString qgisPath = QCoreApplication::applicationDirPath ();
  QgsApplication::setPrefixPath(qgisPath, TRUE);
#ifdef Q_OS_LINUX
  QgsApplication::setPkgDataPath(qgisPath + "/../share/qgis");
  //create some objects that will be used in all tests...

  std::cout << "PrefixPATH: " << QgsApplication::prefixPath().toLocal8Bit().data() << std::endl;
  std::cout << "PluginPATH: " << QgsApplication::pluginPath().toLocal8Bit().data() << std::endl;
  std::cout << "PkgData PATH: " << QgsApplication::pkgDataPath().toLocal8Bit().data() << std::endl;
  std::cout << "User DB PATH: " << QgsApplication::qgisUserDbFilePath().toLocal8Bit().data() << std::endl;

  //create a raster layer that will be used in all tests...
  QString myFileName (TEST_DATA_DIR); //defined in CmakeLists.txt
  myFileName = myFileName + QDir::separator() + "tenbytenraster.asc";
  QFileInfo myRasterFileInfo ( myFileName );
  mpLayer = new QgsRasterLayer ( myRasterFileInfo.filePath(),
  myRasterFileInfo.completeBaseName() );

void TestQgsRasterLayer::cleanupTestCase()
  delete mpLayer;

The above init function illustrates a couple of interesting things.

  1. I needed to manually set the QGIS application data path so that resources such as srs.db can be found properly.
  2. Secondly, this is a data driven test so we needed to provide a way to generically locate the tenbytenraster.asc file. This was achieved by using the compiler define TEST_DATA_PATH. The define is created in the CMakeLists.txt configuration file under <QGIS Source Root>/tests/CMakeLists.txt and is available to all QGIS unit tests. If you need test data for your test, commit it under <QGIS Source Root>/tests/testdata. You should only commit very small datasets here. If your test needs to modify the test data, it should make a copy of it first.

Qt also provides some other interesting mechanisms for data driven testing, so if you are interested to know more on the topic, consult the Qt documentation.

Next lets look at our functional test. The isValid() test simply checks the raster layer was correctly loaded in the initTestCase. QVERIFY is a Qt macro that you can use to evaluate a test condition. There are a few other use macros Qt provide for use in your tests including:

  • QCOMPARE ( actual, expected )
  • QEXPECT_FAIL ( dataIndex, comment, mode )
  • QFAIL ( message )
  • QFETCH ( type, name )
  • QSKIP ( description, mode )
  • QTEST ( actual, testElement )
  • QTEST_APPLESS_MAIN ( TestClass )
  • QTEST_MAIN ( TestClass )
  • QVERIFY2 ( condition, message )
  • QVERIFY ( condition )
  • QWARN ( message )

Some of these macros are useful only when using the Qt framework for data driven testing (see the Qt docs for more detail).

void TestQgsRasterLayer::isValid()
  QVERIFY ( mpLayer->isValid() );

Normally your functional tests would cover all the range of functionality of your classes public API where feasible. With our functional tests out the way, we can look at our regression test example.

Since the issue in bug #832 is a misreported cell count, writing our test is simply a matter of using QVERIFY to check that the cell count meets the expected value:

void TestQgsRasterLayer::regression832()
  QVERIFY ( mpLayer->getRasterXDim() == 10 );
  QVERIFY ( mpLayer->getRasterYDim() == 10 );
  // regression check for ticket #832
  // note getRasterBandStats call is base 1
  QVERIFY ( mpLayer->getRasterBandStats(1).elementCountInt == 100 );

With all the unit test functions implemented, there one final thing we need to add to our test class:

#include "testqgsrasterlayer.moc"

The purpose of these two lines is to signal to Qt’s moc that his is a QtTest (it will generate a main method that in turn calls each test funtion.The last line is the include for the MOC generated sources. You should replace ‘testqgsrasterlayer’ with the name of your class in lower case.

Comparing images for rendering tests

Rendering images on different environments can produce subtle differences due to platform-specific implementations (e.g. different font rendering and antialiasing algorithms), to the fonts available on the system and for other obscure reasons.

When a rendering test runs on Travis and fails, look for the dash link at the very bottom of the Travis log. This link will take you to a cdash page where you can see the rendered vs expected images, along with a “difference” image which highlights in red any pixels which did not match the reference image.

The QGIS unit test system has support for adding “mask” images, which are used to indicate when a rendered image may differ from the reference image. A mask image is an image (with the same name as the reference image, but including a “_mask.png” suffix), and should be the same dimensions as the reference image. In a mask image the pixel values indicate how much that individual pixel can differ from the reference image, so a black pixel indicates that the pixel in the rendered image must exactly match the same pixel in the reference image. A pixel with RGB 2, 2, 2 means that the rendered image can vary by up to 2 in its RGB values from the reference image, and a fully white pixel (255, 255, 255) means that the pixel is effectively ignored when comparing the reference and rendered images.

A utility script to generate mask images is available as scripts/generate_test_mask_image.py. This script is used by passing it the path of a reference image (e.g. tests/testdata/control_images/annotations/expected_annotation_fillstyle/expected_annotation_fillstyle.png) and the path to your rendered image.


scripts/generate_test_mask_image.py tests/testdata/control_images/annotations/expected_annotation_fillstyle/expected_annotation_fillstyle.png /tmp/path_to_rendered_image.png

You can shortcut the path to the reference image by passing a partial part of the test name instead, e.g.

scripts/generate_test_mask_image.py annotation_fillstyle /tmp/path_to_rendered_image.png

(This shortcut only works if a single matching reference image is found. If multiple matches are found you will need to provide the full path to the reference image.)

The script also accepts http urls for the rendered image, so you can directly copy a rendered image url from the cdash results page and pass it to the script.

Be careful when generating mask images - you should always view the generated mask image and review any white areas in the image. Since these pixels are ignored, make sure that these white images do not cover any important portions of the reference image – otherwise your unit test will be meaningless!

Similarly, you can manually “white out” portions of the mask if you deliberately want to exclude them from the test. This can be useful e.g. for tests which mix symbol and text rendering (such as legend tests), where the unit test is not designed to test the rendered text and you don’t want the test to be subject to cross-platform text rendering differences.

To compare images in QGIS unit tests you should use the class QgsMultiRenderChecker or one of its subclasses.

To improve tests robustness here are few tips:

  1. Disable antialiasing if you can, as this minimizes cross-platform rendering differences.
  2. Make sure your reference images are “chunky”… i.e. don’t have 1 px wide lines or other fine features, and use large, bold fonts (14 points or more is recommended).
  3. Sometimes tests generate slightly different sized images (e.g. legend rendering tests, where the image size is dependent on font rendering size - which is subject to cross-platform differences). To account for this, use QgsMultiRenderChecker::setSizeTolerance() and specify the maximum number of pixels that the rendered image width and height differ from the reference image.
  4. Don’t use transparent backgrounds in reference images (CDash does not support them). Instead, use QgsMultiRenderChecker::drawBackground() to draw a checkboard pattern for the reference image background.
  5. When fonts are required, use the font specified in QgsFontUtils::standardTestFontFamily() (“QGIS Vera Sans”).

Adding your unit test to CMakeLists.txt

Adding your unit test to the build system is simply a matter of editing the CMakeLists.txt in the test directory, cloning one of the existing test blocks, and then replacing your test class name into it. For example:

# QgsRasterLayer test
ADD_QGIS_TEST(rasterlayertest testqgsrasterlayer.cpp)

The ADD_QGIS_TEST macro explained

I’ll run through these lines briefly to explain what they do, but if you are not interested, just do the step explained in the above section and section.

MACRO (ADD_QGIS_TEST testname testsrc)
SET(qgis_${testname}_SRCS ${testsrc} ${util_SRCS})
SET(qgis_${testname}_MOC_CPPS ${testsrc})
QT4_WRAP_CPP(qgis_${testname}_MOC_SRCS ${qgis_${testname}_MOC_CPPS})
ADD_CUSTOM_TARGET(qgis_${testname}moc ALL DEPENDS ${qgis_${testname}_MOC_SRCS})
ADD_EXECUTABLE(qgis_${testname} ${qgis_${testname}_SRCS})
ADD_DEPENDENCIES(qgis_${testname} qgis_${testname}moc)
TARGET_LINK_LIBRARIES(qgis_${testname} ${QT_LIBRARIES} qgis_core)
# skip the full RPATH for the build tree
# when building, use the install RPATH already
# (so it doesn't need to relink when installing)
# the RPATH to be used when installing
# add the automatically determined parts of the RPATH
# which point to directories outside the build tree to the install RPATH
# For Mac OS X, the executable must be at the root of the bundle's executable folder
ADD_TEST(qgis_${testname} ${CMAKE_INSTALL_PREFIX}/qgis_${testname})
ADD_TEST(qgis_${testname} ${CMAKE_INSTALL_PREFIX}/bin/qgis_${testname})

Lets look a little more in detail at the individual lines. First we define the list of sources for our test. Since we have only one source file (following the methodology I described above where class declaration and definition are in the same file) its a simple statement:

SET(qgis_${testname}_SRCS ${testsrc} ${util_SRCS})

Since our test class needs to be run through the Qt meta object compiler (moc) we need to provide a couple of lines to make that happen too:

SET(qgis_${testname}_MOC_CPPS ${testsrc})
QT4_WRAP_CPP(qgis_${testname}_MOC_SRCS ${qgis_${testname}_MOC_CPPS})
ADD_CUSTOM_TARGET(qgis_${testname}moc ALL DEPENDS ${qgis_${testname}_MOC_SRCS})

Next we tell cmake that it must make an executable from the test class. Remember in the previous section on the last line of the class implementation I included the moc outputs directly into our test class, so that will give it (among other things) a main method so the class can be compiled as an executable:

ADD_EXECUTABLE(qgis_${testname} ${qgis_${testname}_SRCS})
ADD_DEPENDENCIES(qgis_${testname} qgis_${testname}moc)

Next we need to specify any library dependencies. At the moment, classes have been implemented with a catch-all QT_LIBRARIES dependency, but I will be working to replace that with the specific Qt libraries that each class needs only. Of course you also need to link to the relevant qgis libraries as required by your unit test.

TARGET_LINK_LIBRARIES(qgis_${testname} ${QT_LIBRARIES} qgis_core)

Next I tell cmake to install the tests to the same place as the qgis binaries itself. This is something I plan to remove in the future so that the tests can run directly from inside the source tree.

# skip the full RPATH for the build tree
# when building, use the install RPATH already
# (so it doesn't need to relink when installing)
# the RPATH to be used when installing
# add the automatically determined parts of the RPATH
# which point to directories outside the build tree to the install RPATH
# For Mac OS X, the executable must be at the root of the bundle's executable folder
ADD_TEST(qgis_${testname} ${CMAKE_INSTALL_PREFIX}/qgis_${testname})
ADD_TEST(qgis_${testname} ${CMAKE_INSTALL_PREFIX}/bin/qgis_${testname})

Finally the above uses ADD_TEST to register the test with cmake / ctest. Here is where the best magic happens - we register the class with ctest. If you recall in the overview I gave in the beginning of this section, we are using both QtTest and CTest together. To recap, QtTest adds a main method to your test unit and handles calling your test methods within the class. It also provides some macros like QVERIFY that you can use as to test for failure of the tests using conditions. The output from a QtTest unit test is an executable which you can run from the command line. However when you have a suite of tests and you want to run each executable in turn, and better yet integrate running tests into the build process, the CTest is what we use.

Building your unit test

To build the unit test you need only to make sure that ENABLE_TESTS=true in the cmake configuration. There are two ways to do this:

  1. Run ccmake .. ( or cmakesetup .. under windows) and interactively set the ENABLE_TESTS flag to ON.
  2. Add a command line flag to cmake e.g. cmake -DENABLE_TESTS=true ..

Other than that, just build QGIS as per normal and the tests should build too.

Run your tests

The simplest way to run the tests is as part of your normal build process:

make && make install && make test

The make test command will invoke CTest which will run each test that was registered using the ADD_TEST CMake directive described above. Typical output from make test will look like this:

Running tests...
Start processing tests
Test project /Users/tim/dev/cpp/qgis/build
## 13 Testing qgis_applicationtest***Exception: Other
## 23 Testing qgis_filewritertest *** Passed
## 33 Testing qgis_rasterlayertest*** Passed

## 0 tests passed, 3 tests failed out of 3

The following tests FAILED:
## 1- qgis_applicationtest (OTHER_FAULT)
Errors while running CTest
make: *** [test] Error 8

If a test fails, you can use the ctest command to examine more closely why it failed. Use the -R option to specify a regex for which tests you want to run and -V to get verbose output:

$ ctest -R appl -V

Start processing tests
Test project /Users/tim/dev/cpp/qgis/build
Constructing a list of tests
Done constructing a list of tests
Changing directory into /Users/tim/dev/cpp/qgis/build/tests/src/core
## 13 Testing qgis_applicationtest
Test command: /Users/tim/dev/cpp/qgis/build/tests/src/core/qgis_applicationtest
********* Start testing of TestQgsApplication *********
Config: Using QTest library 4.3.0, Qt 4.3.0
PASS : TestQgsApplication::initTestCase()
PrefixPATH: /Users/tim/dev/cpp/qgis/build/tests/src/core/../
PluginPATH: /Users/tim/dev/cpp/qgis/build/tests/src/core/..//lib/qgis
PkgData PATH: /Users/tim/dev/cpp/qgis/build/tests/src/core/..//share/qgis
User DB PATH: /Users/tim/.qgis/qgis.db
PASS : TestQgsApplication::getPaths()
PrefixPATH: /Users/tim/dev/cpp/qgis/build/tests/src/core/../
PluginPATH: /Users/tim/dev/cpp/qgis/build/tests/src/core/..//lib/qgis
PkgData PATH: /Users/tim/dev/cpp/qgis/build/tests/src/core/..//share/qgis
User DB PATH: /Users/tim/.qgis/qgis.db
QDEBUG : TestQgsApplication::checkTheme() Checking if a theme icon exists:
QDEBUG : TestQgsApplication::checkTheme()
FAIL!: TestQgsApplication::checkTheme() '!myPixmap.isNull()' returned FALSE. ()
Loc: [/Users/tim/dev/cpp/qgis/tests/src/core/testqgsapplication.cpp(59)]
PASS : TestQgsApplication::cleanupTestCase()
Totals: 3 passed, 1 failed, 0 skipped
********* Finished testing of TestQgsApplication *********
-- Process completed

## 0 tests passed, 1 tests failed out of 1

The following tests FAILED:
## 1- qgis_applicationtest (Failed)
Errors while running CTest

Debugging unit tests

For C++ unit tests, QtCreator automatically adds run targets, so you can start them in the debugger.

It’s also possible to start Python unit tests from QtCreator with GDB. For this, you need to go to Projects and choose Run under Build & Run. Then add a new Run configuration with the executable /usr/bin/python3 and the Command line arguments set to the path of the unit test python file, e.g. /home/user/dev/qgis/QGIS/tests/src/python/test_qgsattributeformeditorwidget.py.

Now also change the Run Environment and add 3 new variables:

Variable Value
PYTHONPATH [build]/output/python/:[build]/output/python/plugins:[source]/tests/src/python
QGIS_PREFIX_PATH [build]/output
LD_LIBRARY_PATH [build]/output/lib

Replace [build] with your build directory and [source] with your source directory.

Have fun

Well that concludes this section on writing unit tests in QGIS. We hope you will get into the habit of writing test to test new functionality and to check for regressions. Some aspects of the test system (in particular the CMakeLists.txt parts) are still being worked on so that the testing framework works in a truly platform independent way.