Revision control
Copy as Markdown
// Copyright (C) 2019 The Qt Company Ltd.
// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR GFDL-1.3-no-invariants-only
/*!
\page qttest-best-practices.html
\title Qt Test Best Practices
\brief Guidelines for creating Qt tests.
We recommend that you add Qt tests for bug fixes and new features. Before
you try to fix a bug, add a \e {regression test} (ideally automatic) that
fails before the fix, exhibiting the bug, and passes after the fix. While
you're developing new features, add tests to verify that they work as
intended.
Conforming to a set of coding standards will make it more likely for
Qt autotests to work reliably in all environments. For example, some
tests need to read data from disk. If no standards are set for how this
is done, some tests won't be portable. For example, a test that assumes
its test-data files are in the current working directory only works for
an in-source build. In a shadow build (outside the source directory), the
test will fail to find its data.
The following sections contain guidelines for writing Qt tests:
\list
\li \l {General Principles}
\li \l {Writing Reliable Tests}
\li \l {Improving Test Output}
\li \l {Writing Testable Code}
\li \l {Setting up Test Machines}
\endlist
The advice on \l {Qt Test Security Considerations}{Security Considerations}
should be borne in mind alongside those best practices.
\section1 General Principles
The following sections provide general guidelines for writing unit tests:
\list
\li \l {Verify Tests}
\li \l {Give Test Functions Descriptive Names}
\li \l {Write Self-contained Test Functions}
\li \l {Avoid external dependencies}
\li \l {Test the Full Stack}
\li \l {Make Tests Complete Quickly}
\li \l {Use Data-driven Testing}
\li \l {Use Coverage Tools}
\li \l {Select Appropriate Mechanisms to Exclude Tests}
\li \l {Avoid Q_ASSERT}
\endlist
\section2 Verify Tests
Write and commit your tests along with your fix or new feature on a new
branch. Once you're done, you can check out the branch on which your work
is based, and then check out into this branch the test-files for your new
tests. This enables you to verify that the tests do fail on the prior
branch, and therefore actually do catch a bug or test a new feature.
For example, the workflow to fix a bug in the \c QDateTime class could be
like this if you use the Git version control system:
\list 1
\li Create a branch for your fix and test:
\c {git checkout -b fix-branch dev}
\li Write a test and fix the bug.
\li Build and test with both the fix and the new test, to verify that
the new test passes with the fix.
\li Add the fix and test to your branch:
\c {git add tests/auto/corelib/time/qdatetime/tst_qdatetime.cpp src/corelib/time/qdatetime.cpp}
\li Commit the fix and test to your branch:
\c {git commit -m 'Fix bug in QDateTime'}
\li To verify that the test actually catches something for which you
needed the fix, checkout the branch you based your own branch on:
\c {git checkout dev}
\li Checkout only the test file from the fix branch:
\c {git checkout fix-branch -- tests/auto/corelib/time/qdatetime/tst_qdatetime.cpp}
The rest of the source tree stays on dev, without the fix, but with
the new test ready to try on it.
\li Build and run the test to verify that it fails on dev, and therefore
does indeed catch a bug.
\li You can now return to the fix branch:
\c {git checkout fix-branch}
\li Alternatively, you can restore your work tree to a clean state on
dev: \c{git checkout HEAD -- tests/auto/corelib/time/qdatetime/tst_qdatetime.cpp}
\endlist
When you're reviewing a change, you can adapt this workflow to check that
the change does indeed come with a test for a problem it does fix.
\section2 Give Test Functions Descriptive Names
Naming test cases is important. The test name appears in the failure report
for a test run. For data-driven tests, the name of the data row also appears
in the failure report. Good names can give those reading the report a first
indication of what has gone wrong.
Test function names should make it obvious what the function is trying to
test. Do not simply use the bug-tracking identifier, because the identifiers
become obsolete if the bug-tracker is replaced. Also, developers working
offline cannot access your bug-tracker and some bug-trackers may not be
accessible to all users. When the bug report may be of interest to later
readers of the test code, you can mention it in a comment alongside a
relevant part of the test.
Likewise, when writing data-driven tests, give descriptive names to the
test-cases, that indicate what aspect of the functionality each focuses on.
Do not simply number the test-case, or use bug-tracking identifiers. Someone
reading the test output will have no idea what the numbers or identifiers
mean. You can add a comment on the test-row that mentions the bug-tracking
identifier, when relevant. It's best to avoid spacing characters and
characters that may be significant to command-line shells on which you may
want to run tests. This makes it easier to specify the test and tag on \l{Qt
Test Command Line Arguments}{the command-line} to your test program - for
example, to limit a test run to just one test-case.
\section2 Write Self-contained Test Functions
Within a test program, test functions should be independent of each other
and they should not rely upon previous test functions having been run. You
can check this by running the test function on its own with \c {tst_foo
testname}. For data-driven tests, likewise, avoid dependencies between the
rows of the test's data table so that \c {tst_foo function:tag} can be used
to run a single row in isolation (for example, to investigate why it fails).
Do not re-use instances of the class under test in several tests. Test
instances (for example widgets) should not be member variables of the
tests, but preferably be instantiated on the stack to ensure proper
cleanup even if a test fails, so that tests do not interfere with
each other.
If your test involves making global changes, take care to ensure the prior
state is restored at the end of the test, whether it passes or fails. Since
failure prevents code later than the failing check from running, restoring
at the end of the test doesn't work when the test fails. The robust way to
restore even on failure is to instantiate an RAII (resource acquisition is
initialization) object whose destructor restores the prior state. This can
often be conveniently done using \l qScopeGuard(), for example
\snippet code/src_qtestlib_qtestcase.cpp 36
before the first call to \l QLocale::setDefault() in a test that needs to
control the locale used by the code under test.
\section2 Avoid external dependencies
Test functions should likewise avoid any reliance on external resources.
Such a dependency is apt to make the test fail if that resource is
transiently unavailable. Even when it is available, the resource may block
access by the test system - for example, when tests are run frequently, the
resource may interpret the test as an unwelcome burden on its service.
Skipping the test when inaccessible could hide a problem that would have
been revealed had the test been run, making it a poor remedy for these
issues. It is better to construct a local mimic of the resource, with enough
of its characteristics for the purposes of the test.
External dependencies also present problems for developers working offline,
as well as for testing within a sandbox, such as an isolated virtual
machine, when evaluating code changes from untrusted sources. See \l {Qt
Test Security Considerations} for related concerns.
\section2 Test the Full Stack
If an API is implemented in terms of pluggable or platform-specific backends
that do the heavy-lifting, make sure to write tests that cover the
code-paths all the way down into the backends. Testing the upper layer API
parts using a mock backend is a nice way to isolate errors in the API layer
from the backends, but it is complementary to tests that run the actual
implementation with data that faithfully illustrates real-world conditions.
\section2 Make Tests Complete Quickly
Tests should not waste time by being unnecessarily repetitious, by using
inappropriately large volumes of test data, or by introducing needless
idle time.
This is particularly true for unit testing, where every second of extra
unit test execution time makes CI testing of a branch across multiple
targets take longer. Remember that unit testing is separate from load and
reliability testing, where larger volumes of test data and longer test
runs are expected.
Benchmark tests, which typically execute the same test multiple times,
should be located in a separate \c tests/benchmarks directory and they
should not be mixed with functional unit tests.
\section2 Use Data-driven Testing
\l{Chapter 2: Data Driven Testing}{Data-driven tests} make it easier to add
new tests for boundary conditions found in later bug reports.
Using a data-driven test rather than testing several items in sequence in
a test saves repetition of very similar code and ensures later cases are
tested even when earlier ones fail. It also encourages systematic and
uniform testing, because the same tests are applied to each data sample.
When a test is data-driven, you can specify its data-tag along with the
test-function name, as \c{function:tag}, on the command-line of the test to
run the test on just one specific test-case, rather than all test-cases of
the function. This can be used for either a global data tag or a local tag,
identifying a row from the function's own data; you can even combine them as
\c{function:global:local}.
\section2 Use Coverage Tools
Use a coverage tool such as \l {Coco} or \l {gcov}
to help write tests that cover as many statements, branches, and conditions
as possible in the function or class being tested. The earlier this is done
in the development cycle for a new feature, the easier it will be to catch
regressions later when the code is refactored.
\section2 Select Appropriate Mechanisms to Exclude Tests
It is important to select the appropriate mechanism to exclude inapplicable
tests.
Use \l QSKIP() to handle cases where a whole test function is found at
run-time to be inapplicable in the current test environment. When just a
part of a test function is to be skipped, a conditional statement can be
used, optionally with a \c qDebug() call to report the reason for skipping
the inapplicable part.
When there are known test failures that should eventually be fixed,
\l QEXPECT_FAIL is recommended, as it supports running the rest of the
test, when possible. It also verifies that the issue still exists, and
lets the code's maintainer know if they unwittingly fix it, a benefit
which is gained even when using the \l {QTest::}{Abort} flag.
Test functions or data rows of a data-driven test can be limited to
particular platforms, or to particular features being enabled using
\c{#if}. However, beware of \l moc limitations when using \c{#if} to
skip test functions. The \c moc preprocessor does not have access to
all the \c builtin macros of the compiler that are often used for
feature detection of the compiler. Therefore, \c moc might get a different
result for a preprocessor condition from that seen by the rest of your
code. This may result in \c moc generating meta-data for a test slot that
the actual compiler skips, or omitting the meta-data for a test slot that
is actually compiled into the class. In the first case, the test will
attempt to run a slot that is not implemented. In the second case, the
test will not attempt to run a test slot even though it should.
If an entire test program is inapplicable for a specific platform or unless
a particular feature is enabled, the best approach is to use the parent
directory's build configuration to avoid building the test. For example, if
the \c tests/auto/gui/someclass test is not valid for \macos, wrap its
inclusion as a subdirectory in \c{tests/auto/gui/CMakeLists.txt} in a
platform check:
\badcode
if(NOT APPLE)
add_subdirectory(someclass)
endif
\endcode
or, if using \c qmake, add the following line to \c tests/auto/gui.pro:
\badcode
mac*: SUBDIRS -= someclass
\endcode
See also \l {Chapter 6: Skipping Tests with QSKIP}
{Skipping Tests with QSKIP}.
\section2 Avoid Q_ASSERT
The \l Q_ASSERT macro causes a program to abort whenever the asserted
condition is \c false, but only if the software was built in debug mode.
In both release and debug-and-release builds, \c Q_ASSERT does nothing.
\c Q_ASSERT should be avoided because it makes tests behave differently
depending on whether a debug build is being tested, and because it causes
a test to abort immediately, skipping all remaining test functions and
returning incomplete or malformed test results.
It also skips any tear-down or tidy-up that was supposed to happen at the
end of the test, and might therefore leave the workspace in an untidy state,
which might cause complications for further tests.
Instead of \c Q_ASSERT, the \l QCOMPARE() or \l QVERIFY() macro variants
should be used. They cause the current test to report a failure and
terminate, but allow the remaining test functions to be executed and the
entire test program to terminate normally. \l QVERIFY2() even allows a
descriptive error message to be recorded in the test log.
\section1 Writing Reliable Tests
The following sections provide guidelines for writing reliable tests:
\list
\li \l {Avoid Side-effects in Verification Steps}
\li \l {Avoid Fixed Timeouts}
\li \l {Beware of Timing-dependent Behavior}
\li \l {Avoid Bitmap Capture and Comparison}
\endlist
\section2 Avoid Side-effects in Verification Steps
When performing verification steps in an autotest using \l QCOMPARE(),
\l QVERIFY(), and so on, side-effects should be avoided. Side-effects
in verification steps can make a test difficult to understand. Also,
they can easily break a test in ways that are difficult to diagnose
when the test is changed to use \l QTRY_VERIFY(), \l QTRY_COMPARE() or
\l QBENCHMARK. These can execute the passed expression repeatedly, thus
repeating any side-effects.
When side-effects are unavoidable, ensure that the prior state is restored
at the end of the test function, even if the test fails. This commonly
requires use of an RAII class (see \l {Write Self-contained Test Functions}
above), or a \c cleanup() method. Do not simply put the restoration code at
the end of the test. If part of the test fails, such code will be skipped
and the prior state will not be restored.
\section2 Avoid Fixed Timeouts
Avoid using hard-coded timeouts, such as QTest::qWait() to wait for some
conditions to become true. Consider using the \l QSignalSpy class,
the \l QTRY_VERIFY() or \l QTRY_COMPARE() macros, or the \c QSignalSpy
class in conjunction with the \c QTRY_ macro variants.
The \c qWait() function can be used to set a delay for a fixed period
between performing some action and waiting for some asynchronous behavior
triggered by that action to be completed. For example, changing the state
of a widget and then waiting for the widget to be repainted. However,
such timeouts often cause failures when a test written on a workstation is
executed on a device, where the expected behavior might take longer to
complete. Increasing the fixed timeout to a value several times larger
than needed on the slowest test platform is not a good solution, because
it slows down the test run on all platforms, particularly for table-driven
tests.
If the code under test issues Qt signals on completion of the asynchronous
behavior, a better approach is to use the \l QSignalSpy class to notify
the test function that the verification step can now be performed.
If there are no Qt signals, use the \c QTRY_COMPARE() and \c QTRY_VERIFY()
macros, which periodically test a specified condition until it becomes true
or some maximum timeout is reached. These macros prevent the test from
taking longer than necessary, while avoiding breakages when tests are
developed on faster systems and later executed on slower ones.
If there are no Qt signals, and you are writing the test as part of
developing a new API, consider whether the API could benefit from the
addition of a signal that reports the completion of the asynchronous
behavior. If it would make your testing easier, it may well be useful for
callers of your API, too.
\section2 Beware of Timing-dependent Behavior
Some test strategies are vulnerable to timing-dependent behavior of certain
classes, which can lead to tests that fail only on certain platforms or that
do not return consistent results.
One example of this is text-entry widgets, which often have a blinking
cursor that can make comparisons of captured bitmaps succeed or fail
depending on the state of the cursor when the bitmap is captured. This,
in turn, may depend on the speed of the machine executing the test.
When testing classes that change their state based on timer events, the
timer-based behavior needs to be taken into account when performing
verification steps. Due to the variety of timing-dependent behavior, there
is no single generic solution to this testing problem.
For text-entry widgets, potential solutions include disabling the cursor
blinking behavior (if the API provides that feature), waiting for the
cursor to be in a known state before capturing a bitmap (for example, by
subscribing to an appropriate signal if the API provides one), or
excluding the area containing the cursor from the bitmap comparison.
\section2 Avoid Bitmap Capture and Comparison
While verifying test results by capturing and comparing bitmaps is sometimes
necessary, it can be quite fragile and labor-intensive.
For example, a particular widget may have different appearance on different
platforms or with different widget styles, so reference bitmaps may need to
be created multiple times and then maintained in the future as Qt's set of
supported platforms evolves. Making changes that affect the bitmap thus
means having to recreate the expected bitmaps on each supported platform,
which would require access to each platform.
Bitmap comparisons can also be influenced by factors such as the test
machine's screen resolution, bit depth, active theme, color scheme,
widget style, active locale (currency symbols, text direction, and so
on), font size, transparency effects, and choice of window manager.
Where possible, use programmatic means, such as verifying properties of
objects and variables, instead of capturing and comparing bitmaps.
\section1 Improving Test Output
The following sections provide guidelines for producing readable and
helpful test output:
\list
\li \l {Test for Warnings}
\li \l {Avoid Printing Debug Messages from Autotests}
\li \l {Write Well-structured Diagnostic Code}
\endlist
\section2 Test for Warnings
Just as when building your software, if test output is cluttered with
warnings you will find it harder to notice a warning that really is a clue
to the emergence of a bug. It is thus prudent to regularly check your test
logs for warnings, and other extraneous output, and investigate the
causes. When they are signs of a bug, you can make warnings trigger test
failure.
When the code under test \e should produce messages, such as warnings
about misguided use, it is also important to test that it \e does produce
them when so used. You can test for expected messages from the code under
test, produced by \l qWarning(), \l qDebug(), \l qInfo() and friends,
using \l QTest::ignoreMessage(). This will verify that the message is
produced and filter it out of the output of the test run. If the message
is not produced, the test will fail.
If an expected message is only output when Qt is built in debug mode, use
\l QLibraryInfo::isDebugBuild() to determine whether the Qt libraries were
built in debug mode. Using \c{#ifdef QT_DEBUG} is not enough, as it will
only tell you whether \e{the test} was built in debug mode, and that does
not guarantee that the \e{Qt libraries} were also built in debug mode.
Your tests can verify that they do not trigger calls to \l qWarning() by
calling \l QTest::failOnWarning(). With no parameter (since Qt 6.8), this
makes the test fail if a warning is produced, outputting the warning. You
can optionally pass \c failOnWarning() either a warning message to test for
or a \l QRegularExpression to match against, so as to limit this behavior to
matching warnings. (These filtered version were introduced in Qt 6.3.)
You can also set the environment variable \c QT_FATAL_WARNINGS to cause
warnings to be treated as fatal errors. See \l qWarning() for details; this
is not specific to autotests. If warnings would otherwise be lost in vast
test logs, the occasional run with this environment variable set can help
you to find and eliminate any that do arise.
\section2 Avoid Printing Debug Messages from Autotests
Autotests that pass should not produce any unhandled warning or debug
messages. This will allow the CI Gate to treat new warning or debug messages
as test failures. As with warnings from compilers when building your code,
if warnings are rare they are useful clues to problems, but if they are
routine they may hide meaningful issues from developers testing their
changes.
Adding debug messages during development is fine, but these should be
either disabled or removed before a test is checked in.
\section2 Write Well-structured Diagnostic Code
Any diagnostic output that would be useful if a test fails should be part
of the regular test output rather than being commented-out, disabled by
preprocessor directives, or enabled only in debug builds. If a test fails
during continuous integration, having all of the relevant diagnostic output
in the CI logs could save you a lot of time compared to enabling the
diagnostic code and testing again. Epecially, if the failure was on a
platform that you don't have on your desktop.
Diagnostic messages in tests should use Qt's output mechanisms, such as
\c qDebug() and \c qWarning(), rather than \c stdio.h or \c iostream.h output
mechanisms. The latter bypass Qt's message handling and prevent the
\c -silent command-line option from suppressing the diagnostic messages.
This could result in important failure messages being hidden in a large
volume of debugging output.
Where there is a need, on failure, to augment the output of the Qt Test
macros such as \l QCOMPARE(), one useful pattern is to have the diagnostic
code run by a \l qScopeGuard() instantiated before, and then
\l {QScopeGuard::} {dismiss()} the instance after, the check or checks whose
failure would be easier to understand with that output. The diagnostic code
is then only run when a check, by failing, skips the later \c dismiss()
call. For example:
\snippet code/src_qtestlib_qtestcase.cpp scope-diagnostic
This augments the \c QCOMPARE() output on failure with a helpfully-formatted
representation of the \c actual result, to make it easier to compare it with
a test-case's \c expected result (already available in the test's own source
code). Other situations may need more complex diagnostics.
\section1 Writing Testable Code
The following sections provide guidelines for writing code that is easy to
test:
\list
\li \l {Break Dependencies}
\li \l {Compile All Classes into Libraries}
\endlist
\section2 Break Dependencies
The idea of unit testing is to use every class in isolation. Since many
classes instantiate other classes, it is not possible to instantiate one
class separately. Therefore, you should use a technique called
\e {dependency injection} that separates object creation from object use.
A factory is responsible for building object trees. Other objects manipulate
these objects through abstract interfaces.
This technique works well for data-driven applications. For GUI
applications, this approach can be difficult as objects are frequently
created and destructed. To verify the correct behavior of classes that
depend on abstract interfaces, \e mocking can be used. For example, see
\l {Googletest Mocking (gMock) Framework}.
\section2 Compile All Classes into Libraries
In small to medium sized projects, a build script typically lists all
source files and then compiles the executable in one go. This means that
the build scripts for the tests must list the needed source files again.
It is easier to list the source files and the headers only once in a
script to build a static library. Then the \c main() function will be
linked against the static library to build the executable and the tests
will be linked against the static libraries.
For projects where the same source files are used in building several
programs, it may be more appropriate to build the shared classes into
a dynamically-linked (or shared object) library that each program,
including the test programs, can load at run-time. Again, having the
compiled code in a library helps to avoid duplication in the description
of which components to combine to make the various programs.
\section1 Setting up Test Machines
The following sections discuss common problems caused by test machine setup:
\list
\li \l {Screen Savers}
\li \l {System Dialogs}
\li \l {Display Usage}
\li \l {Window Managers}
\endlist
All of these problems can typically be solved by the judicious use of
virtualisation.
\section2 Screen Savers
Screen savers can interfere with some of the tests for GUI classes, causing
unreliable test results. Screen savers should be disabled to ensure that
test results are consistent and reliable.
\section2 System Dialogs
Dialogs displayed unexpectedly by the operating system or other running
applications can steal input focus from widgets involved in an autotest,
causing unreproducible failures.
Examples of typical problems include online update notification dialogs
on \macos, false alarms from virus scanners, scheduled tasks such as virus
signature updates, software updates pushed out to workstations, and chat
programs popping up windows on top of the stack.
\section2 Display Usage
Some tests use the test machine's display, mouse, and keyboard, and can
thus fail if the machine is being used for something else at the same
time or if multiple tests are run in parallel.
The CI system uses dedicated test machines to avoid this problem, but if
you don't have a dedicated test machine, you may be able to solve this
problem by running the tests on a second display.
On Unix, one can also run the tests on a nested or virtual X-server, such as
Xephyr. For example, to run the entire set of tests on Xephyr, execute the
following commands:
\code
Xephyr :1 -ac -screen 1920x1200 >/dev/null 2>&1 &
sleep 5
DISPLAY=:1 icewm >/dev/null 2>&1 &
cd tests/auto
make
DISPLAY=:1 make -k -j1 check
\endcode
Users of NVIDIA binary drivers should note that Xephyr might not be able to
provide GLX extensions. Forcing Mesa libGL might help:
\code
export LD_PRELOAD=/usr/lib/mesa-diverted/x86_64-linux-gnu/libGL.so.1
\endcode
However, when tests are run on Xephyr and the real X-server with different
libGL versions, the QML disk cache can make the tests crash. To avoid this,
use \c QML_DISABLE_DISK_CACHE=1.
Alternatively, use the offscreen plugin:
\code
TESTARGS="-platform offscreen" make check -k -j1
\endcode
\section2 Window Managers
On Unix, at least two autotests (\c tst_examples and \c tst_gestures)
require a window manager to be running. Therefore, if running these
tests under a nested X-server, you must also run a window manager
in that X-server.
Your window manager must be configured to position all windows on the
display automatically. Some windows managers, such as Tab Window Manager
(twm), have a mode for manually positioning new windows, and this prevents
the test suite from running without user interaction.
\note Tab Window Manager is not suitable for running the full suite of
Qt autotests, as the \c tst_gestures autotest causes it to forget its
configuration and revert to manual window placement.
*/