A unit testing framework for Arduino platforms inspired by by
ArduinoUnit and Google
Test. AUnit is almost a drop-in
replacement of ArduinoUnit (v2.2) with some advantages. AUnit supports timeouts
and test fixtures. It sometimes consumes 50% less flash memory on the AVR
platform, and it has been tested to work on the AVR, SAMD21, STM32, ESP8266,
ESP32 and Teensy platforms. The assertion error messages were updated in v1.7 to
provide seamless integration with Unix tools like vim
.
Originally, the AUnit tests were designed to run on the embedded controller itself which allows detection of architecture-specific problems. But the uploading, flashing, and execution process is often slow and flaky, causing the iteration cycle to take too much time. It is often more effective to execute the AUnit tests natively on a host machine (running Linux, MacOS, or FreeBSD) using the EpoxyDuino companion project. Once the unit tests are running on the Linux or MacOS host machine, they can be incorporated into a continuous integration system like Jenkins system or a cloud-based system like GitHub Actions.
Version: 1.7.1 (2023-06-15)
Changelog: CHANGELOG.md
- Summary
- Installation
- Documentation
- Usage
- Header and Namespace
- Verbose Mode
- Defining the Tests
- Generated Class and Instance Names
- Binary Assertions
- Approximate Comparisons
- Boolean Assertions
- Test Fixtures
- Early Return and Delayed Assertions
- Meta Assertions
- Unconditional Termination
- Overridable Methods
- Running the Tests
- Filtering Test Cases
- Output Printer
- Controlling Verbosity
- Line Number Mismatch
- Test Framework Messages
- Test Timeout
- GoogleTest Adapter
- Command Line Tools
- Continuous Integration
- Tips
- Benchmarks
- System Requirements
- License
- Feedback and Support
- Authors
AUnit was created to solve 3 problems with ArduinoUnit 2.2:
- ArduinoUnit consumes too much flash memory on an AVR platform (e.g. Arduino UNO, Nano) as explained in ArduinoUnit#70.
- ArduinoUnit does not compile on the ESP8266 platform (see ArduinoUnit#68, ArduinoUnit#55, ArduinoUnit#54).
- ArduinoUnit does not provide an easy way to create tests using fixtures,
equivalent to the
TEST_F()
macro in Google Test.
In contrast:
- AUnit consumes as much as 65% less flash memory than ArduinoUnit 2.2 on the AVR platform. On Teensy-ARM, the savings can be as much as 30%.
- AUnit has been tested on AVR, Teensy-ARM and ESP8266.
- AUnit implements the
testF()
andtestingF()
macros to use fixtures.
The AUniter command line tools can compile the unit tests, upload them to Arduino boards attached to the serial ports of the local machine, and validate the output of the AUnit test runner. In addition, the AUniter script can be integrated into a Jenkins continuous integration service running on the local machine, and the unit tests can be monitored automatically.
Unit tests written using AUnit can often be compiled and executed natively on
Linux or MacOS using the
EpoxyDuino library. The output on
the Serial
object is redirected to the stdout
of the Unix host. This
provides another avenue for implementing continuous builds or integration.
For basic unit tests written using ArduinoUnit 2.2, only two changes are required to convert to AUnit:
#include <ArduinoUnit.h>
->#include <AUnit.h>
Test::run()
->aunit::TestRunner::run()
Most of the core macros are compatible between ArduinoUnit and AUnit:
test()
testing()
assertXxx()
- Meta Assertions
checkTestXxx()
assertTestXxx()
externTest()
externTesting()
AUnit also supports exclude and include filters:
TestRunner::exclude()
TestRunner::include()
Filters can be accessed through Command Line Flags on desktop machines using EpoxyDuino
The various assertion and test status messages can be enabled or disabled using
the Verbosity
flags on per test basis:
enableVerbosity()
disableVerbosity()
Here are the features which have not been ported over from ArduinoUnit 2.2:
- ArduinoUnit supports multiple
*
wildcards in itsexclude()
andinclude()
methods. AUnit supports only a single*
wildcard and it must occur at the end if present.
Here are the features in AUnit which are not available in ArduinoUnit 2.2:
- Configurable timeout parameter to prevent
testing()
test cases from running forever:TestRunner::setTimeout(seconds)
Test::expire()
assertTestExpire()
assertTestNotExpire()
checkTestExpire()
checkTestNotExpire()
- Case-insensitive string comparisons:
assertStringCaseEqual()
assertStringCaseNotEqual()
- Approximate comparisons:
assertNear()
asssertNotNear()
- Supports 64-bit integer
assertXxx()
support bothlong long
andunsigned long long
test()
andtesting()
macros support both 1 and 2 argumentstest(testName)
andtest(suiteName, testName)
testing(testName)
andtesting(suiteName, testName)
- Test fixtures using the "F" variations of existing macros:
testF()
testingF()
assertTestXxxF()
checkTestXxxF()
externTestF()
externTestingF()
- Unconditional termination:
passTestNow()
failTestNow()
skipTestNow()
expireTestNow()
teardown()
method which mirrors thesetup()
method:teardown()
- Test filters support 2-arguments, matching
testF()
andtestingF()
:TestRunner::include(testClass, name)
TestRunner::exclude(testClass, name)
- No dynamic allocation of heap memory:
- The library does not create a
String
object. - It reads from a
String
if the calling code provides it.
- The library does not create a
- Terse and verbose modes:
#include <AUnit.h>
- terse messages use less flash memory#include <AUnitVerbose.h>
- verbose messages use more flash memory
Every feature of AUnit is unit tested using AUnit itself.
The latest stable release is available in the Arduino IDE Library Manager. Search for "unit test" or "AUnit", select "AUnit", then click the "Install" button.
The development version can be installed by cloning the
GitHub repository, checking out the
develop
branch, then manually copying over the contents to the ./libraries
directory used by the Arduino IDE. (The result is a directory named
./libraries/AUnit
.) See the Preferences menu for the location of the Arduino
sketch directory. The master
branch contains the stable release.
Using either installation method, you may need to restart the Arduino IDE to pick up the new library.
The source files are organized as follows:
src/AUnit.h
- main header filesrc/AUnitVerbose.h
- verbose version of main header filesrc/aunit/
- all implementation filestests/
- unit tests written using AUnit itselfexamples/
- example sketches
- README.md - this file
- Doxygen docs published on GitHub Pages can help navigate the classes and header files
The examples/
directory has a number of examples:
- Basic
- Intermediate
- filter
- how to filter tests using
TestRunner::include()
andTestRunner::exclude()
- how to filter tests using
- meta_asserts
- how to use
assertTestXxx()
andcheckTestXxx()
- how to use
- filter
- Advanced
- advanced
- how to subclass
Test
andTestOnce
manually
- how to subclass
- continuous
- using the
testing()
macro
- using the
- advanced
- Benchmarks
- These are internal benchmark programs, not meant as examples (they are in
the
examples/
directory because of the Arduino IDE). - MemoryBenchmark
- Determines the flash and static memory consumption of AUnit for various microcontroller
- These are internal benchmark programs, not meant as examples (they are in
the
In the tests/
directory, there are unit tests to test the AUnit framework
itself:
- AUnitTest
- the unit test for core
AUnit
functions
- the unit test for core
- AUnitMoreTest
- more tests
- AUnitMetaTest
- the unit test for meta assertions and
extern*()
macros
- the unit test for meta assertions and
- CompareTest
- tests for the low-level compare functions
- FailingTest
- tests that are expected to fail
- FilterTest
- manual tests for
include()
andexclude()
filters
- manual tests for
- Print64Test
- manual tests for
include()
andexclude()
filters
- manual tests for
- SetupAndTeardownTest
- tests to verify that
setup()
andteardown()
are called properly by the finite state machine
- tests to verify that
- tests/Makefile
- Runs the tests on a Linux or Mac machine using EpoxyDuino
Perhaps the best way to see AUnit in action is through real life examples. All my libraries use AUnit for testing and for continuous integration through EpoxyDuino. Here are some of my libraries:
- AceButton
- My first Arduino library, which originally used ArduinoUnit 2.2.
- I kept many of the original ArduinoUnit tests for backwards compatibility testing. But over time, I started to use more AUnit features so I'm not sure if they work with ArduinoUnit anymore.
- AceCRC
- AceCommon
- AceRoutine
- AceSegment
- AceSorting
- AceTime
In this section, information about differences between AUnit and ArduinoUnit will appear in a note marked by ArduinoUnit Compatibility.
To prevent name clashes with other libraries and code, all classes in the AUnit
library are defined in the aunit
namespace. The user will mostly interact with
the TestRunner
class. It can be referenced with an explicit namespace
qualifier (i.e. aunit::TestRunner
), or we can use a using
directive like
this:
#include <AUnit.h>
using aunit::TestRunner;
or we can import the entire aunit
namespace:
#include <AUnit.h>
using namespace aunit;
Similar to ArduinoUnit, many of the "functions" in this framework (e.g.
test()
, testing()
, assertXxx()
) are defined as #define
macros which live
in the global namespace, so it is usually not necessary to import the entire
aunit
namespace.
By default, AUnit generates terse assertion messages by leaving out
the string arguments of the various assertXxx()
macros. If you would like
to get the same verbose output as ArduinoUnit, use the following header
instead:
#include <AUnitVerbose.h>
The flash memory consumption on an 8-bit AVR may go up by 20-25% for medium to large tests. On Teensy ARM, ESP8266 or ESP32, the increased memory size probably does not matter too much because these microcontrollers have far more flash and static memory.
The usage of AUnit is basically identical to ArduinoUnit. The following macros are used to create a test:
test(name) {...}
- creates a subclass ofTestOnce
testing(name) {...}
- creates a subclass ofTestAgain
test(suiteName, name) {...}
- creates a subclass ofTestOnce
testing(suiteName, name) {...}
- creates a subclass ofTestAgain
testF(classname, name) {...}
- creates a subclass ofclassname
testingF(classname, name) {...}
- creates a subclass ofclassname
The code in { }
following these macros becomes the body of a method in a
subclass derived from the base class indicated above. The test()
and testF()
macros place the code body into the TestOnce::once()
method. The testing()
and testingF()
macros place the code body into the TestAgain::again()
method.
The test()
and testing()
macros support 1 or 2 arguments. The one-argument
version is inherited from ArduinoUnit. The two-argument version is
analogous to the TEST()
macro in GoogleTest, where the suiteName
can
be used to organize multiple tests into a collection of similar tests. The
grouping is purely in the naming scheme of the generated code, there is no
functional relationship between these tests.
During static initialization, the constructor of the object adds itself to an
internal list. The root of that list is given by Test::getRoot()
. The
TestRunner::run()
method traverses the linked list, executing each test case
until it passes, fails or is skipped.
Here is a rough outline of an AUnit unit test sketch:
#line 2 AUnitTest.ino
#include <AUnit.h>
using namespace aunit;
test(example) {
...
assertXxx(...)
...
}
test(ExampleTest, example) {
...
}
testing(looping) {
...code...
if (...) {
pass();
} else if (...) {
failTestNow();
} else {
skipTestNow();
}
}
testing(LoopingTest, looping) {
...
}
class CustomTestOnce: public TestOnce {
protected:
// optional
void setup() override {
TestOnce::setup();
...setup code...
}
// optional
void teardown() override {
...teardown code...
TestOnce::teardown();
}
void assertBigStuff() {
...higher level assertions...
}
};
testF(CustomTestOnce, example_test) {
...
assertBigStuff();
...
}
class CustomTestAgain: public TestAgain {
protected:
// optional
void setup() override {
TestAgain::setup();
...setup code...
}
// optional
void teardown() override {
...teardown code...
TestAgain::teardown();
}
void assertBigStuff() {
...various assertions...
}
};
testingF(CustomTestAgain, example_test) {
...
assertBigStuff();
...
}
void setup() {
Serial.begin(115200);
while (! Serial); // Wait until Serial is ready - Leonardo/Micro
TestRunner::exclude("*");
TestRunner::include("looping*");
TestRunner::include("CustomTestAgain", "example*");
}
void loop() {
TestRunner::run();
}
ArduinoUnit Compatibility: The basic structure of the unit test is
identical to ArduinoUnit. AUnit adds the testF()
and testingF
() macros,
and the two-argument versions of test()
and testing()
which
are not available in ArduinoUnit. The Test
class in ArduinoUnit has been
replaced with the TestAgain
class in AUnit.
The arguments to the various test*()
macros are used to generate the name for
the subclasses of TestOnce
or TestAgain
, and generate the names of the
instances of those classes. For reference, here are the rules:
test(name)
- class:
"test_"
+ name - instance:
"test_"
+ name +"_instance"
- class:
testing(name)
- class:
"test_"
+ name - instance:
"test_"
+ name +"_instance"
- class:
test(suiteName, name)
- class:
suiteName
+"_"
+ name - instance:
suiteName
+"_"
+ name +"_instance"
- class:
testing(suiteName, name)
- class:
suiteName
+"_"
+ name - instance:
suiteName
+"_"
+ name +"_instance"
- class:
testF(className, name)
- class:
className
+"_"
+ name - instance:
className
+"_"
+ name +"_instance"
- class:
testingF(className, name)
- class:
className
+"_"
+ name - instance:
className
+"_"
+ name +"_instance"
- class:
The instance name is available within the test code using the Test::getName()
method.
Inside the test()
and testing()
macros, the following assertions
are available. These are essentially identical to ArduinoUnit:
assertEqual(a, b)
assertNotEqual(a, b)
assertLess(a, b)
assertMore(a, b)
assertLessOrEqual(a, b)
assertMoreOrEqual(a, b)
When the assertion passes, nothing is printed by default. This can be controlled
by the TestRunner::setVerbosity()
method. See Controlling
Verbosity.
When the assertion fails, an error message of the following format is printed:
SampleTest.ino:10: Assertion failed: (2) == (1)
The format of the assertion failure messages was changed in v1.7 to the following:
{fileName}:{lineNumber}: Assertion failed: {expression}
This format is a widely used in many other programs, for example, the C compiler
gcc
, the C++ compiler g++
, the Python 3 interpreter python3
, grep
, and
the GNU Make program make
. In particular, the
quickfix feature in the vim
text
editor can parse this error format and jump directly to the fileName
and
lineNumber
indicated by the error message. See the instructions in
EpoxyDuino to see how to run unit tests
on a Linux or MacOS machine inside the vim
editor so that the editor jumps
directly to the files and line numbers where the assertion failure occurred.
The 6 core assert macros (assertEqual()
, assertNotEqual()
, assertLess()
,
assertMore()
, assertLessOrEqual()
, assertMoreOrEqual()
) support the
following 18 combinations for their parameter types:
(bool, bool)
(char, char)
(int, int)
(unsigned int, unsigned int)
(long, long)
(unsigned long, unsigned long)
(long long, long long)
(unsigned long long, unsigned long long)
(double, double)
(const char *, const char *)
(const char *, const String&)
(const char *, const __FlashStringHelper*)
(const String&, const char*)
(const String&, const String&)
(const String&, const __FlashStringHelper*)
(const __FlashStringHelper*, const char*)
(const __FlashStringHelper*, const String&)
(const __FlashStringHelper*, const __FlashStringHelper*)
The assertEqual()
and assertNotEqual()
support arbitrary pointer types
through implicit casts to const void*
:
(const void*, const void*)
(since v1.4)
All 9 combinations of the 3 string types (char*
, String
, and
__FlashStringHelper*
) are supported.
These macros perform deep comparisons for string types instead of just comparing
their pointer values. This is different than the ASSERT_EQ()
and ASSERT_NE()
macros in Google Test which perform only pointer comparisons. In other words,
assertEqual()
with string types is equivalent to ASSERT_STREQ()
in Google
Test.
Also for string types, these macros support nullptr
(unlike the underlying
strcmp()
function from the C-library). The nullptr
string is defined to be
"smaller" than any non-null string, including the empty string. Two nullptr
strings are considered to be equal however.
Additionally, the usual C++ implicit type conversion and function overloading matching algorithms apply to support additional argument types. For example, the following type conversions will occur:
signed char
->int
unsigned char
->int
short
->int
unsigned short
->int
orunsigned int
(depending onsizeof(int)
)char*
->const char*
.char[N]
->const char*
float
->double
- pointer types ->
const void*
Note that char
, signed char
, and unsigned char
are 3 distinct types in
C++, so a (char, char)
will match exactly to one of the assertXxx()
methods.
ArduinoUnit Compatibility:
The names of the macros are identical. However, the
type inference logic of two (a, b)
arguments in the assertXxx(a, b)
is
slightly different. ArduinoUnit allows the two parameters to be slightly
different types, at the expense of a compiler warning. In AUnit, the
warning becomes a compiler error. See the "Parameters Must Match Types" section
below.
In ArduinoUnit, the assertXxx()
macros could be slightly different types, for
example:
unsigned int uintValue = 5;
assertEqual(5, uintValue);
If the compiler warnings are enabled in the Preferences box of the IDE, a warning from the compiler is printed:
../ArduinoUnit/src/ArduinoUnitUtility/Compare.h:17:28: warning:
comparison between signed and unsigned integer expressions [-Wsign-compare]
return (!(a<b)) && (!(b<a));
In AUnit, the above code produces a compiler error (not a warning) like this:
.../AUnit/src/aunit/Assertion.h:29:66: error: call of overloaded
'assertion(const char [14], int, int, const char[3], <unresolved overloaded
function type>, unsigned int&)' is ambiguous
if (!aunit::assertion(__FILE__,__LINE__,(arg1),opName,op,(arg2)))\
...
The compiler cannot find an appropriate overloaded version of assertEqual()
.
The solution is to make the parameters the same type:
assertEqual(5U, uintValue);
On the AVR platform, both a (short) and (int) are 16-bit types, so the following will produce a compiler error:
unsigned short ushortValue = 5;
assertEqual(5U, ushortValue);
But on Teensy-ARM, ESP8266, and ESP32, a 16-bit (short) can be promoted to a 32-bit (int) without loss of precision, so the above will compile just fine. For portability, the following should be used on all platforms:
unsigned short ushortValue = 5;
assertEqual((unsigned short) 5, ushortValue);
The integer type promotion rules and function overload matching rules can be difficult to remember (and sometimes difficult to understand). The best way to avoid these compiler errors is to make sure that the assertion parameter types are identical, potentially using explicit casting.
Version 1.4 adds pointer comparison to assertEqual()
and assertNotEqual()
.
Arbritary pointers are implicitly cast to a const void*
and compared to
each other. If the assertion fails, the pointer is converted to an integer type,
and the hexadecimal value of the pointer is printed. For example,
test(voidPointer) {
const int aa[] = {1, 2};
const long bb[] = {1, 2};
assertEqual(aa, bb);
}
This test will fail with the following error message:
AUnitTest.ino:338: Assertion failed: (aa=0x3FFFFF38) == (bb=0x3FFFFF30).
Test voidPointer failed.
Comparison against the nullptr
will work:
test(nullPointer) {
const int aa[] = {1, 2};
assertEqual(aa, nullptr);
}
This will print the following:
AUnitTest.ino:348: Assertion failed: (aa=0x3FFFFF58) == (nullptr=0x0).
Test nullPointer failed.
Comparing a string type (i.e. const char*
, or const __FlashStringHelper*
)
to a nullptr
will cause an error due to ambiguous matches on overloaded
functions. The solution is to explicitly cast the nullptr
to the corresponding
string type:
test(stringPointer) {
const char aa[] = "abc";
// assertEqual(aa, nullptr); // Causes errors
assertEqual(aa, (const char*) nullptr); // Works.
}
Two macros provide case-insensitive string comparisons (analogous to
ASSERT_STRCASEEQ()
and ASSERT_STRCASENE()
in Google Test):
assertStringCaseEqual(a, b)
assertStringCaseNotEqual(a, b)
The supported types for (a, b)
are all 9 combinations of Arduino string types:
(const char *, const char *)
(const char *, const String&)
(const char *, const __FlashStringHelper*)
(const String&, const char*)
(const String&, const String&)
(const String&, const __FlashStringHelper*)
(const __FlashStringHelper*, const char*)
(const __FlashStringHelper*, const String&)
(const __FlashStringHelper*, const __FlashStringHelper*)
ArduinoUnit Compatibility: Not available in ArduinoUnit.
Floating point values are difficult to compare because of internal rounding errors. Google Test provides two types of macros to handle floating points:
ASSERT_FLOAT_EQ(a, b)
,ASSERT_DOUBLE_EQ(a, b)
- determine if the floating point numbers are within 4 Units in the Last Place (ULPs)ASSERT_NEAR(a, b, error)
- determine if the absolute distance betweena
andb
is within the givenerror
Since floating point operations are relatively rare in Arduino programming,
AUnit offers only the equivalent of ASSERT_NEAR()
function:
assertNear(a, b, error)
assertNotNear(a, b, error)
Upon failure, the error messages will look something like:
AUnitTest.ino:517: Assertion failed: |(1.00) - (1.10)| > (0.20).
AUnitTest.ino:527: Assertion failed: |(4.00) - (1.10)| <= (0.20).
Unlike Google Test where ASSERT_NEAR()
supports only the double
type, both
assertNear()
and assertNotNear()
support integral types as well. The full
list of supported types is:
int
unsigned int
long
unsigned long
double
Other primitive types (e.g. char
or float
) will be automatically converted
to one of these supported types by the compiler. (64-bit integer types
long long
and unsigned long long
are not supported here.)
Note that the abs()
of 2 values of a signed integer type can be larger than
the maximum value that can be represented by the given signed type. Since signed
integer overflow is an
undefined behavior
in C and C++11, I cannot predict what the compiler will do in that case.
Unsigned types should not have this problem because the distance between two
values of an unsigned type should always fit inside the given unsigned type.
Technically, a similar problem exists for the floating point types (which are
naturally signed), but it is unlikely that you are dealing with floating point
values so close to the maximum values.
The following boolean asserts are also available:
assertTrue(condition)
assertFalse(condition)
ArduinoUnit Compatibility: These are identical to ArduinoUnit.
When the unit tests become more complex, using test fixtures will allow you to place common data objects and methods into a class that can be shared among multiple test cases. This concept matches very closely to the test fixtures in Google Test.
To create a test fixture:
- Derives a new class from either
TestOnce
(if you want to run the test just once), orTestAgain
(if you want to run the test repeatedly). - Add any data objects inside the class.
- Optionally add a
virtual void setup() {...}
method to perform any common initialization code. Be sure to call the parent'ssetup()
method in the first line to chain anysetup()
methods defined by the parents. There may be multiple parent classes. - Optionally add a
virtual void teardown() {...}
method to perform any common clean up code. Be sure to call the parent'steardown()
method in the last line to chain anyteardown()
methods defined by the parents. There may be multiple parent classes. - Add any additional shared methods into this new class.
To define your tests, use the testF()
macro like this:
class CustomTestOnce: public TestOnce {
protected:
// optional
void setup() override {
TestOnce::setup();
...setup code...
}
// optional
void teardown() override {
...teardown code...
TestOnce::teardown();
}
void assertCustomStuff() {
...common code...
}
int sharedValue;
};
testF(CustomTestOnce, calculate) {
...test code here...
}
No constructor for CustomTestOnce
needs to be defined.
Underneath the covers, the testF()
macro creates a subclass named
CustomTestOnce_calculate
which inherits from TestOnce
class. The test code
becomes the body of the CustomTestOnce_calculate::once()
method. The name of
this test has the class name prepended, so it is CustomTestOnce_calculate
,
which prevents name collision with other testF()
tests with the same
name using a different test fixture class.
To define a continuous test, use the testingF()
macro like this:
class CustomTestAgain: public TestAgain {
protected:
// optional
void setup() override {
TestAgain::setup();
...setup code...
}
// optional
void teardown() override {
...teardown code...
TestAgain::teardown();
}
void assertCustomStuff() {
...common code...
}
int sharedValue;
};
testingF(CustomTestAgain, calculate) {
...test code here...
}
Similarly, the testingF()
macro creates a subclass named
CustomTestAgain_calculate
, and the test code becomes the body of the
CustomTestAgain_calculate::again()
method.
See examples/fixtures/fixtures.ino
to see a working example of the testF()
macro.
ArduinoUnit Compatibility: The testF()
and testingF()
macros,
and the teardown()
virtual method are available only in AUnit (and Google
Test), not ArduinoUnit.
AUnit (like ArduinoUnit and Google Test) does not use C++ exceptions. Instead,
the various assertXxx()
macros perform an early return
if the condition
evaluates to false
. That means that the assertions can only bail out of the
current method, not the calling method. If you are using test fixtures, and
create a shared custom assert function, e.g. the assertCustomStuff()
method
above, any assertXxx()
statements in shared method will bail out of that
method only. The statement after the assertCustomStuff()
will continue to
execute.
In other words, in the following example, if the assertCustomStuff()
fails,
then assertMoreStuff()
inside testF()
will execute:
class CustomTestOnce: public TestOnce {
protected:
void assertCustomStuff() {
assertEqual(sharedValue, 3);
// This will not execute if the assertEqual() above fails.
assertLess(...);
}
void assertMoreStuff() {
assertEqual(...);
}
int sharedValue;
};
// DON'T DO THIS
testF(CustomTestOnce, dontDoThis) {
assertCustomStuff();
// This will execute even if assertCustomStuff() fails.
assertMoreStuff();
}
// DO THIS INSTEAD
testF(CustomTestOnce, doThis) {
assertNoFatalFailure(assertCustomStuff());
assertNoFatalFailure(assertMoreStuff());
}
The solution is to use the assertNoFatalFailure(statement)
macro which checks
whether the inner statement
returned with a fatal assertion. If so, then it
returns immediately, preventing execution from continuing to the code that
follows. This macro is modeled after the
ASSERT_NO_FATAL_FAILURE(statement)
macro in Google Test that provides the same functionality.
The following methods from ArduinoUnit have also been implemented:
checkTestDone(name)
checkTestNotDone(name)
checkTestPass(name)
checkTestNotPass(name)
checkTestFail(name)
checkTestNotFail(name)
checkTestSkip(name)
checkTestNotSkip(name)
checkTestExpire(name)
[*]checkTestNotExpire(name)
[*]assertTestDone(name)
assertTestNotDone(name)
assertTestPass(name)
assertTestNotPass(name)
assertTestFail(name)
assertTestNotFail(name)
assertTestSkip(name)
assertTestNotSkip(name)
assertTestExpire(name)
[*]assertTestNotExpire(name)
[*]
The checkTestXxx()
methods check the status of the test named name
and returns a bool
. The execution continues even if false
.
The assertTestXxx()
methods stops the unit test if the status check
returns false
, and prints assertion messages that look like this:
AUnitTest.ino:366: Assertion passed: Test slow_pass is done.
AUnitTest.ino:372: Assertion passed: Test slow_pass is not failed.
AUnitTest.ino:448: Assertion passed: Test slow_skip is skipped.
AUnitTest.ino:451: Assertion passed: Test slow_skip is not timed out.
(The human readable version of being expired
will always be timed out
or
not timed out
on the Serial
output.)
The following macros define extern
references to test case objects which live
in other .cpp
files. These are required for the above meta assertions if the
test cases are defined in another file:
externTest()
externTesting()
externTestF()
externTestingF()
ArduinoUnit Compatibility: The methods marked by [*] are only
available in AUnit. Also, the assertion messages are different. ArduinoUnit
reuses the format used by the assertXxx()
macros, so prints something like
the following:
AUnitTest.ino:439: Assertion passed: (test_slow_skip_instance.state=2) >= (Test::DONE_SKIP=2).
AUnit has a separate message handler to print a customized message for the assertTestXxx() meta assertion macros.
The following macros can be used inside the body of test()
or testing()
macro to terminate a test unconditionally. Each macro prints a short message,
and returns immediately from the test, much like an assertXxx()
macro that
fails.
passTestNow()
[*]failTestNow()
[*]skipTestNow()
[*]expireTestNow()
[*]
The messages look like:
AUnitTest.ino:360: Status passed.
AUnitTest.ino:378: Status failed.
AUnitTest.ino:380: Status skipped.
AUnitTest.ino:391: Status timed out.
The following methods on the Test
class also set the status
of the test, but
these methods do not print any messages (which makes debugging difficult) and
they do not terminate the test immediately.
pass()
- test passedfail()
- test failedskip()
- test skippedexpire()
- test timed out [*]
In most cases, the failTestNow()
, skipTestNow()
and expireTestNow()
macros
are more useful than the equivalent methods in the Test
class. However, in a
testing()
loop test, the pass()
method is probably better than the
passTestNow()
macro because we usually don't want to see an error message
from a passing test.
ArduinoUnit Compatibility: The method(s) marked by [*] are only available in AUnit.
The following methods are defined at the Test
base class level:
setup()
teardown()
The TestOnce
class defines:
once()
The TestAgain
class defines:
again()
ArduinoUnit Compatibility: These are functionally the same as ArduinoUnit
except with different class names. Instead of Test
use TestAgain
. Instead
of Test::loop
use TestAgain::again()
. ArduinoUnit does not support a
teardown()
method.
We run the test cases in the global loop()
method by calling
TestRunner::run()
. The tests are sorted according to the name of the test
given in the argument in the test()
or testing()
macro.
Each call to the run()
method causes one test case to run and be resolved. The
next call to run()
executes the next test case. This design allows the
loop()
method to perform a small amount of work and return periodically to
allow the system to perform some actions. On some systems, such as the ESP8266,
an error is generated if loop()
takes too much CPU time.
...
void loop() {
TestRunner::run();
}
ArduinoUnit Compatibility: This is equivalent to called Test::run()
in
ArduinoUnit. AUnit sorts the tests in the same way as ArduinoUnit. In
ArduinoUnit, each call to Test::run()
will process the entire list of
currently active test cases. In AUnit, each call to TestRunner::run()
performs
only a single test case, then returns.
Six filtering methods are available on the TestRunner
class:
TestRunner::include(pattern)
- prefix matchTestRunner::include(testClass, pattern)
- prefix matchTestRunner::exclude(pattern)
- prefix matchTestRunner::exclude(testClass, pattern)
- prefix matchTestRunner::includesub(substring)
- substring match (v1.6)TestRunner::excludesub(substring)
- substring match (v1.6)
These methods are called from the global setup()
method, for example:
void setup() {
TestRunner::include("looping*");
TestRunner::exclude("CustomTestAgain", "*");
TestRunner::include("CustomTestAgain", "test*");
TestRunner::include("CustomTestAgain", "test*");
TestRunner::includesub("net");
TestRunner::excludesub("net");
...
}
Excluded tests bypass their Test::setup()
and Test::teardown()
methods and
terminate immediately. For the purposes of reporting, excluded tests are counted
as "skipped".
The 2-argument versions of include()
and exclude()
correspond to the
2 arguments of testF()
and testingF()
.
The filtering methods are also available as command line flags and arguments
(--include
, --exclude
, --includesub
--excludesub
) if the test
program is compiled using EpoxyDuino under a Unix-like environment. See
the EpoxyDuino section below.
Implicit Exclude All: If the first filtering request is an "include" (i.e.
include(pattern)
, include(testClass, pattern)
, includesub(substring)
),
all tests are excluded by default initially, instead of being included by
default. Otherwise, the first "include" statement would have no effect.
ArduinoUnit Compatibility:
The equivalent versions in ArduinoUnit are Test::exclude()
and
Test::include()
The matching algorithm in AUnit is not as powerful as the one
in ArduinoUnit. AUnit supports only a single wildcard character *
and that
character can appear only at the end if it is present. For example, the
following are accepted:
TestRunner::exclude("*");
TestRunner::include("f*");
TestRunner::exclude("flash_*");
TestRunner::include("looping*");
TestRunner::include("CustomTestOnce", "flashTest*");
AUnit provides 2-argument versions of include()
and exclude()
The default output printer is the Serial
instance. This can be
changed using the TestRunner::setPrinter()
method:
#include <AUnit.h>
using aunit::TestRunner;
...
void setup() {
Serial1.begin(...);
TestRunner::setPrinter(&Serial1);
...
}
void loop() {
TestRunner::run();
}
ArduinoUnit Compatibility:
This is the equivalent of the Test::out
static member variable in
ArduinoUnit.
The default verbosity of the test results can be controlled using the
TestRunner::setVerbosity()
method:
#include <AUnit.h>
using aunit::TestRunner;
using aunit::Verbosity;
...
void setup() {
...
TestRunner::setVerbosity(Verbosity::kAll);
...
}
Every test is assigned this default verbosity just before its Test::setup()
is called. A unit test can choose to modify the verbosity calling one of the
following methods:
void enableVerbosity(uint8_t verbosity);
- enables the given verbosity, retaining all the others
void disableVerbosity(uint8_t verbosity);
- disables the given verbosity, retaining all the others
at the beginning of the test definition, like this:
test(enable_assertion_passed_messages) {
enableVerbosity(Verbosity::kAssertionPassed);
...
}
The values of verbosity
are defined by the static constants of the Verbosity
utility class:
Verbosity::kAssertionPassed
Verbosity::kAssertionFailed
Verbosity::kTestPassed
Verbosity::kTestFailed
Verbosity::kTestSkipped
Verbosity::kTestExpired
Verbosity::kTestRunSummary
Verbosity::kAssertionAll
- enables all assert messagesVerbosity::kTestAll
- same as
(kTestPassed | kTestFailed | kTestSkipped | kTestExpired)
- same as
Verbosity::kDefault
- same as
(kAssertionFailed | kTestAll | kTestRunSummary )
- same as
Verbosity::kAll
- enables all messagesVerbosity::kNone
- disables all messages
ArduinoUnit Compatibility: The following ArduinoUnit variables do not exist:
Test::min_verbosity
Test::max_verbosity
The bit field constants have slightly different names:
TEST_VERBOSITY_TESTS_SUMMARY
->Verbosity::kTestRunSummary
TEST_VERBOSITY_TESTS_FAILED
->Verbosity::kTestFailed
TEST_VERBOSITY_TESTS_PASSED
->Verbosity::kTestPassed
TEST_VERBOSITY_TESTS_SKIPPED
->Verbosity::kTestSkipped
TEST_VERBOSITY_TESTS_ALL
->Verbosity::kTestAll
TEST_VERBOSITY_ASSERTIONS_FAILED
->Verbosity::kAssertionFailed
TEST_VERBOSITY_ASSERTIONS_PASSED
->Verbosity::kAssertionPassed
TEST_VERBOSITY_ASSERTIONS_ALL
->Verbosity::kAssertionAll
TEST_VERBOSITY_ALL
->Verbosity::kAll
TEST_VERBOSITY_NONE
->Verbosity::kNone
- {no equivalent} <-
Verbosity::kDefault
- {no equivalent} <-
Verbosity::kTestExpired
AUnit suffers from the same compiler/preprocessor bug as ArduinoUnit that causes
the built-in __LINE__
macro to be off by one. The solution is to add:
#line 2 {file.ino}
as the first line of a unit test sketch.
ArduinoUnit Compatibility: This problem is identical to ArduinoUnit.
The various assertXxx()
macros in AUnit print a message upon pass or fail. For
example, if the assertion was:
int expected = 3;
int counter = 4;
assertEquals(expected, counter);
The error message (if enabled, which is the default) is:
AUnitTest.ino:134: Assertion failed: (3) == (4).
Asserts with bool
values produce customized messages, printing "true" or
"false" instead of using the Print class default conversion to int
:
assertEquals(true, false);
AUnitTest.ino:134: Assertion failed: (true) == (false).
Similarly, the assertTrue()
and assertFalse()
macros provide more customized
messages:
bool ok = false;
assertTrue(ok);
AUnitTest.ino:134: Assertion failed: (false) is true.
and
bool ok = true;
assertFalse(ok);
AUnitTest.ino:134: Assertion failed: (true) is false.
ArduinoUnit Compatibility:
ArduinoUnit captures the arguments of the assertEqual()
macro
and prints:
AUnitTest.ino:134: Assertion failed: (expected=3) == (counter=4).
Each capture of the parameter string consumes flash memory space. If the unit
test has numerous assertXxx()
statements, the flash memory cost is expensive.
AUnit omits the parameters to reduce flash memory space by about 33%.
The messages for asserts with bool values are customized for better clarity (partially to compensate for the lack of capture of the string of the actual arguments, and are different from ArduinoUnit.
If you use the verbose header:
#include <AUnitVerbose.h>
the assertion message will contain the string fragments of the arguments
passed into the assertXxx()
macros, like this:
AUnitTest.ino:134: Assertion failed: (expected=3) == (counter=4).
AUnitTest.ino:134: Assertion failed: (ok=false) is true.
instead of:
AUnitTest.ino:134: Assertion failed: (3) == (4).
AUnitTest.ino:134: Assertion failed: (false) is true.
ArduinoUnit Compatibility: _As of v1.7, the assertion message format is compatible with the vim editor and other Linux/MacOS/Unix tools, and no longer compatible with ArduinoUnit
As each test case finishes, the TestRunner
prints out the summary of the test
case like this:
Test bad failed.
Test looping_pass passed.
Test looping_skip skipped.
Test looping_until timed out.
ArduinoUnit Compatibility: These are identifcal to ArduinoUnit, except that the "timed out" status is new to AUnit. See Test Timeout section below.
At the end of the test run, the TestRunner
prints out the summary
of all test cases, like this:
TestRunner duration: 0.05 seconds.
TestRunner summary: 12 passed, 0 failed, 2 skipped, 1 timed out, out of 15 test(s).
ArduinoUnit Compatibility: The message format is slightly different than
ArduinoUnit. I changed "Test summary" to "TestRunner summary" because the former
looks identical to the message that could have been printed by a test(summary)
test case. AUnit also adds information about tests which timed out. See below.
ArduinoUnit Compatibility: Only available in AUnit.
From my experience, it seems incredibly easy to write a testing()
test case
which accidentally runs forever because the code forgets to call an explicit
pass()
, fail()
or skip()
.
The TestRunner
in AUnit applies a timeout value to all the test cases that it
runs. The default timeout is 10 seconds. A timeout value of 0
means an
infinite timeout, which means that the testing()
test case may run
forever. The value of the timeout is stored as a uint16_t
type, so the maximum
timeout is 65535 seconds or a bit over 18 hours.
Currently, the timeout value is global to all test cases. Time out values for
individual tests cannot be set independently. If a test does not finish before
that time, then the test is marked as timed out
(internally implemented by the
Test::expire()
method) and a message is printed like this:
Test looping_until timed out.
The timeout value can be changed by calling the static
TestRunner::setTimeout()
method. Here is an example that sets the timeout to
30 seconds instead:
void setup() {
...
TestRunner::setTimeout(30);
...
}
ArduinoUnit Compatibility: Only available in AUnit.
It may be possible to run simple unit tests written using Google Test API on an Arduino platform by using the aunit/contrib/gtest.h adapter. This adapter layer provides a number of Google Test macros which map to their equivalent macros in AUnit:
ASSERT_EQ(e, a)
-assertEqual()
ASSERT_NE(e, a)
-assertNotEqual()
ASSERT_LT(e, a)
-assertLess()
ASSERT_GT(e, a)
-assertMore()
ASSERT_LE(e, a)
-assertLessOrEqual()
ASSERT_GE(e, a)
-assertMoreOrEqual()
ASSERT_STREQ(e, a)
-assertEqual()
ASSERT_STRNE(e, a)
-assertNotEqual()
ASSERT_STRCASEEQ(e, a)
-assertStringCaseEqual()
ASSERT_STRCASENE(e, a)
-assertStringCaseNotEqual()
ASSERT_TRUE(x)
-assertTrue()
ASSERT_FALSE(x)
-assertFalse()
ASSERT_NEAR(e, a, error)
-assertNear()
To use the gtest.h
adapter, include the following headers:
#include <AUnit.h>
#include <aunit/contrib/gtest.h>
or
#include <AUnitVerbose.h>
#include <aunit/contrib/gtest.h>
Each unit test is an independent *.ino
program. You can run it using your
Arduino IDE. But there are 2 command line tools that can be used to run them.
The auniter.sh
script used to be part of this project, but now lives in
its own AUniter (https://github.com/bxparks/AUniter) project. The auniter.sh
script is a wrapper around:
Using auniter.sh
, you can compile, upload and validate multiple AUnit tests on
multiple Arduino boards using a single command.
The script can monitor the serial port and determine if the unit test passed or
failed, and it will print out a summary of all unit tests at the end. Full
details are given in the AUniter project, but here are some quick examples
copied from the AUniter/README.md
file:
$ auniter envs
- list the environments configured in the
auniter.ini
config file
- list the environments configured in the
$ auniter ports
- list the available serial ports and devices
$ auniter verify nano Blink.ino
- verify (compile)
Blink.ino
using theenv:nano
environment
- verify (compile)
$ auniter upload nano:/dev/ttyUSB0 Blink.ino
- upload
Blink.ino
to theenv:nano
target environment connected to/dev/ttyUSB0
- upload
$ auniter test nano:USB0 BlinkTest.ino
- compile and upload
BlinkTest.ino
using theenv:nano
environment, upload it to the board at/dev/ttyUSB0
, then validate the output of the AUnit unit test
- compile and upload
$ auniter test nano:USB0 BlinkTest/ ClockTest/
- upload and verify the 2 unit tests (
BlinkTest/BlinkTest.ino
,ClockTest/ClockTest.ino
) on the target environment (env:nano
)
- upload and verify the 2 unit tests (
$ auniter upmon nano:USB0 Blink.ino
- upload the
Blink.ino
sketch and monitor the serial port using a user-configurable terminal program (e.g.picocom
) on/dev/ttyUSB0
- upload the
Instead of running the unit tests on the actual microcontrollers themselves, you
can compile and execute AUnit unit tests natively on Linux or MacOS machines
using the EpoxyDuino (https://github.com/bxparks/EpoxyDuino) project.
EpoxyDuino provides a minimal Arduino programming environment that is usually
sufficient to compile and run AUnit test units on the Unix host machine. It
relies on the native C++ compiler, GNU Make, and Makefile
files for each
*.ino
unit test like this:
APP_NAME := SampleTest
ARDUINO_LIBS := AUnit
include ../../../EpoxyDuino/EpoxyDuino.mk
The unit test is compiled into a binary (SampleTest.out
) using the `make
command, and the binary can be executed like this:
$ make
$ ./SampleTest.out
Here are a few tips when writing unit tests to run under EpoxyDuino:
Delay(1000)
For real Arduino boards, you get more reliable unit tests if you add a
delay(1000)
at the start of the program. For EpoxyDuino, this is not
necessary, so I recommend calling this only on real Arduino boards, like this:
void setup() {
#if ! defined(EPOXY_DUINO)
delay(1000); // Wait for stability on some boards, otherwise garage on Serial
#endif
Serial.begin(115200);
while (! Serial); // Wait until Serial is ready - Leonardo/Micro
#if defined(EPOXY_DUINO)
Serial.setLineModeUnix(); // use Unix line terminator instead of DOS
#endif
...
}
void loop() {
aunit::TestRunner::run();
}
Exit() Status Code
On real Arduino boards, the unit test (or any program for that matter) never
terminates. The loop()
function executes forever. On Linux or MacOS using
EpoxyDuino, the test program will terminate at the end through the
exit()
function. If the tests are successful (i.e. passing or skipped), it
will call exit(0)
. If there are any failing tests (i.e. failed or timed out),
it will call exit(1)
.
(Added in v1.6)
The standard Arduino environment does not provide command line arguments, since
a microcontroller does not normally provide a command line environment. However,
if the AUnit test program is compiled under EpoxyDuino, the standard Unix
command line parameters (argc
and argv
) become available through the
extern int epoxy_argc
and extern const char* const* epoxy_argv
global
variables. These allow the TestRunner
class to provide command line flags and
arguments as follows:
$ ./test.out --help
Usage: ./test.out [--help] [--include pattern,...] [--exclude pattern,...]
[--includesub substring,...] [--excludesub substring,...]
[--] [substring ...]
Example, the following runs all tests with substring "net" or "led" in its name, and skips all others:
$ ./test.out net led
Flags:
--include pattern,...
- Comma-separated list of patterns to pass to the
TestRunner::include(pattern)
method
- Comma-separated list of patterns to pass to the
--exclude pattern,...
- Comma-separated list of patterns to pass to the
TestRunner::exclude(pattern)
method
- Comma-separated list of patterns to pass to the
--includesub substring,...
- Comma-separated list of substrings to pass to the
TestRunner::includesub(substring)
method
- Comma-separated list of substrings to pass to the
--excludesub substring,...
- Comma-separated list of substrings to pass to the
TestRunner::excludesub(substring)
method
- Comma-separated list of substrings to pass to the
Arguments:
- Any Space-separated list of words after the optional flags are passed to
the
TestRunner::includesub(substring)
method.
The command line flags and arguments are processed after any hardcoded calls
to TestRunner::include()
and TestRunner::exclude()
methods in the global
setup()
method.
The flags and command line arguments are processed in order of appearance on the command line.
Similar to the hardcoded calls to TestRunner::include()
and
TestRunner::exclude()
, if the first command line flag is an --include
or
--includesub
, then all tests are excluded by default initially. Otherwise,
the first include flag would have no effect.
There are several ways to incorporate AUnit into a continuous integration system. At the infrastructure level, you can use either a system like Jenkins running on a local machine, or use a cloud-based continuous integration system like GitHub Actions. For each of those infrastructures, you can choose to use the Arduino IDE or CLI build tools, or you can use EpoxyDuino to compile and run against a Linux or MacOS environment.
The option matrix looks like this:
+----------------+-------------------------+------------------------+
|\ CI environment| | |
| \____________ | Local (e.g. Jenkins) | Cloud (e.g. GitHub) |
| \ | | |
| Build tool \| | |
+----------------+-------------------------+------------------------+
| | * can execute tests | * verify compile-only |
| | on microcontroller | * cannot actually run |
| Arduino | * complex setup and | tests on |
| IDE/CLI | * maintenance | microcontroller |
| | * can be slow | * complex installation |
| | * (not recommended) | and setup |
| | | * (unverified) |
+----------------+-------------------------+------------------------+
| | * verify execution on | * verify execution on |
| | Unix environment | Unix environment |
| EpoxyDuino | * faster than IDE/CLI | * simple setup and |
| | * complex setup and | maintenance |
| | and maintenance | * very fast |
| | * (unnecessary) | * (recommended) |
+----------------+-------------------------+------------------------+
The big advantage of using the Arduino IDE/CLI to run the AUnit unit tests is
that you can compile them using the precise compiler and tool chain that will be
used against the specific microcontroller that you are interested in verifying.
There can be subtle compiler differences (e.g. size of int
) or differences in
how the Arduino programming environment was implemented (e.g. availability of
the FPSTR()
macro). If you use a cloud-based CI infrastructure, then you must
install the complete Arduino IDE/CLI environment into your cloud test runner.
Although I think it's theoretically possible, I have never actually verified
that this can be done.
This setup is described in Continuous Integration with Jenkins, and it worked reasonably well for small number of unit tests. The problem is that the Arduino IDE is far too slow when the number of unit tests become non-trivial. And it also takes too much effort to maintain the local Jenkins infrastructure. The Jenkins environment seems to be brittle due to its complexity of all of its moving parts. This is the only environment where you can connect a real Arduino microcontroller to the local machine and have the unit tests run on the actual microcontroller. Perhaps for certain situations, running the unit tests on actual hardware is a requirement. But for most people, I no longer recommend this environment.
Once the Jenkins environment is up and running, I have verified that it is easy to run the unit tests using EpoxyDuino, since it needs just a C++ compiler and GNU Make. Things will compile and run a lot faster than using the Arduino IDE/CLI. However, this combination suffers from the same problem of maintaining the Jenkins environment. If the unit tests are running in an Unix environment anyway, it seems far easier to just run them in the cloud. So I don't recommend using this setup. Just use a cloud CI provider as described below.
A cloud-based continuous integration service like GitHub
Actions is easy to setup for
EpoxyDuino. Often the C++ compiler and GNU make
tools are already installed
in the Docker container used by the CI system. The only additional setup is
to install EpoxyDuino, AUnit and other dependent Arduino libraries.
Here are some example YAML files for GitHub Actions:
- https://github.com/bxparks/AceTime/tree/develop/.github/workflows
- https://github.com/bxparks/AceButton/tree/develop/.github/workflows
- https://github.com/bxparks/AceRoutine/tree/develop/.github/workflows
- https://github.com/bxparks/AceCRC/tree/develop/.github/workflows
In the various aunit_tests.yml
files, the Setup
step installs the various
dependent libraries using the git clone
command, for example:
git clone https://github.com/bxparks/EpoxyDuino
git clone https://github.com/bxparks/AUnit
These commands install the default branch for those repositories, which for most
of my libraries will be the develop
branch. This makes sense for me because I
want the unit tests to run against the latest commits. However, for many others,
it is probably better to use the master
branch because it contains the stable
releases:
git clone --branch master https://github.com/bxparks/EpoxyDuino
git clone --branch master https://github.com/bxparks/AUnit
Using EpoxyDuino with a cloud CI provider (like GitHub Actions) is my recommended configuration for running AUnit tests because it is easy to setup and maintain and the tests run fast.
Collection of useful tidbits.
When using test fixtures with the testF()
and testingF()
macros, it's often
useful to create helper assertions, such as the assertCustomStuff()
below.
Debugging such assertion statements can be tricky. I've found that turning on
messages for successful assertions (with a
enableVerbosity(Verbosity::kAssertionPassed)
) statement can be very helpful:
class CustomTestOnce: public TestOnce {
protected:
// optional
void setup() override {
TestOnce::setup();
...setup code...
}
// optional
void teardown() override {
...teardown code...
TestOnce::teardown();
}
void assertCustomStuff() {
assertEqual(...);
...
for (...) {
...
assertEqual(...);
...
}
}
int sharedValue;
};
testF(CustomTestOnce, calculate) {
enableVerbosity(Verbosity::kAssertionPassed);
...test code here...
assertCustomStuff();
}
To support test fixtures in a more natural way, the class hierarchy in AUnit is slightly different than ArduinoUnit. In ArduinoUnit we have a two level hierarchy:
Test ::loop()
^
|
TestOnce ::once()
In AUnit, the functionality that supports the testing()
macro has been
migrated to a separate class called TestAgain
, like this:
Test ::loop()
^
|
Assertion
^
|
MetaAssertion
^ ^
/ \
/ \
TestAgain TestOnce
::again() ::once()
Normally, deep inheritance hierarchies like this should be avoided. However,
placing the Assertion
and MetaAssertion
classes inside the Test
hierarchy
allowed those assertion statements to have access to the internal states of the
Test
instance. This made certain features (like the early return upon delayed
failure) slightly easier to implement. For the most part, the end-users can
ignore the existence of the Assertion
and MetaAssertion
classes and think of
this as a simple 2-level inheritance tree.
There is a school of thought which says that unit tests should test only the
publically exposed methods of a class or library. I agree mostly with that
sentiment, but not rigidly. I think it is sometimes useful to write unit tests
for protected
or private
methods. For example, when creating a chain of
small helper methods, which build up to larger publically exposed methods, it is
extremely useful to write unit tests for the helper methods in isolation.
Normally those helper methods would be private
because they are used
only within that class, and we don't want to expose them to the public API. One
option is to make them public
but add a comment in the function to say that it
is exposed only for testing purposes. This does not seem satisfactory because
users will tend to ignore such comments if the helper functions are useful.
I think a better way is to keep the helper functions private
but make
the unit tests a friend class
of the target class. The syntax for doing this
can be tricky, it took me a number of attempts to get this right, especially if
when using namespaces for the target class:
//------------------- Target.h -------------
// Auto-generated test class names.
class Test_helper;
class TargetSuite_helper;
class TargetTest_helper;
namespace mylib {
class Target {
public:
void publicMethod() {
...
int a = helper();
...
}
private:
// Must have the global scope operator '::'
friend class ::Test_helper;
friend class ::TargetSuite_helper;
friend class ::TargetTest_helper;
static int helper() {...}
};
}
//------------------- TargetTest.ino -------------
#include <AUnit.h>
#include "Target.h"
using namespace aunit;
using namespace mylib;
test(helper) {
assertEqual(1, Target::helper(...));
}
test(TargetSuite, helper) {
assertEqual(1, Target::helper(...));
}
class TargetTest: public TestOnce {
...
};
testF(TargetTest, helper) {
assertEqual(1, Target::helper(...));
}
The tricky part is that Target.h
must have forward declarations of the various
auto-generated AUnit test classes. And within the Target
class itself, the
friend
declarations need to have a global scope ::
specifier before the name
of the test class.
The MemoryBenchmark directory collects the flash and static RAM usage of the AUnit library on various microcontroller boards, and renders the results as tables embedded in the README.md file. Here are 2 highlights:
Arduino Nano (8-bit)
+---------------------------------------------------------------------+
| Functionality | flash/ ram | delta |
|----------------------------------------+--------------+-------------|
| Baseline | 1586/ 185 | 0/ 0 |
|----------------------------------------+--------------+-------------|
| AUnit Single Test | 4456/ 366 | 2870/ 181 |
| AUnit Single Test Verbose | 4500/ 366 | 2914/ 181 |
+---------------------------------------------------------------------+
ESP8266 (32-bit)
+---------------------------------------------------------------------+
| Functionality | flash/ ram | delta |
|----------------------------------------+--------------+-------------|
| Baseline | 264949/27984 | 0/ 0 |
|----------------------------------------+--------------+-------------|
| AUnit Single Test | 268021/28148 | 3072/ 164 |
| AUnit Single Test Verbose | 268081/28148 | 3132/ 164 |
+---------------------------------------------------------------------+
AUnit consumes as much as 65% less flash memory than ArduinoUnit 2.2 on an AVR platform (e.g. Arduino UNO, Nano), and 30% less flash on the Teensy-ARM platform (e.g. Teensy LC ). (ArduinoUnit 2.3 reduces the flash memory by 30% or so, which means that AUnit can still consume significantly less flash memory.)
Here are the resource consumption (flash and static) numbers
from
AceButtonTest
containing 26 test cases using 331 assertXxx()
statements, compiled using AUnit and ArduinoUnit 2.2 on 5 different
microcontrollers:
---------------------------+---------+-------------+-------------+
Platform (resource) | Max | ArduinoUnit | AUnit |
---------------------------+---------+-------------+-------------|
Arduino Nano (flash) | 30720 | 54038 | 18928 |
Arduino Nano (static) | 2048 | 1061 | 917 |
---------------------------+---------+-------------+-------------|
Teensy LC (flash) | 63488 | 36196 | 26496 |
Teensy LC (static) | 8192 | 2980 | 2780 |
---------------------------+---------+-------------+-------------|
Teensy 3.2 (flash) | 262144 | 51236 | 37920 |
Teensy 3.2 (static) | 65536 | 5328 | 5236 |
---------------------------+---------+-------------+-------------|
ESP8266 - ESP-12E (flash) | 1044464 | does not | 268236 |
ESP8266 - ESP-12E (static) | 81920 | compile | 33128 |
---------------------------+---------+-------------+-------------|
ESP8266 - ESP-01 (flash) | 499696 | does not | 268236 |
ESP8266 - ESP-01 (static) | 47356 | compile | 33128 |
---------------------------+---------+-------------+-------------+
Not all unit test sketches will experience a savings of 65% of flash memory with AUnit, but a savings of 30-50% seems to be common.
Tier 1: Fully supported
These boards are tested on each release:
- Arduino Nano clone (16 MHz ATmega328P)
- SparkFun Pro Micro clone (16 MHz ATmega32U4)
- Seeeduino XIAO M0 (SAMD21, 48 MHz ARM Cortex-M0+)
- STM32 Blue Pill (STM32F103C8, 72 MHz ARM Cortex-M3)
- Adafruit ItsyBitsy M4 (SAMD51, 120 MHz ARM Cortext-M4)
- NodeMCU 1.0 (ESP-12E module, 80 MHz ESP8266)
- WeMos D1 Mini (ESP-12E module, 80 MHz ESP8266)
- ESP32 dev board (ESP-WROOM-32 module, 240 MHz dual core Tensilica LX6)
Tier 2: Should work
These boards should work but I don't test them as often:
- Arduino Pro Mini (16 MHz ATmega328P)
- Mini Mega 2560 (Arduino Mega 2560 compatible, 16 MHz ATmega2560)
- Teensy LC (48 MHz ARM Cortex-M0+)
- Teensy 3.2 (96 MHz ARM Cortex-M4)
Tier 3: May work, but not supported
- Other 3rd party SAMD21 and SAMD51 boards may work if their board software uses the traditional Arduino API, instead of the ArduinoCore-API
Tier Blacklisted
The following boards are not supported and are explicitly blacklisted to allow the compiler to print useful error messages instead of hundreds of lines of compiler errors:
- Any platform using the
ArduinoCore-API, such as:
- Arduino-branded megaAVR using
ArduinoCore-megaavr
- Nano Every
- Arduino-branded SAMD21 or SAMD51 boards using
ArduinoCore-samd after
version >= 1.8.10
- MKRZero
- Nano 33 IoT
- Raspberry Pi Pico (RP2040) using Arduino-Pico
- Arduino-branded megaAVR using
ArduinoCore-megaavr
This library was validated using:
- Arduino IDE 1.8.19
- Arduino CLI 0.33.0
- Arduino AVR Boards 1.8.6
- Arduino SAMD Boards 1.8.9 (versions >= 1.8.10 not supported)
- SparkFun AVR Boards 1.1.13
- SparkFun SAMD Boards 1.8.6
- STM32duino 2.5.0
- ESP8266 Arduino 3.0.2
- ESP32 Arduino 2.0.9
- Teensyduino 1.57
This library is not compatible with:
- Any platform using the ArduinoCore-API, for example:
(See Issue #56 and Issue #66).
It should work with PlatformIO but I have not tested it extensively.
The library works under Linux or MacOS (using both g++ and clang++ compilers) using the EpoxyDuino (https://github.com/bxparks/EpoxyDuino) emulation layer.
I use Ubuntu 20.04 for the vast majority of my development. I expect that the library will work fine under MacOS and Windows, but I have not tested them.
If you have any questions, comments, or feature requests for this library, please use the GitHub Discussions for this project. If you have bug reports, please file a ticket in GitHub Issues. Feature requests should go into Discussions first because they often have alternative solutions which are useful to remain visible, instead of disappearing from the default view of the Issue tracker after the ticket is closed.
Please refrain from emailing me directly unless the content is sensitive. The problem with email is that I cannot reference the email conversation when other people ask similar questions later.
- Created by Brian T. Park ([email protected]).
- The Google Test adapter (
gtest.h
) was created by Chris Johnson ([email protected]). - @brewmanz increased the maximum allowed value of
TestRunner::setTimeout()
from 255 seconds to 65535 seconds (18.2 hours). (See Issue #57). - The design and syntax of many macros (e.g.
test()
,assertXxx()
) were borrowed from the ArduinoUnit project to allow AUnit to be almost a drop-in replacement. Many thanks to the ArduinoUnit team for creating such an easy-to-use API.