Tests and testing process#

The testing process uses a Phoenix-RTOS testing framework written in Python. The framework provides an environment for running both unit and functional tests. Unit tests are written using Unity and the process is adapted to it.

Tests, in general, are launched using test runner, placed in phoenix-rtos-tests repository. Read more about the reference project repository here.

High level overview of the test runner#

Testing parts of operating systems involves interacting with emulators and physical devices. To simplify this process, we have developed a framework called the test runner, which automates running tests and facilitates their creation.

Most tests consist of three components:

  • An executable binary that is being tested.

  • A Python harness that supervises the test execution and interacts with the test.

  • A YAML configuration file that connects the harnesses to binaries and contains additional information, such as the list of target devices on which the test should be run, whether it is a nightly test, or any extra arguments to be passed to the test harness.

The test runner iterates over the YAML configurations, loads them, and initiates the execution of tests based on the provided information. A test campaign is run for a specific target device specified by the user. The test runner performs the following steps:

  • Based on the target device, it spawns an emulator process or attempts to connect to a physical device using the provided serial port.

  • If the target device is a physical device, the runner flashes the image (depending on --no-flash argument).

  • The test campaign is then executed. For each test, the corresponding executable is flashed onto the device (if needed), the executables are executed, and the harnesses supervise the test execution. If something does not go as expected, the runner tries to provide as much information as possible based on the encountered error.

  • Finally, the overall test result is printed.

The test runner does not build the project for you. You need to add the test parameter to the build script to ensure that the tests will be built. If you want to test the phoenix-rtos-ports, make sure that the ports are built and include the LONG_TEST=y environment variable during the building process.

Some targets require extra steps to make them compatible with the test runner. For example, the ia32-generic-qemu target must be built with a custom syspage that includes the UART driver. Here is an example command to build an image with ports for the ia32-generic-qemu target:

TARGET=ia32-generic-qemu CONSOLE=serial LONG_TEST=y ./phoenix-rtos-build/build.sh all test

But most of the targets can be build using simple:

TARGET=target ./phoenix-rtos-build/build.sh core fs project image test.

Currently, only user space tests are supported.

How to write and run tests?#

One time setup#

Python packages required by the testing script have to be installed. To do this, create the virtual environment, change the directory to phoenix-rtos-tests and run the following command:

pip3 install -r requirements.txt

Example 1: Hello world#

In this section, we will provide a few basic examples to clarify how to create tests that can be run using the test runner.

Let’s start with the simplest example: a program that prints “Hello, world!” (this example is also included in the repository in the sample directory). We will need four components: a program written in C, a Makefile that builds the executable, a Python harness to test the program execution, and a YAML configuration file to connect everything together.

First, let’s create the executable to test. Create a directory called hello in the phoenix-rtos-tests/hello directory and place the hello.c file inside it:

#include <stdio.h>

int main(void)
{
	puts("Hello, world!");
	return 0;
}

Now, we need to ensure that it is built within our project. To do that, we define Makefile in the same directory:

NAME := hello
LOCAL_SRCS := hello.c

include $(binary.mk)

On most targets, programs located in the phoenix-rtos-tests directory will be built and included in the filesystem. We have the first component - the binary that we are going to test.

Next, let’s focus on creating the harness that supervises the test. The harness is a program written in Python, and its main task is to communicate with the test binary and read its output. For this purpose, we will use the pexpect library that is built-in inside the runner. Create a file named hello_harness.py in the phoenix-rtos-tests/hello directory and place the following code there:

from trunner.dut import Dut

# The "main" function for test is the harness. Every harness must define a function with this name.
def harness(dut: Dut): # Dut wraps pexpect object, but you should use it as a normal pexpect spawn object.
    expected = "Hello, world!"
    line = dut.readline().rstrip() # Read the line that has been printed by a device.
    assert line == expected, f"Line mismatch! Expected: {expected}, got: {line}" # Assure that the read line is `Hello, world!` message.

Notice that you don’t have to worry about following:

  • How the test is flashed into the board.

  • What kind of shell is running inside the device.

  • How the test binary should be executed (whether it is a filesystem or a syspage?).

All of these things are handled by the runner. However, it’s important to be aware that harnesses are not sandboxes. For example, if you send a SIGKILL signal, you may kill your test binary process and end up in a shell. We also support the “fast” path of test execution, which means if there is no need to restart the device, we don’t do it. Because of that, if you spawn any new processes on your own or create some resources, you must clean them up.

Now, let’s move on to the last step. We have the binary that we want to test, and we have defined the harness that ensures everything executes correctly. The next step is to define the YAML configuration file that connects them together.

Define a YAML configuration file called phoenix-rtos-tests/hello/test_config.yaml:

test:
  tests:
    - name: hello-world
      harness: hello_harness.py
      execute: hello

In this configuration, we start with the keyword test, which indicates that we are defining the configuration for the test. Then, we can define the main configuration that will be applied to all tests (which is not included here, but have it in mind, we will use it in other examples). Next, we define the tests using the tests keyword. In this example, the test consists of a name that the runner will use to print the results, the harness file, and the binary that will be executed. If we do not define any target, the runner will use the default list of targets on which the test will run.

Important note: Every YAML configuration must start with the test_ prefix.

Finally, we can verify that everything works correctly. First, let’s build the project:

TARGET=ia32-generic-qemu CONSOLE=serial ./phoenix-rtos-build/build.sh core fs project image test

Once the project is built, we are ready to run our test:

./phoenix-rtos-tests/runner.py --target ia32-generic-qemu -t phoenix-rtos-tests/hello/test_config.yaml

We should expect the following output:

Flashing an image to device...
Done!
phoenix-rtos-tests/hello/hello-world: OK
TESTS: 1 PASSED: 1 FAILED: 0 SKIPPED: 0

The output provides information that the device has been flashed correctly (even if it’s an emulator process), followed by the list of tests with their results. At the end, we get a summary of all the tests.

Now, let’s go back to the hello_harness.py and change the expected message:

# We removed a comma to force a failed assertion.
expected = "Hello world!"

Run tests one more time. Now the output should differ from the previous one:

Flashing an image to device...
Done!
phoenix-rtos-tests/hello/hello-world: FAIL
ASSERTION TRACEBACK (most recent call last):
  File "/home/user/phoenix-rtos-project/phoenix-rtos-tests/hello/hello_harness.py", line 7, in harness
    assert line == expected, f"Line mismatch! Expected: {expected}, got: {line}" # Assure that the read line is `Hello, world!` message.

ASSERTION MESSAGE:
Line mismatch! Expected: Hello world!, got: Hello, world!
TESTS: 1 PASSED: 0 FAILED: 1 SKIPPED: 0

As we can see, the test finished with a failure. As a result, the runner printed more information, including a trace back and an assertion message. Depending on the exception type, the runner will try to provide as much information as possible.

Example 2: unit tests using C#

In this section, we will explore another type of test: unit testing. When we talk about testing, unit testing often comes to mind, which is why the test runner has native support for unit testing. Similar to the previous section, we start by creating a C file named dummy.c located in the phoenix-rtos-tests/dummy directory. To write unit tests, we will use the modified Unity Test, a third party unit testing framework built for C.

#include <string.h>

#include "unity_fixture.h"

TEST_GROUP(dummy);

TEST_SETUP(dummy)
{
}

TEST_TEAR_DOWN(dummy)
{
}

TEST(dummy, memset)
{
	unsigned char buffer[2] = {0, 0};
	memset(buffer, 0xFF, 2);
	for (int i = 0; i < 2; i++) {
		TEST_ASSERT_EQUAL_CHAR(buffer[i], 0xFF);
	}
}

TEST(dummy, strlen)
{
	const char *s = "Dummy string";
	TEST_ASSERT_EQUAL(strlen(s), 12);
}

TEST(dummy, ok)
{
	TEST_PASS_MESSAGE("dummy test that always pass");
}

TEST_GROUP_RUNNER(dummy)
{
	RUN_TEST_CASE(dummy, memset);
	RUN_TEST_CASE(dummy, strlen);
	RUN_TEST_CASE(dummy, ok);
}

void runner(void)
{
	RUN_TEST_GROUP(dummy);
}

int main(int argc, char *argv[])
{
	UnityMain(argc, (const char**)argv, runner);
	return 0;
}

This test runs the test group called dummy, which consists of three tests: memset, strlen, and ok. If you want to learn more about what unity framework offers, you can visit their GitHub website.

As in the previous example, we need to have the executable binary as a result, so let’s create the following Makefile in the same location:

NAME := test-dummy
LOCAL_SRCS := dummy.c
DEP_LIBS := unity

include $(binary.mk)

The last step is to create the YAML configuration for the test. In the same location, create the YAML configuration file called test_dummy.yaml:

test:
  tests:
    - name: test-dummy
      type: unity
      execute: test-dummy

The configuration above is similar to the configuration we defined in the previous section. The main difference is lack of keyword harness, and instead, we use the keyword type with the string unity as value. This line instructs the runner to use internally defined harness to parse the output. Therefore, the runner works in the same manner, but instead defining the harness ourselves, we have ready-to-use harness that will be used when type: unity is specified.

Now, you can run the tests and expect the success:

./phoenix-rtos-tests/runner.py --target ia32-generic-qemu -t phoenix-rtos-tests/dummy/test_dummy.yaml
Flashing an image to device...
Done!
phoenix-rtos-tests/dummy/test-dummy: OK
TESTS: 1 PASSED: 1 FAILED: 0 SKIPPED: 0

To see more information, add the --verbose flag:

./phoenix-rtos-tests/runner.py --target ia32-generic-qemu -t phoenix-rtos-tests/dummy/test_dummy.yaml --verbose
Flashing an image to device...
Done!
phoenix-rtos-tests/dummy/test-dummy: OK
	TEST(dummy, memset) OK
	TEST(dummy, strlen) OK
	TEST(dummy, ok) OK
TESTS: 1 PASSED: 1 FAILED: 0 SKIPPED: 0

Finally, we can force runner to fail to see how the output will look like. Change the line in the dummy ok test to TEST_FAIL_MESSAGE("dummy test that always pass");. Rebuild the project and run tests one more time:

/phoenix-rtos-tests/runner.py --target ia32-generic-qemu -t phoenix-rtos-tests/dummy/test_dummy.yaml
Flashing an image to device...
Done!
phoenix-rtos-tests/dummy/test-dummy: FAIL
	TEST(dummy, ok) FAIL at phoenix-rtos-tests/dummy/dummy.c:27: dummy test that always pass
TESTS: 1 PASSED: 0 FAILED: 1 SKIPPED: 0

As we can see, the runner caught failed test and provided information about the failure.

Example 3: more advanced test case#

In this example, we will demonstrate what we can achieve using additional features of the runner and its YAML configurations. Without surprise, we create a new C program in the same location, which is the phoenix-rtos-tests/hello directory. Let’s name this file hello_arg.c and put the following content inside it:

#include <stdio.h>

int main(int argc, char *argv[])
{
	char word[10];

	if (argc <= 1) {
		puts("One argument is required! You passed zero.");
		return 1;
	}

	if (argc > 2) {
		puts("One argument is required! You passed more than one.");
		return 2;
	}

	printf("Hello, %s!\n", argv[1]);

	fgets(word, 10, stdin);
	printf("%s", word);

	return 0;
}

This program checks the number of arguments. If there is exactly one argument, it continues its execution; otherwise, the program ends with a message describing how many arguments must be passed. After that, it prints the argument that was passed within the greeting sentence. Finally, it reads input (up to 9 characters) from stdin and prints it out. Thus, we can pass arguments and provide the input to the test.

Of course, we need to add some lines to the Makefile:

NAME := test-hello-arg
LOCAL_SRCS := hello_arg.c

include $(binary.mk)

Now, let’s create hello_arg_harness.py in the same directory with the following content:

from trunner.dut import Dut

def harness(dut: Dut, **kwargs):
    argc = kwargs.get("argc", 1)
    input = kwargs.get("input", "Bye!")

    assert argc >= 0

    if argc < 1:
        dut.expect_exact("One argument is required! You passed zero.")
        return
    elif argc > 1:
        dut.expect_exact("One argument is required! You passed more than one.")
        return

    dut.expect(r"Hello, \w+!\r*\n")
    dut.sendline(input)
    dut.expect(rf"{input}\r*\n")

    if len(input) > 10:
        output = input[:9]
    else:
        output = input + r"\r*\n"

    dut.expect(output)

This harness differs from the previous one in the definition of the harness function. Instead of using only dut as an argument, we have added **kwargs in the definition. Using kwargs, we can pass user-defined input to the test by adding a kwargs dictionary to the YAML configuration or by passing them to the runner using the --kwargs argument. We will describe this in more detail later.

In the harness function, we read the arguments, ensuring they have default values if they are not passed. Then we perform different actions based on the argc argument. If argc differs from 1, we expect the error messages and finish the harness execution. If there is exactly one argument, we expect the greeting sentence and send the input that should be printed back to us.

Now, let’s create some test cases based on our harness using the YAML configuration file:

test:

  targets:
    value: [ia32-generic-qemu, armv7a9-zynq7000-qemu, host-generic-pc]

  nightly: true

  tests:
    - name: arg_zero
      execute: test-hello-arg
      harness: hello_arg_harness.py
      kwargs:
        argc: 0
      targets:
        exclude: [armv7a9-zynq7000-qemu]

    - name: arg_two
      execute: test-hello-arg arg1 arg2
      harness: hello_arg_harness.py
      kwargs:
        argc: 2
      targets:
        value: [ia32-generic-qemu]

    - name: arg_hello
      execute: test-hello-arg world
      harness: hello_arg_harness.py
      nightly: false
      kwargs:
        input: 'Adios!'

    - name: arg_hello_too_long_input
      execute: test-hello-arg space
      harness: hello_arg_harness.py
      kwargs:
        input: 'String that is too long!'

This time, the configuration is a bit longer and requires more explanation. Let’s start from the beginning. Before the keyword tests, we defined targets, which is a dictionary with allowed keys: value, include and exclude. As we mentioned in the previous sections, tests are allowed to run on a list of default targets. If we want to change the targets by defining a strict list of targets or including/excluding them, we can do exactly that by defining a targets dictionary.

As you can see, targets is used in the test scope. We call this scope the main config. The set of keywords that you can use in the main config is a subset of the keywords allowed to be used in a single test configuration (elements of the tests list). These keywords are: targets, type, nightly, ignore. If these keywords are not defined in the single test configuration, then they will be taken from the main configuration (if present). However, targets are joined by a special rule.

If targets with the key value is defined in the single test configuration, the targets from the main configuration is not considered. If the targets are used with exclude or include keyword, the test takes the value from the main config (or the default targets if there is no main config) and applies include/exclude operation on that set.

In this example, we specify that this test campaign is allowed to run only on three targets: ia32-generic-qemu, armv7a9-zynq7000-qemu, and host-generic-pc. We also set the nightly option to true, which means the tests will only be run if the --nightly flag is passed as an argument to the runner.

Now let’s go through the tests and try to understand the final configuration:

  • The arg_zero test specifies that the test-hello-arg executable should be executed without any arguments (execute: test-hello-arg). We provide the hello_arg_harness.py as the harness. In the kwargs section, we set argc to 0. This dictionary is passed later to the harness as kwargs parameter. Additionally, we exclude the armv7a9-zynq7000-qemu target for this specific test. As a result, it will be run on the ia32-generic-qemu and host-generic-pc targets.

  • The arg_two test specifies that the test-hello-arg should be executed with two arguments: arg1 and arg2 (execute: test-hello-arg arg1 arg2). We provide the hello_arg_harness.py as the harness. In the kwargs section, we set argc to 2. Additionally, we specify that this test should only run on the ia32-generic-qemu target.

  • The arg_hello test specifies that the test-hello-arg executable should be executed with the argument world. We provide the hello_arg_harness.py as the harness. In the kwargs section, we set input to Adios!. This word will be used as the input to the test-hello-arg. We also set nightly to false for this specific test. Thanks to that, the test can be launched always, no matter if --nightly is provided.

  • The last one, arg_hello_too_long_input, is similar to the previous one, but we set input to String that is too long!.

These test cases demonstrate different scenarios of executing the teset-hello-arg program with different arguments and inputs and validate the expected behavior using hello_arg_harness.py harness.

Now, let’s check how the runner will handle this test configuration based on the arguments provided by the user. Let’s start with following command:

./phoenix-rtos-tests/runner.py -T ia32-generic-qemu -t phoenix-rtos-tests/hello/test_hello_arg.yaml

The result should be as follows:

Flashing an image to device...
Done!
phoenix-rtos-tests/hello/arg_two: OK
TESTS: 1 PASSED: 1 FAILED: 0 SKIPPED: 0

The result is expected because there is only one test that does not run in nightly mode and is allowed to run on the ia32-generic-qemu target.

Add the --nightly flag to the previous command and compare the output:

./phoenix-rtos-tests/runner.py -T ia32-generic-qemu -t phoenix-rtos-tests/hello/test_hello_arg.yaml --nightly
Flashing an image to device...
Done!
phoenix-rtos-tests/hello/arg_zero: OK
phoenix-rtos-tests/hello/arg_two: OK
phoenix-rtos-tests/hello/arg_hello: OK
phoenix-rtos-tests/hello/arg_hello_too_long_input: OK
TESTS: 4 PASSED: 4 FAILED: 0 SKIPPED: 0

As you can see, all tests were run, and additional, it took more time to execute these tests. In nightly mode, the runner reboots the device after each test.

We can change the target to verify what targets keyword looks like. Change ia32-generic-qemu to armv7a9-zynq7000-qemu in the previous command:

./phoenix-rtos-tests/runner.py -T armv7a9-zynq7000-qemu -t phoenix-rtos-tests/hello/test_hello_arg.yaml --nightly
Flashing an image to device...
Done!
phoenix-rtos-tests/hello/arg_hello: OK
phoenix-rtos-tests/hello/arg_hello_too_long_input: OK
TESTS: 2 PASSED: 2 FAILED: 0 SKIPPED: 0

As we could expect, the arg_zero wasn’t launched because the armv7a9-zynq7000-qemu target is excluded in this test case. The arg_two test is also omitted because it can only be run on the ia32-generic-qemu target.

If you try to run this test case on a different target not listed in this YAML configuration, the test runner will finish its execution without executing any test:

./phoenix-rtos-tests/runner.py -T riscv64-generic-qemu -t phoenix-rtos-tests/hello/test_hello_arg.yaml --nightly
Flashing an image to device...
Done!
TESTS: 0 PASSED: 0 FAILED: 0 SKIPPED: 0

See also#

  1. Table of Contents