Writing a test

A test is one C file in ktest/. The build makes one ELF for it, and kst runs it as a child process on the test image.

The file

Name the file ktest/test.<name>.c. The program returns 0 for success and uses the checks from ktest/ktest.h:

#include "ktest.h"

int main(int argc, char** argv)
{
    ktest_printf("hello from %s\n", argv[0]);
    ktest_assert(argc == 1);

    return 0;
}

• ktest_expect(expression, expected) compares the value and keeps the test running after a mismatch.
• ktest_assert(condition) stops the test with a message.
• ktest_result() returns 0 when every check held, and 1 otherwise. A test that uses ktest_expect returns it from main.

ktest.h also has file helpers, formatted output, thread helpers and raw system call wrappers (ktest_syscall1 to ktest_syscall6). The user ABI is in kernel/include/usr/vex/; a test includes it as <vex/...>.

Add it to the build

Add the name to ktest_programs in ktest/meson.build:

ktest_programs = [
    'hello',
    'uname',
    # ...
]

The order in that list is the order that kst all runs the tests. A test that takes the machine over, such as the framebuffer test, goes last.

The build copies each ELF to /kst/<name>.elf on the DATA partition, and kst finds it by name.

Run it

Build and boot the image:

ninja -C build ktest-run

Then run one test or all of them at the kst prompt:

> run hello
> all

kst prints [ OK ] or [FAIL] with a timestamp for each program. A failure names the exit status, the signal or the failed waitpid.

Rules

• A test is a separate process, forked by kst. It cannot see the memory of another test.
• A test runs with the full capability set, because kst is init. A test can mount, format and corrupt the DATA partition.
• The tests share the DATA partition. A test that writes leaves the files for the tests that follow.
• There is no unit test suite. Check a kernel change on the test image.
• After a run, check the ext2 partition with e2fsck. See the test image.

Where the code is