src/third_party/llvm-project/lldb/docs/testsuite/a-detailed-walkthrough.txt

Let's pick test/settings/TestSettings.py as our example.  First, notice the file
name "TestSettings.py", the Test*.py pattern is the default mechanism that the
test driver uses for discovery of tests.  As to TestSettings.py, it defines a
class:

class SettingsCommandTestCase(TestBase):

derived from TestBase, which is defined in test/lldbtest.py and is itself
derived from Python's unittest framework's TestCase class.  See also
http://docs.python.org/library/unittest.html for more details.

To just run the TestSettings.py test, chdir to the lldb test directory, and then
type the following command:

/Volumes/data/lldb/svn/trunk/test $ ./dotest.py settings
----------------------------------------------------------------------
Collected 6 tests

----------------------------------------------------------------------
Ran 6 tests in 8.699s

OK (expected failures=1)
/Volumes/data/lldb/svn/trunk/test $ 

Pass '-v' option to the test driver to also output verbose descriptions of the
individual test cases and their test status:

/Volumes/data/lldb/svn/trunk/test $ ./dotest.py -v settings
----------------------------------------------------------------------
Collected 6 tests

test_set_auto_confirm (TestSettings.SettingsCommandTestCase)
Test that after 'set auto-confirm true', manual confirmation should not kick in. ... ok
test_set_output_path (TestSettings.SettingsCommandTestCase)
Test that setting target.process.output-path for the launched process works. ... expected failure
test_set_prompt (TestSettings.SettingsCommandTestCase)
Test that 'set prompt' actually changes the prompt. ... ok
test_set_term_width (TestSettings.SettingsCommandTestCase)
Test that 'set term-width' actually changes the term-width. ... ok
test_with_dsym (TestSettings.SettingsCommandTestCase)
Test that run-args and env-vars are passed to the launched process. ... ok
test_with_dwarf (TestSettings.SettingsCommandTestCase)
Test that run-args and env-vars are passed to the launched process. ... ok

----------------------------------------------------------------------
Ran 6 tests in 5.735s

OK (expected failures=1)
/Volumes/data/lldb/svn/trunk/test $ 

Underneath, the '-v' option passes keyword argument verbosity=2 to the
Python's unittest.TextTestRunner (see also
http://docs.python.org/library/unittest.html#unittest.TextTestRunner).  For very
detailed descriptions about what's going on during the test, pass '-t' to the
test driver, which asks the test driver to trace the commands executed and to
display their output.  For brevity, the '-t' output is not included here.

Notice the 'expected failures=1' message at the end of the run.  This is because
of a bug currently in lldb such that setting target.process.output-path to
'stdout.txt' does not have any effect on the redirection of the standard output
of the subsequent launched process.  We are using unittest2 (a backport of new
unittest features for Python 2.4-2.6) to decorate (mark) the particular test
method as such:

    @unittest2.expectedFailure
    # rdar://problem/8435794
    # settings set target.process.output-path does not seem to work
    def test_set_output_path(self):

See http://pypi.python.org/pypi/unittest2 for more details.

Now let's look inside the test method:

    def test_set_output_path(self):
        """Test that setting target.process.output-path for the launched process works."""
        self.buildDefault()

        exe = os.path.join(os.getcwd(), "a.out")
        self.runCmd("file " + exe, CURRENT_EXECUTABLE_SET)

        # Set the output-path and verify it is set.
        self.runCmd("settings set target.process.output-path 'stdout.txt'")
        self.expect("settings show target.process.output-path",
            startstr = "target.process.output-path (string) = 'stdout.txt'")

        self.runCmd("run", RUN_SUCCEEDED)

        # The 'stdout.txt' file should now exist.
        self.assertTrue(os.path.isfile("stdout.txt"),
                        "'stdout.txt' exists due to target.process.output-path.")

        # Read the output file produced by running the program.
        with open('stdout.txt', 'r') as f:
            output = f.read()

        self.expect(output, exe=False,
            startstr = "This message should go to standard out.")

The self.buildDefault() statement is used to build a default binary for this
test instance.  For this particular test case, since we don't really care what
debugging format is used, we instruct the build subsystem to build the default
binary for us.  The base class TestBase has defined three instance methods:

    def buildDefault(self, architecture=None, compiler=None, dictionary=None):
        """Platform specific way to build the default binaries."""
        module = __import__(sys.platform)
        if not module.buildDefault(self, architecture, compiler, dictionary):
            raise Exception("Don't know how to build default binary")

    def buildDsym(self, architecture=None, compiler=None, dictionary=None):
        """Platform specific way to build binaries with dsym info."""
        module = __import__(sys.platform)
        if not module.buildDsym(self, architecture, compiler, dictionary):
            raise Exception("Don't know how to build binary with dsym")

    def buildDwarf(self, architecture=None, compiler=None, dictionary=None):
        """Platform specific way to build binaries with dwarf maps."""
        module = __import__(sys.platform)
        if not module.buildDwarf(self, architecture, compiler, dictionary):
            raise Exception("Don't know how to build binary with dwarf")

And the test/plugins/darwin.py provides the implementation for all three build
methods using the makefile mechanism.  We envision that linux plugin can use a
similar approach to accomplish the task of building the binaries.

Mac OS X provides an additional way to manipulate archived DWARF debug symbol
files and produces dSYM files.  The buildDsym() instance method is used by the
test method to build the binary with dsym info.  For an example of this,
see test/array_types/TestArrayTypes.py:

    @unittest2.skipUnless(sys.platform.startswith("darwin"), "requires Darwin")
    def test_with_dsym_and_run_command(self):
        """Test 'frame variable var_name' on some variables with array types."""
        self.buildDsym()
        self.array_types()

This method is decorated with a skipUnless decorator so that it will only gets
included into the test suite if the platform it is running on is 'darwin', a.k.a.
macOS.

Type 'man dsymutil' for more details. 

After the binary is built, it is time to specify the file to be used as the main
executable by lldb:

        # Construct the path to a file "a.out" inside the test's build folder.
        exe = self.getBuildArtifact("a.out")
        self.runCmd("file " + exe, CURRENT_EXECUTABLE_SET)

This is where the attribute assignment:

class SettingsCommandTestCase(TestBase):

    mydir = "settings"

which happens right after the SettingsCommandTestCase class declaration comes
into place. It specifies the relative directory to the top level 'test' so that
the test harness can change its working directory in order to find the
executable as well as the source code files. The runCmd() method is defined in
the TestBase base class (within test/lldbtest.py) and its purpose is to pass the
specified command to the lldb command interpreter. It's like you're typing the
command within an interactive lldb session.

The CURRENT_EXECUTABLE_SET is an assert message defined in the lldbtest module
so that it can be reused from other test modules.

By default, the runCmd() is going to check the return status of the command
execution and fails the test if it is not a success.  The assert message, in our
case CURRENT_EXECUTABLE_SET, is used in the exception printout if this happens.

There are cases when we don't care about the return status from the command
execution.  This can be accomplished by passing the keyword argument pair
'check=False' to the method.

After the current executable is set, we'll then execute two more commands:

        # Set the output-path and verify it is set.
        stdout = self.getBuildArtifact('stdout.txt')
        self.runCmd("settings set target.process.output-path '%s'" %stdout)
        self.expect("settings show target.process.output-path",
                    SETTING_MSG("target.process.output-path"),
            startstr = "target.process.output-path (string) = '.*stdout.txt'")

The first uses the 'settings set' command to set the static setting
target.process.output-path to be 'stdout.txt', instead of the default
'/dev/stdout'.  We then immediately issue a 'settings show' command to check
that, indeed, the setting did take place.  Notice that we use a new method
expect() to accomplish the task, which in effect issues a runCmd() behind the
door and grabs the output from the command execution and expects to match the
start string of the output against what we pass in as the value of the keyword
argument pair:

            startstr = "target.process.output-path (string) = '%s'" %stdout

Take a look at TestBase.expect() within lldbtest.py for more details.  Among
other things, it can also match against a list of regexp patterns as well as a
list of sub strings.  And it can also perform negative matching, i.e., instead
of expecting something from the output of command execution, it can perform the
action of 'not expecting' something.

This will launch/run the program:

        self.runCmd("run", RUN_SUCCEEDED)

And this asserts that the file 'stdout.txt' should be present after running the
program.

        # The 'stdout.txt' file should now exist.
        self.assertTrue(os.path.isfile(stdout),
                        "stdout.txt' exists due to target.process.output-path.")

Also take a look at main.cpp which emits some message to the stdout.  Now, if we
pass this assertion, it's time to examine the contents of the file to make sure
it contains the same message as programmed in main.cpp:

        # Read the output file produced by running the program.
        with open(stdout, 'r') as f:
            output = f.read()

        self.expect(output, exe=False,
            startstr = "This message should go to standard out.")

We open the file and read its contents into output, then issue an expect()
method.  The 'exe=False' keyword argument pair tells expect() that don't try to
execute the first arg as a command at all.  Instead, treat it as a string to
match against whatever is thrown in as keyword argument pairs!

There are also other test methods present in the TestSettings.py mode:
test_set_prompt(), test_set_term_width(), test_set_auto_confirm(),
test_with_dsym(), and test_with_dwarf().  We are using the default test loader
from unittest framework, which uses the 'test' method name prefix to identify
test methods automatically.

This finishes the walkthrough of the test method test_set_output_path(self).
Before we say goodbye, notice the little method definition at the top of the
file:

    @classmethod
    def classCleanup(cls):
        system(["/bin/sh", "-c", "rm -f "+self.getBuildArtifact("output.txt")])
        system(["/bin/sh", "-c", "rm -f "+self.getBuildArtifact("stdout.txt")])

This is a classmethod (as shown by the @classmethod decorator) which allows the
individual test class to perform cleanup actions after the test harness finishes
with the particular test class.  This is part of the so-called test fixture in
the unittest framework.  From http://docs.python.org/library/unittest.html:

A test fixture represents the preparation needed to perform one or more tests,
and any associate cleanup actions. This may involve, for example, creating
temporary or proxy databases, directories, or starting a server process.

The TestBase class uses such fixture with setUp(self), tearDown(self),
setUpClass(cls), and tearDownClass(cls).  And within teraDownClass(cls), it
checks whether the current class has an attribute named 'classCleanup', and
executes as a method if present.  In this particular case, the classCleanup()
calls a utility function system() defined in lldbtest.py in order to remove the
files created by running the program as the tests are executed.

This system() function uses the Python subprocess module to spawn the process
and to retrieve its results.  If the test instance passes the keyword argument
pair 'sender=self', the detailed command execution through the operating system
also gets recorded in a session object.  If the test instance fails or errors,
the session info automatically gets dumped to a file grouped under a directory
named after the timestamp of the particular test suite run.

For simple cases, look for the timestamp directory in the same directory of the
test driver program dotest.py.  For example, if we comment out the
@expectedFailure decorator for TestSettings.py, and then run the test module:

/Volumes/data/lldb/svn/trunk/test $ ./dotest.py -v settings
----------------------------------------------------------------------
Collected 6 tests

test_set_auto_confirm (TestSettings.SettingsCommandTestCase)
Test that after 'set auto-confirm true', manual confirmation should not kick in. ... ok
test_set_output_path (TestSettings.SettingsCommandTestCase)
Test that setting target.process.output-path for the launched process works. ... FAIL
test_set_prompt (TestSettings.SettingsCommandTestCase)
Test that 'set prompt' actually changes the prompt. ... ok
test_set_term_width (TestSettings.SettingsCommandTestCase)
Test that 'set term-width' actually changes the term-width. ... ok
test_with_dsym (TestSettings.SettingsCommandTestCase)
Test that run-args and env-vars are passed to the launched process. ... ok
test_with_dwarf (TestSettings.SettingsCommandTestCase)
Test that run-args and env-vars are passed to the launched process. ... ok

======================================================================
FAIL: test_set_output_path (TestSettings.SettingsCommandTestCase)
Test that setting target.process.output-path for the launched process works.
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Volumes/data/lldb/svn/trunk/test/settings/TestSettings.py", line 125, in test_set_output_path
    "'stdout.txt' exists due to target.process.output-path.")
AssertionError: False is not True : 'stdout.txt' exists due to target.process.output-path.

----------------------------------------------------------------------
Ran 6 tests in 8.219s

FAILED (failures=1)
/Volumes/data/lldb/svn/trunk/test $ ls 2010-10-19-14:10:49.059609

NOTE: This directory name has been changed to not contain the ':' character
      which is not allowed in windows platforms.  We'll change the ':' to '_'
      and get rid of the microsecond resolution by modifying the test driver.

TestSettings.SettingsCommandTestCase.test_set_output_path.log
/Volumes/data/lldb/svn/trunk/test $ 

We get one failure and a timestamp directory 2010-10-19-14:10:49.059609.
For education purposes, the directory and its contents are reproduced here in
the same directory as the current file.