diff options
author | Chris St. Pierre <chris.a.st.pierre@gmail.com> | 2012-10-05 11:57:16 -0400 |
---|---|---|
committer | Chris St. Pierre <chris.a.st.pierre@gmail.com> | 2012-10-05 11:57:16 -0400 |
commit | 1a3ced3f45423d79e08ca7d861e8118e8618d3b2 (patch) | |
tree | 72dcb775987fff65471b6e27bc2c6313d8fcb409 /doc | |
parent | cc2e101cd6a1ef8c29ef2481c03274011f321a77 (diff) | |
download | bcfg2-1a3ced3f45423d79e08ca7d861e8118e8618d3b2.tar.gz bcfg2-1a3ced3f45423d79e08ca7d861e8118e8618d3b2.tar.bz2 bcfg2-1a3ced3f45423d79e08ca7d861e8118e8618d3b2.zip |
wrote more detailed unit testing documentation
Diffstat (limited to 'doc')
-rw-r--r-- | doc/conf.py | 4 | ||||
-rw-r--r-- | doc/development/testing.txt | 31 | ||||
-rw-r--r-- | doc/development/unit-testing.txt | 354 |
3 files changed, 372 insertions, 17 deletions
diff --git a/doc/conf.py b/doc/conf.py index 538cd236d..4dda8327f 100644 --- a/doc/conf.py +++ b/doc/conf.py @@ -20,6 +20,7 @@ import time # add these directories to sys.path here. If the directory is relative to the # documentation root, use os.path.abspath to make it absolute, like shown here. sys.path.insert(0, os.path.abspath('../src/lib')) +sys.path.insert(0, os.path.abspath('..')) # -- General configuration ----------------------------------------------------- @@ -263,7 +264,8 @@ def setup(app): versions = ["3.2", "2.7", "2.6"] cur_version = '.'.join(str(v) for v in sys.version_info[0:2]) -intersphinx_mapping = dict() +intersphinx_mapping = dict(mock=('http://www.voidspace.org.uk/python/mock', + None)) for pyver in versions: if pyver == cur_version: key = 'py' diff --git a/doc/development/testing.txt b/doc/development/testing.txt index 1f0842053..f00193574 100644 --- a/doc/development/testing.txt +++ b/doc/development/testing.txt @@ -8,22 +8,24 @@ Testing Testing Prereleases ------------------- -Before each release, several prereleases will be tagged. It is -helpful to have users test these releases (when feasible) because -it is hard to replicate the full range of potential reconfiguration -situations; between different operating systems, system management -tools, and configuration specification variation, there can be -large differences between sites. +Before each release, several prereleases will be tagged. It is helpful +to have users test these releases (when feasible) because it is hard +to replicate the full range of potential reconfiguration situations; +between different operating systems, system management tools, and +configuration specification variation, there can be large differences +between sites. -For more details please visit `Tracking Development Releases of Bcfg2 <http://trac.mcs.anl.gov/projects/bcfg2/wiki/TrackingDevelopmentTrunk>`_ . +For more details please visit `Tracking Development Releases of Bcfg2 +<http://trac.mcs.anl.gov/projects/bcfg2/wiki/TrackingDevelopmentTrunk>`_ +. Upgrade Testing --------------- -This section describes upgrade procedures to completely test the -client and server. These procedures can be used for either pre-release +This section describes upgrade procedures to completely test the +client and server. These procedures can be used for either pre-release testing, or for confidence building in a new release. @@ -33,9 +35,9 @@ Server Testing 1. Ensure that the server produces the same configurations for clients * Before the upgrade, generate all client configurations using the - buildall subcommand of bcfg2-info. This subcommand takes a directory - argument; it will generate one client configuration in each file, - naming each according to the client name. + buildall subcommand of bcfg2-info. This subcommand takes a + directory argument; it will generate one client configuration in + each file, naming each according to the client name. .. code-block:: sh @@ -54,7 +56,8 @@ Server Testing * Upgrade the server software * Generate all client configurations in a second location using the new software. Any tracebacks reflect bugs, and should be filed in - the ticketing system. Any new messages should be carefully examined. + the ticketing system. Any new messages should be carefully + examined. * Compare each file in the old directory to those in the new directory using ``bcfg2-admin compare -r /old/directory /new/directory`` @@ -78,5 +81,5 @@ Server Testing Client Testing ^^^^^^^^^^^^^^ -Run the client in dry-run and non-dry-run mode; ensure that multiple +Run the client in dry-run and non-dry-run mode; ensure that multiple runs produce consistent results. diff --git a/doc/development/unit-testing.txt b/doc/development/unit-testing.txt index 7af969686..a4ec2ad5c 100644 --- a/doc/development/unit-testing.txt +++ b/doc/development/unit-testing.txt @@ -6,12 +6,14 @@ Bcfg2 unit testing ================== -.. _Python Mock Module: http://python-mock.sourceforge.net/ +.. _Python Mock Module: http://www.voidspace.org.uk/python/mock .. _Python Nose: http://readthedocs.org/docs/nose/en/latest/ You will first need to install the `Python Mock Module`_ and `Python Nose`_ modules. You can then run the existing tests with the -following.:: +following: + +.. code-block: bash cd testsuite nosetests @@ -26,3 +28,351 @@ You should see output something like the following:: Unit tests are also run by Travis-CI, a free continuous integration service, at http://travis-ci.org/#!/Bcfg2/bcfg2/ + +Testing in a virtualenv +======================= + +Travis-CI runs the unit tests in a virtual environment, so to emulate +that testing environment as closely as possible you can also use a +virtual environment. To do so, you must have `virtualenv +<http://www.virtualenv.org/>`_ installed. + +There are two ways to test: Either with just the bare essential +packages installed, or with optional packages installed as well. +(Optional packages are things like Genshi; you can run Bcfg2 with them +or without them.) For completeness, the tests should be run in both +manners. (On Python 3, almost none of the optional packages are +available, so it can only be run with just the required packages.) To +install the optional packages, set: + +.. code-block:: bash + + export WITH_OPTIONAL_DEPS=yes + +This flag tells the install script to install optional dependencies as +well as requirements. + +This assumes that you will create a virtual environment in +``~/venvs/``, and that the Bcfg2 source tree is cloned into +``~/bcfg2``. + +First, create a new virtual environment and activate it: + +.. code-block:: bash + + cd ~/venvs + virtualenv travis + source travis/bin/activate + +Get the test suite from bcfg2: + +.. code-block:: bash + + cp -R ~/bcfg2/* ~/venvs/travis/ + +Next, you must install prerequisite packages that are required to +build some of the required Python packages, and some optional packages +that are much easier to install from binary (rather than from source). +If you are running on Ubuntu (the platform Travis-CI runs on) and have +sudo, you can simply run: + +.. code-block:: bash + + testsuite/before_install.sh + +If not, you will need to examine ``testsuite/before_install.sh`` +and install the packages manually. The equivalent for Fedora, for +instance, would be: + +.. code-block:: bash + + yum -y update + yum -y install swig pylint + if [[ "$WITH_OPTIONAL_DEPS" == "yes" ]]; then + sudo yum -y install libselinux-python pylibacl python-inotify \ + PyYAML + fi + +Next, install required Python packages: + +.. code-block:: bash + + testsuite/install.sh + +Install Bcfg2 itself to the virtualenv: + +.. code-block:: bash + + pip install -e . + +Now you can run tests: + +.. code-block:: bash + + nosetests testsuite + +Writing Unit Tests +================== + +Bcfg2 makes extremely heavy use of object inheritance, which can make +it challenging at times to write reusable tests. For instance, when +writing tests for the base :class:`Bcfg2.Server.Plugin.base.Plugin` +class, which all Bcfg2 :ref:`server-plugins-index` inherit from via +the :mod:`Plugin interfaces <Bcfg2.Server.Plugin.interfaces>`, +yielding several levels of often-multiple inheritance. To make this +easier, our unit tests adhere to several design considerations: + +Inherit Tests +------------- + +Our test objects should have inheritance trees that mirror the +inheritance trees of their tested objects. For instance, the +:class:`Bcfg2.Server.Plugins.Metadata.Metadata` class definition is: + +.. code-block:: python + + class Metadata(Bcfg2.Server.Plugin.Metadata, + Bcfg2.Server.Plugin.Statistics, + Bcfg2.Server.Plugin.DatabaseBacked): + +Consequently, the ``TestMetadata`` class definition is: + +.. code-block:: python + + class TestMetadata(TestPlugin.TestMetadata, + TestPlugin.TestStatistics, + TestPlugin.TestDatabaseBacked): + +.. note:: + + The test object names are abbreviated because of the system of + relative imports in the ``testsuite`` tree, described below. + +This gives us a large number of tests basically "for free": all core +:class:`Bcfg2.Server.Plugin.interfaces.Metadata`, +:class:`Bcfg2.Server.Plugin.interfaces.Statistics`, and +:class:`Bcfg2.Server.Plugin.helpers.DatabaseBacked` functionality is +automatically tested on the ``Metadata`` class, which gives the test +writer a lot of free functionality and also an easy list of which +tests must be overridden to provide tests appropriate for the ``Metadata`` +class implementation. + +Additionally, a test class should have a class variable that describes +the class that is being tested, and tests in that class should use +that class variable to instantate the tested object. For instance, +the test for :class:`Bcfg2.Server.Plugin.helpers.DirectoryBacked` +looks like this: + +.. code-block:: python + + class TestDirectoryBacked(Bcfg2TestCase): + test_obj = DirectoryBacked + ... + + + def test_child_interface(self): + """ ensure that the child object has the correct interface """ + self.assertTrue(hasattr(self.test_obj.__child__, "HandleEvent")) + +Then test objects that inherit from ``TestDirectoryBacked`` can +override that object, and the ``test_child_interface`` test (e.g.) +will still work. For example: + +.. code-block:: python + + class TestPropDirectoryBacked(TestDirectoryBacked): + test_obj = PropDirectoryBacked + +Finally, each test class must also provide a ``get_obj`` method that +takes no required arguments and produces an instance of ``test_obj``. +All test methods must use ``self.get_obj()`` to instantiate an object +to be tested. + +An object that does not inherit from any other tested Bcfg2 objects +should inherit from :class:`testsuite.common.Bcfg2TestCase`, described +below. + +.. _development-unit-testing-relative-imports: + +Relative Imports +---------------- + +In order to reuse test code and allow for test inheritance, each test +module should add all parent module paths to its ``sys.path``. For +instance, assuming a test in +``testsuite/Testsrc/Testlib/TestServer/TestPlugins/TestMetadata.py``, +the following paths should be added to ``sys.path``:: + + testsuite + testsuite/Testsrc + testsuite/Testsrc/Testlib + testsuite/Testsrc/Testlib/TestServer + testsuite/Testsrc/Testlib/TestServer/TestPlugins + +This must be done because Python 2.4, one of our target platforms, +does not support relative imports. An easy way to do this is to add +the following snippet to the top of each test file: + +.. code-block:: python + + import os + import sys + + # add all parent testsuite directories to sys.path to allow (most) + # relative imports in python 2.4 + path = os.path.dirname(__file__) + while path != "/": + if os.path.basename(path).lower().startswith("test"): + sys.path.append(path) + if os.path.basename(path) == "testsuite": + break + path = os.path.dirname(path) + +In addition, each new directory created in ``testsuite`` must contain +an empty ``__init__.py``. + +This will allow you, within ``TestMetadata.py``, to import common test +code and the parent objects the ``TestMetadata`` class will inherit from: + +.. code-block:: python + + from common import inPy3k, call, builtins, u, can_skip, \ + skip, skipIf, skipUnless, Bcfg2TestCase, DBModelTestCase, syncdb, \ + patchIf, datastore + from TestPlugin import TestXMLFileBacked, TestMetadata as _TestMetadata, \ + TestStatistics, TestDatabaseBacked + +Avoid Patching Where Possible +----------------------------- + +The `Python Mock Module`_ provides a ``patch`` decorator that can be +used to replace tested objects with ``Mock`` objects. This is +wonderful and necessary, but due to differences in the way various +versions of Python and Python Mock handle object scope, it's not +always reliable when combined with our system of test object +inheritance. Consequently, you should follow these rules when +considering whether to use ``patch``: + +* If you need to mock an object that is not part of Bcfg2 (e.g., a + builtin or an object in another Python library), use ``patch``. +* If you need to patch an object being tested in order to instantiate + it, use ``patch``, but see below. +* If you need to patch a function (not a method) that is part of + Bcfg2, use ``patch``. +* If you need to mock an object that is part of the object being + tested, do not use ``patch``. + +As an example of the last rule, assume you are writing tests for +:class:`Bcfg2.Server.Plugin.helpers.FileBacked`. +:func:`Bcfg2.Server.Plugin.helpers.FileBacked.HandleEvent` calls +:func:`Bcfg2.Server.Plugin.helpers.FileBacked.Index`, so we need to +mock the ``Index`` function. This is the **wrong** way to do that: + +.. code-block:: python + + class TestFileBacked(Bcfg2TestCase): + @patch("%s.open" % builtins) + @patch("Bcfg2.Server.Plugin.helpers.FileBacked.Index") + def test_HandleEvent(self, mock_Index, mock_open): + ... + +Tests that inherit from ``TestFileBacked`` will not reliably patch the +correct ``Index`` function. Instead, assign the object to be mocked +directly: + +.. code-block:: python + + class TestFileBacked(Bcfg2TestCase): + @patch("%s.open" % builtins) + def test_HandleEvent(self, mock_open): + fb = self.get_obj() + fb.Index = Mock() + +.. note:: + + ``@patch`` decorations are evaluated at compile-time, so a + workaround like this does **not** work: + + .. code-block:: python + + class TestFileBacked(Bcfg2TestCase): + @patch("%s.open" % builtins) + @patch("%s.%s.Index" % (self.test_obj.__module__, + self.test_obj.__name)) + def test_HandleEvent(self, mock_Index, mock_open): + ... + + But see below about patching objects before instantiation. + +In some cases, you will need to patch an object in order to +instantiate it. For instance, consider +:class:`Bcfg2.Server.Plugin.helpers.DirectoryBacked`, which attempts +to set a file access monitor watch when it is instantiated. This +won't work during unit testing, so we have to patch +:func:`Bcfg2.Server.Plugin.helpers.DirectoryBacked.add_directory_monitor` +in order to successfully instantiate a ``DirectoryBacked`` object. In +order to do that, we need to patch the object being tested, which is a +variable, but we need to evaluate the patch at run-time, not at +compile time, in order to deal with inheritance. This can be done +with a ``@patch`` decorator on an inner function, e.g.: + +.. code-block:: python + + class TestDirectoryBacked(Bcfg2TestCase): + test_obj = DirectoryBacked + + def test__init(self): + @patch("%s.%s.add_directory_monitor" % (self.test_obj.__module__, + self.test_obj.__name__)) + def inner(mock_add_monitor): + db = self.test_obj(datastore, Mock()) + mock_add_monitor.assert_called_with('') + + inner() + +``inner()`` is patched when ``test__init()`` is called, and so +``@patch()`` is called with the module and the name of the object +being tested as defined by the test object (i.e., not as defined by +the parent object). If this is not done, then the patch will be +applied at compile-time and ``add_directory_monitor`` will be patched +on the ``DirectoryBacked`` class instead of on the class to be tested. + +Some of our older unit tests do not follow these rules religiously, so +as more tests are written that inherit from larger portions of the +``testsuite`` tree they may need to be refactored. + +Naming +------ + +In order to make the system of inheritance we implement possible, we +must follow these naming conventions fairly religiously. + +* Test classes are given the name of the object to be tested with + ``Test`` prepended. E.g., the test for the + :class:`Bcfg2.Server.Plugins.Metadata.Metadata` is named + ``TestMetadata``. +* Test classes that test miscellaneous functions in a module are named + ``TestFunctions``. +* Test modules are given the name of the module to be tested with + ``Test`` prepended. Tests for ``__init__.py`` are named + ``Test_init.py`` (one underscore). +* Tests for methods or functions are given the name of the method or + function to be tested with ``test_`` prepended. E.g., the test for + :class:`Bcfg2.Server.Plugin.helpers.StructFile.Match` is called + ``test_Match``; the test for + :class:`Bcfg2.Server.Plugin.helpers.StructFile._match` is called + ``test__match``. +* Tests for magic methods -- those that start and end with double + underscores -- are named ``test__<name>``, where name is the name of + the magic method without underscores. E.g., a test for ``__init__`` + is called ``test__init``, and a test for ``__getitem__`` is called + ``test__getitem``. If this causes a collision with a non-magic + function (e.g., if a class also has a function called + ``_getitem()``, the test for which would also be called + ``test__getitem``, seriously consider refactoring the code for the + class. + +Common Test Code +---------------- + +.. automodule:: testsuite.common |