In addition to the YAML-based tests, OpenERP uses the unittest2 testing framework to test both the core openerp package and its addons. For the core and each addons, tests are divided between three (overlapping) sets:
A test suite that comprises all the tests that can be run right after the addons is installed (or, for the core, right after a database is created). That suite is called fast_suite and must contain only tests that can be run frequently. Actually most of the tests should be considered fast enough to be included in that fast_suite list and only tests that take a long time to run (e.g. more than a minute) should not be listed. Those long tests should come up pretty rarely.
A test suite called checks provides sanity checks. These tests are invariants that must be full-filled at any time. They are expected to always pass: obviously they must pass right after the module is installed (i.e. just like the fast_suite tests), but they must also pass after any other module is installed, after a migration, or even after the database was put in production for a few months.
The third suite is made of all the tests: those provided by the two above suites, but also tests that are not explicitely listed in fast_suite or checks. They are not explicitely listed anywhere and are discovered automatically.
As the sanity checks provide stronger guarantees about the code and database structure, new tests must be added to the checks suite whenever it is possible. Said with other words: one should try to avoid writing tests that assume a freshly installed/unaltered module or database.
It is possible to have tests that are not listed in fast_suite or checks. This is useful if a test takes a lot of time. By default, when using the testing infrastructure, tests should run fast enough so that people can use them frequently. One can also use that possiblity for tests that require some complex setup before they can be successfuly run.
As a rule of thumb when writing a new test, try to add it to the checks suite. If it really needs that the module it belongs to is freshly installed, add it to fast_suite. Finally, if it can not be run in an acceptable time frame, don't add it to any explicit list.
The tests must be developed under <addons-name>.tests (or openerp.tests for the core). For instance, with respect to the tests, a module foo should be organized as follow:
foo/ __init__.py # does not import .tests tests/ __init__.py # import some of the tests sub-modules, and # list them in fast_suite or checks test_bar.py # contains unittest2 classes test_baz.py # idem ... and so on ...
The two explicit lists of tests are thus the variables foo.tests.fast_suite and foo.tests.checks. As an example, you can take a look at the openerp.tests module (which follows exactly the same conventions even if it is not an addons).
Note that the fast_suite and checks variables are really lists of module objects. They could be directly unittest2 suite objects if necessary in the future.
Running the tests¶
To run the tests (see above to learn how tests are organized), the simplest way is to use the oe command (provided by the openerp-command project).
> oe run-tests # will run all the fast_suite tests > oe run-tests -m openerp # will run all the fast_suite tests defined in `openerp.tests` > oe run-tests -m sale # will run all the fast_suite tests defined in `openerp.addons.sale.tests` > oe run-tests -m foo.test_bar # will run the tests defined in `openerp.addons.foo.tests.test_bar`
In addition to the above possibilities, when invoked with a non-existing module (or module.sub-module) name, oe will reply with a list of available test sub-modules.
Depending on the unittest2 class that is used to write the tests (see openerp.tests.common for some helper classes that you can re-use), a database may be created before the test is run, and the module providing the test will be installed on that database.
Because creating a database, installing modules, and then dropping it is expensive, it is possible to interleave the run of the fast_suite tests with the initialization of a new database: the dabase is created, and after each requested module is installed, its fast_suite tests are run. The database is thus created and dropped (and the modules installed) only once.
The setUp and tearDown methods are not part of the tests. Uncaught exceptions in those methods are errors, not test failures. In particular, a failing setUp will not be followed by a tearDown causing any allocated resource in the setUp to not be released by the tearDown.
In the openerp.tests.common.TransactionCase and openerp.tests.common.SingleTransactionCase, this means the test suite can hang because of unclosed cursors.