Abjad includes an extensive battery of tests. Abjad is in a state of rapid development and extension. Major refactoring efforts are common every six to eight months, and are likely to remain so at least 2012. And yet Abjad continues to allow the creation of complex pieces of fully notated score in the midst of these changes. We believe this is due to the extensive coverage provided by the automated regression battery described in the following sections. [1]
A battery is any collection of tests. Regression tests differ from other types of test in that they are designed to be run again and again during many different stages of the development process. Regression tests help ensure that the system continues to function correctly as we make changes to it. An automated regression battery is one that can be run automatically by some sort of driver with minimal manual intervention.
Several different test drivers are now in use in the Python community. Of these, Abjad uses py.test. The py.test distribution is not included in the Python standard library, so one of the first thing new contributors to Abjad should do is download and install py.test, and then run the existing battery.
Change to the directory where you have Abjad installed. Then run py.test.
abjad$ py.test
============================= test process starts ==============================
executable: /Library/Frameworks/Python.framework/Versions/2.5/Resources/Python
.app/Contents/MacOS/Python (2.5.0-final-0)
using py lib: /Library/Frameworks/Python.framework/Versions/2.5/lib/python2.5/sit
e-packages/py <rev unknown>
accidental/test/test_accidental_compare.py[3] ...
accidental/test/test_accidental_eq.py[3] ...
accidental/test/test_accidental_init.py[2] ..
accidental/test/test_accidental_interface_grob_handling.py[2] ..
accidental/test/test_accidental_interface_style.py[2] ..
... (many lines omitted) ...
tuplet/test/test_tuplet_number_grob_handling.py[3] ...
update/test/test_update_interface.py[10] ..........
voice/interface/test/test_voice_interface_explicit.py[2] ..
voice/interface/test/test_voice_interface_number.py[5] .....
voice/test/test_voice_len.py[2] ..
================ tests finished: 2165 passed in 232.53 seconds =================
py.test crawls the entire directory structure from which you call it, running tests in alphabetical order. py.test prints the total number of tests per file in square brackets and prints test results as a single . dot for success or else an F for failure.
Project check-in standards ask that tests accompany all code committed to the Abjad repository. If you add a new function, class or method to Abjad, you should add a new test file for that function, class or method. If you fix or extend an existing function, class or method, you should find the existing test file that covers that code and then either add a completely new test to the test file or else update an existing test already present in the test file.
When py.test first starts up it crawls the entire directory structure from which you call it prior to running a single test. As py.test executes this preflight work, it looks for any files beginning or ending with the string test and then collects and alphabetizes these. Only after making such a catalog of tests does py.test begin execution. This collect-and-cache behavior leads to the important point about naming, below.
Note
The names of test functions must be absolutely unique across the entire directory structure on which you call py.test. You must never share names between test functions. For example, you must not have two tests named test_grob_handling_01( ) even if both tests live in different test files. That is, a test named test_grob_handling_01( ) living in the file test_accidental_grob_handling.py and a second test named test_grob_handling_01( ) living in the file test_notehead_grob_handling.py will conflict with the each other when py.test runs. And, unfortunately, ``py.test is silent about such conflicts when it runs. That is, should you run py.test with the duplicate naming situation described here, what will happen is that py.test will correctly run and report results for the first such test it finds. However, when py.test encounters the second like-named test, py.test will incorrectly report cached results for the first test rather than the second. The take-away is to include some sort of namespacing indicators in every test name and not to be afraid of long test names. The test_grob_handling_01( ) example given here fixes easily when the two tests rename to test_accidental_grob_handling_01( ) and test_notehead_grob_handling_01( ).
Footnotes
[1] | Abjad r2371 includes 2165 tests. |