Dev: Testing

takluyver edited this page Dec 8, 2014 · 14 revisions
Clone this wiki locally

Testing IPython for users and developers

Overview

It is extremely important that all code contributed to IPython has tests. Tests should be written as unittests, doctests or other entities that the IPython test system can detect. See below for more details on this.

Each subpackage in IPython should have its own tests directory that contains all of the tests for that subpackage. All of the files in the tests directory should have the word "tests" in them to enable the testing framework to find them.

In docstrings, examples (either using IPython prompts like In [1]: or 'classic' python >>> ones) can and should be included. The testing system will detect them as doctests and will run them; it offers control to skip parts or all of a specific doctest if the example is meant to be informative but shows non-reproducible information (like filesystem data).

If a subpackage has any dependencies beyond the Python standard library, the tests for that subpackage should be skipped if the dependencies are not found. This is very important so users don't get tests failing simply because they don't have dependencies.

The testing system we use is an extension of the nose test runner. In particular we've developed a nose plugin that allows us to paste verbatim IPython sessions and test them as doctests, which is extremely important for us.

Running the test suite

You can run IPython from the source download directory without even installing it system-wide or having configure anything, by typing at the terminal:

   python2 -c "import IPython; IPython.start_ipython();"

To start the webbased notebook you can use:

   python2 -c "import IPython; IPython.start_ipython(['notebook']);"

In order to run the test suite, you must at least be able to import IPython, even if you haven't fully installed the user-facing scripts yet (common in a development environment). You can then run the tests with:

  python -c "import IPython; IPython.test()"

Once you have installed IPython either via a full install or using:

  python setup.py develop

you will have available a system-wide script called iptest that runs the full test suite. You can then run the suite with:

   iptest  [args]

By default, this excludes the relatively slow tests for IPython.parallel. To run these, use iptest --all.

Please note that the iptest tool will run tests against the code imported by the Python interpreter. If the command python setup.py symlink has been previously run then this will always be the test code in the local directory via a symlink. However, if this command has not been run for the version of Python being tested, there is the possibility that iptest will run the tests against an installed version of IPython.

Regardless of how you run things, you should eventually see something like:

   **********************************************************************
   Test suite completed for system with the following information:
   {'commit_hash': '144fdae',
    'commit_source': 'repository',
    'ipython_path': '/home/fperez/usr/lib/python2.6/site-packages/IPython',
    'ipython_version': '0.11.dev',
    'os_name': 'posix',
    'platform': 'Linux-2.6.35-22-generic-i686-with-Ubuntu-10.10-maverick',
    'sys_executable': '/usr/bin/python',
    'sys_platform': 'linux2',
    'sys_version': '2.6.6 (r266:84292, Sep 15 2010, 15:52:39) \n[GCC 4.4.5]'}

   Tools and libraries available at test time:
      curses matplotlib pymongo qt sqlite3 tornado wx wx.aui zmq

   Ran 9 test groups in 67.213s

   Status:
   OK

If not, there will be a message indicating which test group failed and how to rerun that group individually. For example, this tests the IPython.utils subpackage, the -v option shows progress indicators:

   $ iptest IPython.utils -- -v
   ..........................SS..SSS............................S.S...
   .........................................................
   ----------------------------------------------------------------------
   Ran 125 tests in 0.119s

   OK (SKIP=7)

Because the IPython test machinery is based on nose, you can use all nose syntax. Options after -- are passed to nose. For example, this lets you run the specific test test_rehashx inside the test_magic module:

   $ iptest IPython.core.tests.test_magic:test_rehashx -- -vv
   IPython.core.tests.test_magic.test_rehashx(True,) ... ok
   IPython.core.tests.test_magic.test_rehashx(True,) ... ok

   ----------------------------------------------------------------------
   Ran 2 tests in 0.100s

   OK

When developing, the --pdb and --pdb-failures of nose are particularly useful, these drop you into an interactive pdb session at the point of the error or failure respectively: iptest mymodule -- --pdb.

The system information summary printed above is accessible from the top level package. If you encounter a problem with IPython, it's useful to include this information when reporting on the mailing list; use::

    from IPython import sys_info
    print sys_info()

and include the resulting information in your query.

Testing pull requests

We have a script that fetches a pull request from Github, merges it with master, and runs the test suite on different versions of Python. This uses a separate copy of the repository, so you can keep working on the code while it runs. To run it:

    python tools/test_pr.py -p 1234

The number is the pull request number from Github; the -p flag makes it post the results to a comment on the pull request. Any further arguments are passed to iptest.

This requires the requests and keyring packages.

For developers: writing tests

By now IPython has a reasonable test suite, so the best way to see what's available is to look at the tests directory in most subpackages. But here are a few pointers to make the process easier.

Main tools: IPython.testing

The IPython.testing package is where all of the machinery to test IPython (rather than the tests for its various parts) lives. In particular, the iptest module in there has all the smarts to control the test process. In there, the make_exclude function is used to build a blacklist of exclusions, these are modules that do not get even imported for tests. This is important so that things that would fail to even import because of missing dependencies don't give errors to end users, as we stated above.

The decorators module contains a lot of useful decorators, especially useful to mark individual tests that should be skipped under certain conditions (rather than blacklisting the package altogether because of a missing major dependency).

Our nose plugin for doctests

The plugin subpackage in testing contains a nose plugin called ipdoctest that teaches nose about IPython syntax, so you can write doctests with IPython prompts. You can also mark doctest output with # random for the output corresponding to a single input to be ignored (stronger than using ellipsis and useful to keep it as an example). If you want the entire docstring to be executed but none of the output from any input to be checked, you can use the # all-random marker. The IPython.testing.plugin.dtexample module contains examples of how to use these; for reference here is how to use # random:

    def ranfunc():
    """A function with some random output.

       Normal examples are verified as usual:
       >>> 1+3
       4

       But if you put '# random' in the output, it is ignored:
       >>> 1+3
       junk goes here...  # random

       >>> 1+2
       again,  anything goes #random
       if multiline, the random mark is only needed once.

       >>> 1+2
       You can also put the random marker at the end:
       # random

       >>> 1+2
       # random
       .. or at the beginning.

       More correct input is properly verified:
       >>> ranfunc()
       'ranfunc'
    """
    return 'ranfunc'

and an example of # all-random:

    def random_all():
    """A function where we ignore the output of ALL examples.

    Examples:

      # all-random

      This mark tells the testing machinery that all subsequent examples
      should be treated as random (ignoring their output).  They are still
      executed, so if a they raise an error, it will be detected as such,
      but their output is completely ignored.

      >>> 1+3
      junk goes here...

      >>> 1+3
      klasdfj;

    In [8]: print 'hello'
    world  # random

    In [9]: iprand()
    Out[9]: 'iprand'
    """
    return 'iprand'

When writing docstrings, you can use the @skip_doctest decorator to indicate that a docstring should not be treated as a doctest at all. The difference between # all-random and @skip_doctest is that the former executes the example but ignores output, while the latter doesn't execute any code. @skip_doctest should be used for docstrings whose examples are purely informational.

If a given docstring fails under certain conditions but otherwise is a good doctest, you can use code like the following, that relies on the 'null' decorator to leave the docstring intact where it works as a test:

  # The docstring for full_path doctests differently on win32 (different path
  # separator) so just skip the doctest there, and use a null decorator
  # elsewhere:

  doctest_deco = dec.skip_doctest if sys.platform == 'win32' else dec.null_deco

  @doctest_deco
  def full_path(startPath,files):
      """Make full paths for all the listed files, based on startPath..."""

      # function body follows...

With our nose plugin that understands IPython syntax, an extremely effective way to write tests is to simply copy and paste an interactive session into a docstring. You can writing this type of test, where your docstring is meant only as a test, by prefixing the function name with doctest_ and leaving its body absolutely empty other than the docstring. In IPython.core.tests.test_magic you can find several examples of this, but for completeness sake, your code should look like this (a simple case):

    def doctest_time():
    """
    In [10]: %time None
    CPU times: user 0.00 s, sys: 0.00 s, total: 0.00 s
    Wall time: 0.00 s
    """

This function is only analyzed for its docstring but it is not considered a separate test, which is why its body should be empty.

JavaScript Tests

We currently use casperjs for testing the notebook javascript user interface.

To run the JS test suite by itself, you can either use iptest js, which will start up a new notebook server and test against it, or you can open up a notebook server yourself, and then:

cd IPython/html/tests/casperjs;
casperjs test --includes=util.js test_cases

If your testing notebook server uses something other than the default port (8888), you will have to pass that as a parameter to the test suite as well.

casperjs test --includes=util.js --port=8889 test_cases

Running individual tests

To speed up development, you usually are working on getting one test passing at a time. To do this, just pass the filename directly to the casperjs test command like so:

casperjs test --includes=util.js  test_cases/execute_code_cell.js

Wrapping your head around the javascript within javascript:

CasperJS is a browser that's written in javascript, so we write javascript code to drive it. The Casper browser itself also has a javascript implementation (like the ones that come with Firefox and Chrome), and in the test suite we get access to those using this.evaluate, and it's cousins (this.theEvaluate, etc). Additionally, because of the asynchronous / callback nature of everything, there are plenty of this.then calls which define steps in test suite. Part of the reason for this is that each step has a timeout (default of 5 or 10 seconds). Additionally, there are already convenience functions in util.js to help you wait for output in a given cell, etc. In our javascript tests, if you see functions which look_like_pep8_naming_convention, those are probably coming from util.js, whereas functions that come with casper haveCamelCaseNamingConvention

Each file in test_cases looks something like this (this is test_cases/check_interrupt.js):

casper.notebook_test(function () {
    this.evaluate(function () {
        var cell = IPython.notebook.get_cell(0);
        cell.set_text('import time\nfor x in range(3):\n    time.sleep(1)');
        cell.execute();
    });


    // interrupt using menu item (Kernel -> Interrupt)
    this.thenClick('li#int_kernel');

    this.wait_for_output(0);

    this.then(function () {
        var result = this.get_output_cell(0);
        this.test.assertEquals(result.ename, 'KeyboardInterrupt', 'keyboard interrupt (mouseclick)');
    });

    // run cell 0 again, now interrupting using keyboard shortcut
    this.thenEvaluate(function () {
        cell.clear_output();
        cell.execute();
    });

    // interrupt using Ctrl-M I keyboard shortcut
    this.thenEvaluate( function() {
        IPython.utils.press_ghetto(IPython.utils.keycodes.I)
    });

    this.wait_for_output(0);

    this.then(function () {
        var result = this.get_output_cell(0);
        this.test.assertEquals(result.ename, 'KeyboardInterrupt', 'keyboard interrupt (shortcut)');
    });
});

For an example of how to pass parameters to the client-side javascript from casper test suite, see the casper.wait_for_output implementation in IPython/html/tests/casperjs/util.js

Testing system design notes

This section is a set of notes on the key points of the IPython testing needs, that were used when writing the system and should be kept for reference as it eveolves.

Testing IPython in full requires modifications to the default behavior of nose and doctest, because the IPython prompt is not recognized to determine Python input, and because IPython admits user input that is not valid Python (things like %magics and !system commands.

We basically need to be able to test the following types of code:

  • (1) Pure Python files containing normal tests. These are not a problem, since Nose will pick them up as long as they conform to the (flexible) conventions used by nose to recognize tests.

  • (2) Python files containing doctests. Here, we have two possibilities:

    • The prompts are the usual >>> and the input is pure Python.
    • The prompts are of the form In [1]: and the input can contain extended IPython expressions.

    In the first case, Nose will recognize the doctests as long as it is called with the --with-doctest flag. But the second case will likely require modifications or the writing of a new doctest plugin for Nose that is IPython-aware.

  • (3) ReStructuredText files that contain code blocks. For this type of file, we have three distinct possibilities for the code blocks:

    • They use >>> prompts.
    • They use In [1]: prompts.
    • They are standalone blocks of pure Python code without any prompts.

    The first two cases are similar to the situation #2 above, except that in this case the doctests must be extracted from input code blocks using docutils instead of from the Python docstrings.

    In the third case, we must have a convention for distinguishing code blocks that are meant for execution from others that may be snippets of shell code or other examples not meant to be run. One possibility is to assume that all indented code blocks are meant for execution, but to have a special docutils directive for input that should not be executed.

    For those code blocks that we will execute, the convention used will simply be that they get called and are considered successful if they run to completion without raising errors. This is similar to what Nose does for standalone test functions, and by putting asserts or other forms of exception-raising statements it becomes possible to have literate examples that double as lightweight tests.

  • (4) Extension modules with doctests in function and method docstrings. Currently Nose simply can't find these docstrings correctly, because the underlying doctest DocTestFinder object fails there. Similarly to #2 above, the docstrings could have either pure python or IPython prompts.

Of these, only 3-c (reST with standalone code blocks) is not implemented at this point.