Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP
A record-then-assert test double library.
Python
branch: master

Merge pull request #21 from chromy/add-wrapped-to-patch

set __wrapped__ to work with pytest fixtures
Failed to load latest commit information.
examples back to old style classes
tests make patch set wrapped
.gitignore move hgignore to gitignore
CHANGELOG
README.rst Tiny README typofix
README.txt add README to setup.py data_files
dingus.py make patch set wrapped
nosy.py back to old style classes
runtests.py Moved runtests.py to root of repo
setup.py Don't install README.txt as a data file.

README.rst

DINGUSES

A dingus is sort of like a mock object. The main difference is that you don't set up expectations ahead of time. You just run your code, using a dingus in place of another object or class, and it will record what happens to it. Then, once your code has been exercised, you can make assertions about what it did to the dingus.

A new dingus is created from the Dingus class. You can give dinguses names, which helps with debugging your tests, especially when there are multiple dinguses in play.

>>> from dingus import Dingus
>>> d = Dingus('root')
>>> d
<Dingus root>

Accessing any attribute of a dingus will return a new dingus.

>>> d.something
<Dingus root.something>

There are a few exceptions for special dingus methods. We'll see some in a bit.

A dingus can also be called like a function or method. It doesn't care how many arguments you give it or what those arguments are. Calls to a dingus will always return the same object, regardless of the arguments.

>>> d()
<Dingus root()>
>>> d('argument')
<Dingus root()>
>>> d(55)
<Dingus root()>

RECORDING AND ASSERTIONS

At any time we can get a list of calls that have been made to a dingus. Each entry in the call list contains:

  • the name of the method called (or "()" if the dingus itself was called)
  • The arguments, or () if none
  • The keyword arguments, or {} if none
  • The value that was returned to the caller

Here is a list of the calls we've made to d so far:

>>> from pprint import pprint
>>> pprint(d.calls)
[('()', (), {}, <Dingus root()>),
 ('()', ('argument',), {}, <Dingus root()>),
 ('()', (55,), {}, <Dingus root()>)]

You can filter calls by name, arguments, and keyword arguments:

>>> pprint(d.calls('()', 55))
[('()', (55,), {}, <Dingus root()>)]

If you don't care about a particular argument's value, you can use the value DontCare when filtering:

>>> from dingus import DontCare
>>> pprint(d.calls('()', DontCare))
[('()', ('argument',), {}, <Dingus root()>),
 ('()', (55,), {}, <Dingus root()>)]

Dinguses can do more than just have attributes accessed and be called. They support many Python operators. The goal is to allow, and record, any interaction:

>>> d = Dingus('root')
>>> (2 ** d.something)['hello']() / 100 * 'foo'
<Dingus root.something.__rpow__[hello]().__div__.__mul__>

(Hopefully your real-world dingus recordings won't look like this!)

PATCHING

Dingus provides a context manager for patching objects during tests. For example:

>>> from dingus import patch
>>> import urllib2
>>> with patch('urllib2.urlopen'):
...     print urllib2.urlopen.__class__
<class 'dingus.Dingus'>
>>> print urllib2.urlopen.__class__
<type 'function'>

You can also use this as a decorator on your test methods:

>>> @patch('urllib2.urlopen')
... def test_something(self):
...     pass
...

ISOLATION

The opposite of patch is isolate. It patches everything except the named object:

>>> from dingus import isolate
>>> @isolate('urllib2.urlparse')
... def test_urlparse(self):
...     pass
...

When this test runs, everything in the urllib2 module except urlparse will be a dingus. Note that this may be slow to execute if the module contains many objects; performance patches are welcome. :)

DANGEROUS MAGIC

Dingus can also automatically replace a module's globals when running tests. This allows you to write fully isolated unit tests. See examples/urllib2/test_urllib2.py for an example. The author no longer recommends this feature, as it can encourage very brittle tests. You should feel the pain of manually mocking dependencies; the pain will tell you when a class collaborates with too many others.

Something went wrong with that request. Please try again.