path: root/docs/comparing.txt

Comparing objects and sequences
===============================

.. currentmodule:: testfixtures

Python's :mod:`unittest` package often fails to give very useful
feedback when comparing long sequences or chunks of text. It also has
trouble dealing with objects that don't natively support
comparison. The functions and classes described here alleviate these
problems.

The compare function
--------------------

The :func:`compare` function can be used as a replacement for
:meth:`~unittest.TestCase.assertEqual`. It raises an
``AssertionError`` when its parameters are not equal, which will be
reported as a test failure:

>>> from testfixtures import compare
>>> compare(1, 2)
Traceback (most recent call last):
 ...
AssertionError: 1 != 2

However, it allows you to specify a prefix for the message to be used
in the event of failure:

>>> compare(1, 2, prefix='wrong number of orders')
Traceback (most recent call last):
 ...
AssertionError: wrong number of orders: 1 != 2

This is recommended as it makes the reason for the failure more
apparent without having to delve into the code or tests.

You can also optionally specify a suffix, which will be appended to the
message on a new line:

>>> compare(1, 2, suffix='(Except for very large values of 1)')
Traceback (most recent call last):
 ...
AssertionError: 1 != 2
(Except for very large values of 1)

The expected and actual value can also be explicitly supplied, making it
clearer as to what has gone wrong:

>>> compare(expected=1, actual=2)
Traceback (most recent call last):
 ...
AssertionError: 1 (expected) != 2 (actual)

The real strengths of this function come when comparing more complex
data types. A number of common python data types will give more
detailed output when a comparison fails as described below:

sets
~~~~
 
Comparing sets that aren't the same will attempt to
highlight where the differences lie:

>>> compare(set([1, 2]), set([2, 3]))
Traceback (most recent call last):
 ...
AssertionError: set not as expected:
<BLANKLINE>
in first but not second:
[1]
<BLANKLINE>
in second but not first:
[3]
<BLANKLINE>
<BLANKLINE>

dicts
~~~~~

Comparing dictionaries that aren't the same will attempt to
highlight where the differences lie:

>>> compare(dict(x=1, y=2, a=4), dict(x=1, z=3, a=5))
Traceback (most recent call last):
 ...
AssertionError: dict not as expected:
<BLANKLINE>
same:
['x']
<BLANKLINE>
in first but not second:
'y': 2
<BLANKLINE>
in second but not first:
'z': 3
<BLANKLINE>
values differ:
'a': 4 != 5

lists and tuples
~~~~~~~~~~~~~~~~

Comparing lists or tuples that aren't the same will attempt to highlight
where the differences lie:

>>> compare([1, 2, 3], [1, 2, 4])
Traceback (most recent call last):
 ...
AssertionError: sequence not as expected:
<BLANKLINE>
same:
[1, 2]
<BLANKLINE>
first:
[3]
<BLANKLINE>
second:
[4]

namedtuples
~~~~~~~~~~~

When two :func:`~collections.namedtuple` instances are compared, if
they are of the same type, the description given will highlight which
elements were the same and which were different:

>>> from collections import namedtuple
>>> TestTuple = namedtuple('TestTuple', 'x y z')
>>> compare(TestTuple(1, 2, 3), TestTuple(1, 4, 3))
Traceback (most recent call last):
 ...
AssertionError: TestTuple not as expected:
<BLANKLINE>
same:
['x', 'z']
<BLANKLINE>
values differ:
'y': 2 != 4

generators
~~~~~~~~~~

When two generators are compared, they are both first unwound into
tuples and those tuples are then compared.

The :ref:`generator <generator>` helper is useful for creating a
generator to represent the expected results:

>>> from testfixtures import generator
>>> def my_gen(t):
...     i = 0
...     while i<t:
...         i += 1
...         yield i
>>> compare(generator(1, 2, 3), my_gen(2))
Traceback (most recent call last):
 ...
AssertionError: sequence not as expected:
<BLANKLINE>
same:
(1, 2)
<BLANKLINE>
first:
(3,)
<BLANKLINE>
second:
()

.. warning::

  If you wish to assert that a function returns a generator, say, for
  performance reasons, then you should use 
  :ref:`strict comparison <strict-comparison>`.
  
strings and unicodes
~~~~~~~~~~~~~~~~~~~~

Comparison of strings can be tricky, particularly when those strings
contain multiple lines; spotting the differences between the expected
and actual values can be hard.

To help with this, long strings give a more helpful representation
when comparison fails:

>>> compare("1234567891011", "1234567789")
Traceback (most recent call last):
 ...
AssertionError: 
'1234567891011'
!=
'1234567789'

Likewise, multi-line strings give unified diffs when their comparison
fails:

>>> compare("""
...         This is line 1
...         This is line 2
...         This is line 3
...         """,
...         """
...         This is line 1
...         This is another line
...         This is line 3
...         """)
Traceback (most recent call last):
 ...
AssertionError: 
--- first
+++ second
@@ -1,5 +1,5 @@
<BLANKLINE>
         This is line 1
-        This is line 2
+        This is another line
         This is line 3
<BLANKLINE>

Such comparisons can still be confusing as white space is taken into
account. If you need to care about whitespace characters, you can make
spotting the differences easier as follows:

>>> compare("\tline 1\r\nline 2"," line1 \nline 2", show_whitespace=True)
Traceback (most recent call last):
 ...
AssertionError: 
--- first
+++ second
@@ -1,2 +1,2 @@
-'\tline 1\r\n'
+' line1 \n'
 'line 2'

However, you may not care about some of the whitespace involved. To
help with this, :func:`compare` has two options that can be set to
ignore certain types of whitespace.

If you wish to compare two strings that contain blank lines or lines
containing only whitespace characters, but where you only care about
the content, you can use the following:

.. code-block:: python

  compare('line1\nline2', 'line1\n \nline2\n\n',
          blanklines=False)

If you wish to compare two strings made up of lines that may have
trailing whitespace that you don't care about, you can do so with the
following: 

.. code-block:: python

  compare('line1\nline2', 'line1 \t\nline2   \n',
          trailing_whitespace=False)

Recursive comparison
~~~~~~~~~~~~~~~~~~~~

Where :func:`compare` is able to provide a descriptive comparison for
a particular type, it will then recurse to do the same for the
elements contained within objects of that type. 
For example, when comparing a list of dictionaries, the description
will not only tell you where in the list the difference occurred, but
also what the differences were within the dictionaries that weren't
equal:

>>> compare([{'one': 1}, {'two': 2, 'text':'foo\nbar\nbaz'}],
...         [{'one': 1}, {'two': 2, 'text':'foo\nbob\nbaz'}])
Traceback (most recent call last):
 ...
AssertionError: sequence not as expected:
<BLANKLINE>
same:
[{'one': 1}]
<BLANKLINE>
first:
[{'text': 'foo\nbar\nbaz', 'two': 2}]
<BLANKLINE>
second:
[{'text': 'foo\nbob\nbaz', 'two': 2}]
<BLANKLINE>
While comparing [1]: dict not as expected:
<BLANKLINE>
same:
['two']
<BLANKLINE>
values differ:
'text': 'foo\nbar\nbaz' != 'foo\nbob\nbaz'
<BLANKLINE>
While comparing [1]['text']: 
--- first
+++ second
@@ -1,3 +1,3 @@
 foo
-bar
+bob
 baz

This also applies to any comparers you have provided, as can be seen
in the next section.

.. _comparer-register:

Providing your own comparers
~~~~~~~~~~~~~~~~~~~~~~~~~~~~

When using :meth:`compare` frequently for your own complex objects,
it can be beneficial to give more descriptive output when two objects
don't compare as equal.

.. note:: 

    If you are reading this section as a result of needing to test
    objects that don't natively support comparison, or as a result of
    needing to infrequently compare your own subclasses of python
    basic types, take a look at :ref:`comparison-objects` as this may
    well be an easier solution.

.. invisible-code-block: python

  from testfixtures.comparison import _registry, compare_sequence
  from testfixtures import Replacer
  r = Replacer()
  r.replace('testfixtures.comparison._registry', {
      list: compare_sequence
      })

As an example, suppose you have a class whose instances have a
timestamp and a name as attributes, but you'd like to ignore the
timestamp when comparing:

.. code-block:: python

  from datetime import datetime
  
  class MyObject(object):
      def __init__(self, name):
          self.timestamp = datetime.now()
          self.name = name

To compare lots of these, you would first write a comparer:

.. code-block:: python

  def compare_my_object(x, y, context):
      if x.name == y.name:
          return
      return 'MyObject named %s != MyObject named %s' % (
          context.label('x', repr(x.name)),
          context.label('y', repr(y.name)),
          )

Then you'd register that comparer for your type:

.. code-block:: python

  from testfixtures.comparison import register
  register(MyObject, compare_my_object)

.. invisible-code-block: python

  import testfixtures.comparison
  assert testfixtures.comparison._registry == {
      MyObject: compare_my_object, list: compare_sequence,
      }

Now, it'll get used when comparing objects of that type,
even if they're contained within other objects:

>>> compare([1, MyObject('foo')], [1, MyObject('bar')])
Traceback (most recent call last):
 ...
AssertionError: sequence not as expected:
<BLANKLINE>
same:
[1]
<BLANKLINE>
first:
[<MyObject ...>]
<BLANKLINE>
second:
[<MyObject ...>]
<BLANKLINE>
While comparing [1]: MyObject named 'foo' != MyObject named 'bar'

From this example, you can also see that a comparer can indicate that
two objects are equal, for :func:`compare`'s purposes, by returning
``None``:

>>> MyObject('foo') == MyObject('foo')
False
>>> compare(MyObject('foo'), MyObject('foo'))

You can also see that you can, and should, use the context object passed in
to add labels to the representations of the objects being compared if the
comparison fails:

>>> compare(expected=MyObject('foo'), actual=MyObject('bar'))
Traceback (most recent call last):
 ...
AssertionError: MyObject named 'foo' (expected) != MyObject named 'bar' (actual)

.. invisible-code-block: python

  r.restore()

  # set up for the next test
  r = Replacer()
  r.replace('testfixtures.comparison._registry', {})

It may be that you only want to use a comparer or set of
comparers for a particular test. If that's the case, you can pass
:func:`compare` a ``comparers`` parameter consisting of a
dictionary that maps types to comparers. These will be added to the
global registry for the duration of the call:

>>> compare(MyObject('foo'), MyObject('bar'),
...         comparers={MyObject: compare_my_object})
Traceback (most recent call last):
 ...
AssertionError: MyObject named 'foo' != MyObject named 'bar'

.. invisible-code-block: python

  import testfixtures.comparison
  assert testfixtures.comparison._registry == {}
  r.restore()

A full list of the available comparers included can be found below the
API documentation for :func:`compare`. These make good candidates for
registering for your own classes, if they provide the necessary
behaviour, and their source is also good to read when wondering how to
implement your own comparers.

You may be wondering what the ``context`` object passed to the
comparer is for; it allows you to hand off comparison of parts of the
two objects currently being compared back to the :func:`compare`
machinery, it also allows you to pass options to your comparison
function. 

For example, you may have an object that has a couple of dictionaries
as attributes:

.. code-block:: python

  from datetime import datetime
  
  class Request(object):
      def __init__(self, uri, headers, body):
          self.uri = uri
          self.headers = headers
          self.body = body

When your tests encounter instances of these that are not as expected,
you want feedback about which bits of the request or response weren't
as expected. This can be achieved by implementing a comparer as
follows:

.. code-block:: python

   def compare_request(x, y, context):
       uri_different = x.uri != y.uri
       headers_different = context.different(x.headers, y.headers, '.headers')
       body_different = context.different(x.body, y.body, '.body')
       if uri_different or headers_different or body_different:
           return 'Request for %r != Request for %r' % (
               x.uri, y.uri
           )

.. note:: 

  A comparer should always return some text when it considers
  the two objects it is comparing to be different.

This comparer can either be registered globally or passed to each
:func:`compare` call and will give detailed feedback about how the
requests were different:

>>> compare(Request('/foo', {'method': 'POST'}, {'my_field': 'value_1'}),
...         Request('/foo', {'method': 'GET'}, {'my_field': 'value_2'}),
...         comparers={Request: compare_request})
Traceback (most recent call last):
 ...
AssertionError: Request for '/foo' != Request for '/foo'
<BLANKLINE>
While comparing .headers: dict not as expected:
<BLANKLINE>
values differ:
'method': 'POST' != 'GET'
<BLANKLINE>
While comparing .headers['method']: 'POST' != 'GET'
<BLANKLINE>
While comparing .body: dict not as expected:
<BLANKLINE>
values differ:
'my_field': 'value_1' != 'value_2'
<BLANKLINE>
While comparing .body['my_field']: 'value_1' != 'value_2'

As an example of passing options through to a comparer, suppose you
wanted to compare all decimals in a nested data structure by rounding
them to a number of decimal places that varies from test to test. The
comparer could be implemented and registered as follows:

.. invisible-code-block: python

  from testfixtures.comparison import _registry
  r = Replacer()
  r.replace('testfixtures.comparison._registry', dict(_registry))

.. code-block:: python

  from decimal import Decimal
  from testfixtures.comparison import register

  def compare_decimal(x, y, context):
       precision = context.get_option('precision', 2)
       if round(x, precision) != round(y, precision):
           return '%r != %r when rounded to %i decimal places' % (
               x, y, precision
           )

  register(Decimal, compare_decimal)

Now, this comparer will be used for comparing all decimals and the
precision used will be that passed to :func:`compare`:

>>> expected_order = {'price': Decimal('1.234'), 'quantity': 5}
>>> actual_order = {'price': Decimal('1.236'), 'quantity': 5}
>>> compare(expected_order, actual_order, precision=1)
>>> compare(expected_order, actual_order, precision=3)
Traceback (most recent call last):
 ...
AssertionError: dict not as expected:
<BLANKLINE>
same:
['quantity']
<BLANKLINE>
values differ:
'price': Decimal('1.234') != Decimal('1.236')
<BLANKLINE>
While comparing ['price']: Decimal('1.234') != Decimal('1.236') when rounded to 3 decimal places

If no precision is passed, the default of ``2`` will be used:

>>> compare(Decimal('2.006'), Decimal('2.009'))
>>> compare(Decimal('2.001'), Decimal('2.009'))
Traceback (most recent call last):
 ...
AssertionError: Decimal('2.001') != Decimal('2.009') when rounded to 2 decimal places

.. invisible-code-block: python

  r.restore()

.. _strict-comparison:

Ignoring ``__eq__``
~~~~~~~~~~~~~~~~~~~

Some objects, such as those from the Django ORM, have pretty broken
implementations or ``__eq__``. Since :func:`compare` normally relies on this,
it can result in objects appearing to be equal when they are not.

Take this class, for example:

.. code-block:: python

  class OrmObj(object):
      def __init__(self, a):
          self.a = a
      def __eq__(self, other):
          return True
      def __repr__(self):
          return 'OrmObj: '+str(self.a)

If we compare normally, we erroneously understand the objects to be equal:

>>> compare(actual=OrmObj(1), expected=OrmObj(2))

In order to get a sane comparison, we need to both supply a custom comparer
as described above, and use the ``ignore_eq`` parameter:

.. code-block:: python

  def compare_orm_obj(x, y, context):
      if x.a != y.a:
          return 'OrmObj: %s != %s' % (x.a, y.a)

>>> compare(actual=OrmObj(1), expected=OrmObj(2),
...         comparers={OrmObj: compare_orm_obj}, ignore_eq=True)
Traceback (most recent call last):
...
AssertionError: OrmObj: 2 != 1

Strict comparison
~~~~~~~~~~~~~~~~~

If is it important that the two values being compared are of exactly
the same type, rather than just being equal as far as Python is
concerned, then the strict mode of :func:`compare` should be used.

For example, these two instances will normally appear to be equal
provided the elements within them are the same:

>>> TypeA = namedtuple('A', 'x')
>>> TypeB = namedtuple('B', 'x')
>>> compare(TypeA(1), TypeB(1))

If this type difference is important, then the `strict` parameter
should be used:

>>> compare(TypeA(1), TypeB(1), strict=True)
Traceback (most recent call last):
 ...
AssertionError: A(x=1) (<class '__main__.A'>) != B(x=1) (<class '__main__.B'>)

.. _comparison-objects:

Comparison objects
------------------

Another common problem with the checking in tests is that not all
objects support comparison and nor should they need to. For this
reason, TextFixtures provides the :class:`~testfixtures.Comparison`
class.

This class lets you instantiate placeholders that can be used to
compare expected results with actual results where objects in the
actual results do not support useful comparison.  

Comparisons will appear to be equal to any object they are compared
with that matches their specification. For example, take the following
class: 

.. code-block:: python

  class SomeClass:

      def __init__(self, x, y):
         self.x, self.y = x, y

Normal comparison doesn't work, which makes testing tricky:

>>> SomeClass(1, 2) == SomeClass(1, 2)
False

Here's how this comparison can be done:

>>> from testfixtures import Comparison as C
>>> C(SomeClass, x=1, y=2) == SomeClass(1, 2)
True

Perhaps even more importantly, when a comparison fails, its
representation changes to give information about what went wrong. The
common idiom for using comparisons is in conjuction with
:meth:`~unittest.TestCase.assertEqual` or
:meth:`~testfixtures.compare`: 

>>> compare(C(SomeClass, x=2), SomeClass(1, 2))
Traceback (most recent call last):
 ...
AssertionError: 
  <C(failed):...SomeClass>
  x:2 != 1
  y:2 not in Comparison
  </C> != <...SomeClass...>

The key is that the comparison object actually stores information
about what it was last compared with. The following example shows this
more clearly: 

>>> c = C(SomeClass, x=2)
>>> print(repr(c))
<BLANKLINE>
  <C:...SomeClass>
  x:2
  </C>
>>> c == SomeClass(1, 2)
False
>>> print(repr(c))
<BLANKLINE>
  <C(failed):...SomeClass>
  x:2 != 1
  y:2 not in Comparison
  </C>

.. note:: 

   :meth:`~unittest.TestCase.assertEqual` has regressed in Python 3.4
   and will now truncate the text shown in assertions with no way to
   configure this behaviour. Use :func:`compare` instead, which will
   give you other desirable behaviour as well as showing you the full
   output of failed comparisons.

Types of comparison
~~~~~~~~~~~~~~~~~~~

There are several ways a comparison can be set up depending on what
you want to check.

If you only care about the class of an object, you can set up the
comparison with only the class:

>>> C(SomeClass) == SomeClass(1, 2)
True

This can also be achieved by specifying the type of the object as a
dotted name:

>>> import sys
>>> C('types.ModuleType') == sys
True

Alternatively, if you happen to have a non-comparable object already
around, comparison can be done with it:

>>> C(SomeClass(1,2)) == SomeClass(1,2)
True

If you only care about certain attributes, this can also easily be
achieved with the `strict` parameter: 

>>> C(SomeClass, x=1, strict=False) == SomeClass(1, 2)
True

The above can be problematic if you want to compare an object with
attibutes that share names with parameters to the :class:`~testfixtures.Comparison`
constructor. For this reason, you can pass the attributes in a
dictionary:

>>> compare(C(SomeClass, {'strict':3}, strict=False), SomeClass(1, 2))
Traceback (most recent call last):
 ...
AssertionError: 
  <C(failed):...SomeClass>
  strict:3 not in other
  </C> != <...SomeClass...>

Gotchas
~~~~~~~

There are a few things to be careful of when using comparisons:

.. work around Manuel bug :-(
.. invisible-code-block: python

  class NoVars(object):
      __slots__ = ['x']

- The default strict comparison cannot be used with a class such as
  the following:

  .. code-block:: python

    class NoVars(object):
         __slots__ = ['x']

  If you try, you will get an error that explains the problem:

  >>> C(NoVars, x=1) == NoVars()
  Traceback (most recent call last):
   ...
  TypeError: <NoVars object at ...> does not support vars() so cannot do strict comparison

  Comparisons can still be done with classes that don't support
  ``vars()``, they just need to be non-strict:

  >>> nv = NoVars()
  >>> nv.x = 1
  >>> C(NoVars, x=1, strict=False) == nv
  True
  
.. work around Manuel bug :-(
.. invisible-code-block: python
        class SomeModel:
            def __eq__(self,other):
                if isinstance(other,SomeModel):
                    return True
                return False

- If the object being compared has an ``__eq__`` method, such as
  Django model instances, then the :class:`~testfixtures.Comparison`
  must be the first object in the equality check.

  The following class is an example of this:

  .. code-block:: python

        class SomeModel:
            def __eq__(self,other):
                if isinstance(other,SomeModel):
                    return True
                return False
  
  It will not work correctly if used as the second object in the
  expression:

  >>> SomeModel()==C(SomeModel)
  False

  However, if the comparison is correctly placed first, then
  everything will behave as expected:

  >>> C(SomeModel)==SomeModel()
  True

- It probably goes without saying, but comparisons should not be used
  on both sides of an equality check:

  >>> C(SomeClass)==C(SomeClass)
  False

Round Comparison objects
-------------------------

When comparing numerics you often want to be able to compare to a 
given precision to allow for rounding issues which make precise
equality impossible.

For these situations, you can use :class:`RoundComparison` objects
wherever you would use floats or Decimals, and they will compare equal to
any float or Decimal that matches when both sides are rounded to the
specified precision.

Here's an example:

.. code-block:: python

  from testfixtures import compare, RoundComparison as R

  compare(R(1234.5678, 2), 1234.5681)

.. note:: 

  You should always pass the same type of object to the
  :class:`RoundComparison` object as you intend compare it with. If
  the type of the rounded expected value is not the same as the type of
  the rounded value being compared against it, a :class:`TypeError`
  will be raised.

Range Comparison objects
-------------------------

When comparing orderable types just as numbers, dates and time, you may only
know what range a value will fall into. :class:`RangeComparison` objects
let you confirm a value is within a certain tolerance or range.

Here's an example:

.. code-block:: python

  from testfixtures import compare, RangeComparison as R

  compare(R(123.456, 789), Decimal(555.01))

.. note::

  :class:`RangeComparison` is inclusive of both the lower and upper bound.

String Comparison objects
-------------------------

When comparing sequences of strings, particularly those comping from
things like the python logging package, you often end up wanting to
express a requirement that one string should be almost like another,
or maybe fit a particular regular expression.

For these situations, you can use :class:`StringComparison` objects
wherever you would use normal strings, and they will compare equal to
any string that matches the regular expression they are created with.

Here's an example:

.. code-block:: python

  from testfixtures import compare, StringComparison as S

  compare(S('Starting thread \d+'),'Starting thread 132356')

Differentiating chunks of text
------------------------------

TextFixtures provides a function that will compare two strings and
give a unified diff as a result. This can be handy as a third
parameter to :meth:`~unittest.TestCase.assertEqual` or just as a
general utility function for comparing two lumps of text.

As an example:

>>> from testfixtures import diff
>>> print(diff('line1\nline2\nline3',
...            'line1\nlineA\nline3'))
--- first
+++ second
@@ -1,3 +1,3 @@
 line1
-line2
+lineA
 line3