NEWS

testtools NEWS
++++++++++++++

Changes and improvements to testtools_, grouped by release.


NEXT
~~~~

0.9.32
~~~~~~

Regular maintenance release.  Special thanks to new contributor, Xiao Hanyu!

Improvements
------------

* Stacktrace filtering no longer hides unittest frames that are surrounded by
  user frames. We will reenable this when we figure out a better algorithm for
  retaining meaning. (Robert Collins, #1188420)

* The compatibility code for skipped tests with unittest2 was broken.
  (Robert Collins, #1190951)

* Various documentation improvements (Clint Byrum, Xiao Hanyu).

0.9.31
~~~~~~

Improvements
------------

* ``ExpectedException`` now accepts a msg parameter for describing an error,
  much the same as assertEquals etc. (Robert Collins)

0.9.30
~~~~~~

A new sort of TestResult, the StreamResult has been added, as a prototype for
a revised standard library test result API.  Expect this API to change.
Although we will try to preserve compatibility for early adopters, it is
experimental and we might need to break it if it turns out to be unsuitable.

Improvements
------------
* ``assertRaises`` works properly for exception classes that have custom 
  metaclasses

* ``ConcurrentTestSuite`` was silently eating exceptions that propagate from
  the test.run(result) method call. Ignoring them is fine in a normal test
  runner, but when they happen in a different thread, the thread that called
  suite.run() is not in the stack anymore, and the exceptions are lost. We now
  create a synthetic test recording any such exception.
  (Robert Collins, #1130429)

* Fixed SyntaxError raised in ``_compat2x.py`` when installing via Python 3.
  (Will Bond, #941958)

* New class ``StreamResult`` which defines the API for the new result type.
  (Robert Collins)

* New support class ``ConcurrentStreamTestSuite`` for convenient construction
  and utilisation of ``StreamToQueue`` objects. (Robert Collins)

* New support class ``CopyStreamResult`` which forwards events onto multiple
  ``StreamResult`` objects (each of which receives all the events).
  (Robert Collins)

* New support class ``StreamSummary`` which summarises a ``StreamResult``
  stream compatibly with ``TestResult`` code. (Robert Collins)

* New support class ``StreamTagger`` which adds or removes tags from
  ``StreamResult`` events. (RobertCollins)

* New support class ``StreamToDict`` which converts a ``StreamResult`` to a
  series of dicts describing a test. Useful for writing trivial stream
  analysers. (Robert Collins)

* New support class ``TestControl`` which permits cancelling an in-progress
  run. (Robert Collins)

* New support class ``StreamFailFast`` which calls a ``TestControl`` instance
  to abort the test run when a failure is detected. (Robert Collins)

* New support class ``ExtendedToStreamDecorator`` which translates both regular
  unittest TestResult API calls and the ExtendedTestResult API which testtools
  has supported into the StreamResult API. ExtendedToStreamDecorator also
  forwards calls made in the StreamResult API, permitting it to be used
  anywhere a StreamResult is used. Key TestResult query methods like
  wasSuccessful and shouldStop are synchronised with the StreamResult API
  calls, but the detailed statistics like the list of errors are not - a
  separate consumer will be created to support that.
  (Robert Collins)

* New support class ``StreamToExtendedDecorator`` which translates
  ``StreamResult`` API calls into ``ExtendedTestResult`` (or any older
  ``TestResult``) calls. This permits using un-migrated result objects with
  new runners / tests. (Robert Collins)

* New support class ``StreamToQueue`` for sending messages to one
  ``StreamResult`` from multiple threads. (Robert Collins)

* New support class ``TimestampingStreamResult`` which adds a timestamp to
  events with no timestamp. (Robert Collins)

* New ``TestCase`` decorator ``DecorateTestCaseResult`` that adapts the
  ``TestResult`` or ``StreamResult`` a case will be run with, for ensuring that
  a particular result object is used even if the runner running the test doesn't
  know to use it. (Robert Collins)

* New test support class ``testtools.testresult.doubles.StreamResult``, which
  captures all the StreamResult events. (Robert Collins)

* ``PlaceHolder`` can now hold tags, and applies them before, and removes them
  after, the test. (Robert Collins)

* ``PlaceHolder`` can now hold timestamps, and applies them before the test and
  then before the outcome. (Robert Collins)

* ``StreamResultRouter`` added. This is useful for demultiplexing - e.g. for
  partitioning analysis of events or sending feedback encapsulated in
  StreamResult events back to their source. (Robert Collins)

* ``testtools.run.TestProgram`` now supports the ``TestRunner`` taking over
  responsibility for formatting the output of ``--list-tests``.
  (Robert Collins)

* The error message for setUp and tearDown upcall errors was broken on Python
  3.4. (Monty Taylor, Robert Collins, #1140688)

* The repr of object() on pypy includes the object id, which was breaking a
  test that accidentally depended on the CPython repr for object().
  (Jonathan Lange)

0.9.29
~~~~~~

A simple bug fix, and better error messages when you don't up-call.

Changes
-------

* ``testtools.content_type.ContentType`` incorrectly used ',' rather than ';'
  to separate parameters. (Robert Collins)

Improvements
------------

* ``testtools.compat.unicode_output_stream`` was wrapping a stream encoder
  around ``io.StringIO`` and ``io.TextIOWrapper`` objects, which was incorrect.
  (Robert Collins)

* Report the name of the source file for setUp and tearDown upcall errors.
  (Monty Taylor)

0.9.28
~~~~~~

Testtools has moved VCS - https://github.com/testing-cabal/testtools/ is
the new home. Bug tracking is still on Launchpad, and releases are on Pypi.

We made this change to take advantage of the richer ecosystem of tools around
Git, and to lower the barrier for new contributors.

Improvements
------------

* New ``testtools.testcase.attr`` and ``testtools.testcase.WithAttributes``
  helpers allow marking up test case methods with simple labels. This permits
  filtering tests with more granularity than organising them into modules and
  test classes. (Robert Collins)

0.9.27
~~~~~~

Improvements
------------

* New matcher ``HasLength`` for matching the length of a collection.
  (Robert Collins)

* New matcher ``MatchesPredicateWithParams`` make it still easier to create
  ad hoc matchers. (Robert Collins)

* We have a simpler release process in future - see doc/hacking.rst.
  (Robert Collins)

0.9.26
~~~~~~

Brown paper bag fix: failed to document the need for setup to be able to use
extras. Compounded by pip not supporting setup_requires.

Changes
-------

* setup.py now can generate egg_info even if extras is not available.
  Also lists extras in setup_requires for easy_install.
  (Robert Collins, #1102464)

0.9.25
~~~~~~

Changes
-------

* ``python -m testtools.run --load-list`` will now preserve any custom suites
  (such as ``testtools.FixtureSuite`` or ``testresources.OptimisingTestSuite``)
  rather than flattening them.
  (Robert Collins, #827175)

* Testtools now depends on extras, a small library split out from it to contain
  generally useful non-testing facilities. Since extras has been around for a
  couple of testtools releases now, we're making this into a hard dependency of
  testtools. (Robert Collins)

* Testtools now uses setuptools rather than distutils so that we can document
  the extras dependency. (Robert Collins)

Improvements
------------

* Testtools will no longer override test code registered details called
  'traceback' when reporting caught exceptions from test code.
  (Robert Collins, #812793)

0.9.24
~~~~~~

Changes
-------

* ``testtools.run discover`` will now sort the tests it discovered. This is a 
  workaround for http://bugs.python.org/issue16709. Non-standard test suites
  are preserved, and their ``sort_tests()`` method called (if they have such an
  attribute). ``testtools.testsuite.sorted_tests(suite, True)`` can be used by
  such suites to do a local sort. (Robert Collins, #1091512)

* ``ThreadsafeForwardingResult`` now defines a stub ``progress`` method, which
  fixes ``testr run`` of streams containing progress markers (by discarding the 
  progress data). (Robert Collins, #1019165)

0.9.23
~~~~~~

Changes
-------

* ``run.TestToolsTestRunner`` now accepts the verbosity, buffer and failfast
  arguments the upstream python TestProgram code wants to give it, making it
  possible to support them in a compatible fashion. (Robert Collins)

Improvements
------------

* ``testtools.run`` now supports the ``-f`` or ``--failfast`` parameter.
  Previously it was advertised in the help but ignored.
  (Robert Collins, #1090582)

* ``AnyMatch`` added, a new matcher that matches when any item in a collection
  matches the given matcher.  (Jonathan Lange)

* Spelling corrections to documentation.  (Vincent Ladeuil)

* ``TestProgram`` now has a sane default for its ``testRunner`` argument.
  (Vincent Ladeuil)

* The test suite passes on Python 3 again. (Robert Collins)

0.9.22
~~~~~~

Improvements
------------

* ``content_from_file`` and ``content_from_stream`` now accept seek_offset and
  seek_whence parameters allowing them to be used to grab less than the full
  stream, or to be used with StringIO streams. (Robert Collins, #1088693)

0.9.21
~~~~~~

Improvements
------------

* ``DirContains`` correctly exposed, after being accidentally hidden in the
  great matcher re-organization of 0.9.17.  (Jonathan Lange)


0.9.20
~~~~~~

Three new matchers that'll rock your world.

Improvements
------------

* New, powerful matchers that match items in a dictionary:

  - ``MatchesDict``, match every key in a dictionary with a key in a
    dictionary of matchers.  For when the set of expected keys is equal to the
    set of observed keys.

  - ``ContainsDict``, every key in a dictionary of matchers must be
    found in a dictionary, and the values for those keys must match.  For when
    the set of expected keys is a subset of the set of observed keys.

  - ``ContainedByDict``, every key in a dictionary must be found in
    a dictionary of matchers.  For when the set of expected keys is a superset
    of the set of observed keys.

  The names are a little confusing, sorry.  We're still trying to figure out
  how to present the concept in the simplest way possible.


0.9.19
~~~~~~

How embarrassing!  Three releases in two days.

We've worked out the kinks and have confirmation from our downstreams that
this is all good.  Should be the last release for a little while.  Please
ignore 0.9.18 and 0.9.17.

Improvements
------------

* Include the matcher tests in the release, allowing the tests to run and
  pass from the release tarball.  (Jonathan Lange)

* Fix cosmetic test failures in Python 3.3, introduced during release 0.9.17.
  (Jonathan Lange)


0.9.18
~~~~~~

Due to an oversight, release 0.9.18 did not contain the new
``testtools.matchers`` package and was thus completely broken.  This release
corrects that, returning us all to normality.

0.9.17
~~~~~~

This release brings better discover support and Python3.x improvements. There
are still some test failures on Python3.3 but they are cosmetic - the library
is as usable there as on any other Python 3 release.

Changes
-------

* The ``testtools.matchers`` package has been split up.  No change to the
  public interface.  (Jonathan Lange)

Improvements
------------

* ``python -m testtools.run discover . --list`` now works. (Robert Collins)

* Correctly handling of bytes vs text in JSON content type. (Martin [gz])


0.9.16
~~~~~~

Some new matchers and a new content helper for JSON content.

This is the first release of testtools to drop support for Python 2.4 and 2.5.
If you need support for either of those versions, please use testtools 0.9.15.

Improvements
------------

* New content helper, ``json_content`` (Jonathan Lange)

* New matchers:

  * ``ContainsAll`` for asserting one thing is a subset of another
    (Raphaël Badin)

  * ``SameMembers`` for asserting two iterators have the same members.
    (Jonathan Lange)

* Reraising of exceptions in Python 3 is more reliable. (Martin [gz])


0.9.15
~~~~~~

This is the last release to support Python2.4 and 2.5. It brings in a slew of
improvements to test tagging and concurrency, making running large test suites
with partitioned workers more reliable and easier to reproduce exact test
ordering in a given worker. See our sister project ``testrepository`` for a
test runner that uses these features.

Changes
-------

* ``PlaceHolder`` and ``ErrorHolder`` now support being given result details.
  (Robert Collins)

* ``ErrorHolder`` is now just a function - all the logic is in ``PlaceHolder``.
  (Robert Collins)

* ``TestResult`` and all other ``TestResult``-like objects in testtools
  distinguish between global tags and test-local tags, as per the subunit
  specification.  (Jonathan Lange)

* This is the **last** release of testtools that supports Python 2.4 or 2.5.
  These releases are no longer supported by the Python community and do not
  receive security updates. If this affects you, you will need to either
  stay on this release or perform your own backports.
  (Jonathan Lange, Robert Collins)

* ``ThreadsafeForwardingResult`` now forwards global tags as test-local tags,
  making reasoning about the correctness of the multiplexed stream simpler.
  This preserves the semantic value (what tags apply to a given test) while
  consuming less stream size (as no negative-tag statement is needed).
  (Robert Collins, Gary Poster, #986434)

Improvements
------------

* API documentation corrections. (Raphaël Badin)

* ``ConcurrentTestSuite`` now takes an optional ``wrap_result`` parameter
  that can be used to wrap the ``ThreadsafeForwardingResults`` created by
  the suite.  (Jonathan Lange)

* ``Tagger`` added.  It's a new ``TestResult`` that tags all tests sent to
  it with a particular set of tags.  (Jonathan Lange)

* ``testresultdecorator`` brought over from subunit.  (Jonathan Lange)

* All ``TestResult`` wrappers now correctly forward ``current_tags`` from
  their wrapped results, meaning that ``current_tags`` can always be relied
  upon to return the currently active tags on a test result.

* ``TestByTestResult``, a ``TestResult`` that calls a method once per test,
  added.  (Jonathan Lange)

* ``ThreadsafeForwardingResult`` correctly forwards ``tags()`` calls where
  only one of ``new_tags`` or ``gone_tags`` are specified.
  (Jonathan Lange, #980263)

* ``ThreadsafeForwardingResult`` no longer leaks local tags from one test
  into all future tests run.  (Jonathan Lange, #985613)

* ``ThreadsafeForwardingResult`` has many, many more tests.  (Jonathan Lange)


0.9.14
~~~~~~

Our sister project, `subunit <https://launchpad.net/subunit>`_, was using a
private API that was deleted in the 0.9.13 release.  This release restores
that API in order to smooth out the upgrade path.

If you don't use subunit, then this release won't matter very much to you.


0.9.13
~~~~~~

Plenty of new matchers and quite a few critical bug fixes (especially to do
with stack traces from failed assertions).  A net win for all.

Changes
-------

* ``MatchesAll`` now takes an ``first_only`` keyword argument that changes how
  mismatches are displayed.  If you were previously passing matchers to
  ``MatchesAll`` with keyword arguments, then this change might affect your
  test results.  (Jonathan Lange)

Improvements
------------

* Actually hide all of the testtools stack for assertion failures. The
  previous release promised clean stack, but now we actually provide it.
  (Jonathan Lange, #854769)

* ``assertRaises`` now includes the ``repr`` of the callable that failed to raise
  properly. (Jonathan Lange, #881052)

* Asynchronous tests no longer hang when run with trial.
  (Jonathan Lange, #926189)

* ``Content`` objects now have an ``as_text`` method to convert their contents
  to Unicode text.  (Jonathan Lange)

* Failed equality assertions now line up. (Jonathan Lange, #879339)

* ``FullStackRunTest`` no longer aborts the test run if a test raises an
  error.  (Jonathan Lange)

* ``MatchesAll`` and ``MatchesListwise`` both take a ``first_only`` keyword
  argument.  If True, they will report only on the first mismatch they find,
  and not continue looking for other possible mismatches.
  (Jonathan Lange)

* New helper, ``Nullary`` that turns callables with arguments into ones that
  don't take arguments.  (Jonathan Lange)

* New matchers:

  * ``DirContains`` matches the contents of a directory.
    (Jonathan Lange, James Westby)

  * ``DirExists`` matches if a directory exists.
    (Jonathan Lange, James Westby)

  * ``FileContains`` matches the contents of a file.
    (Jonathan Lange, James Westby)

  * ``FileExists`` matches if a file exists.
    (Jonathan Lange, James Westby)

  * ``HasPermissions`` matches the permissions of a file.  (Jonathan Lange)

  * ``MatchesPredicate`` matches if a predicate is true.  (Jonathan Lange)

  * ``PathExists`` matches if a path exists.  (Jonathan Lange, James Westby)

  * ``SamePath`` matches if two paths are the same.  (Jonathan Lange)

  * ``TarballContains`` matches the contents of a tarball.  (Jonathan Lange)

* ``MultiTestResult`` supports the ``tags`` method.
  (Graham Binns, Francesco Banconi, #914279)

* ``ThreadsafeForwardingResult`` supports the ``tags`` method.
  (Graham Binns, Francesco Banconi, #914279)

* ``ThreadsafeForwardingResult`` no longer includes semaphore acquisition time
  in the test duration (for implicitly timed test runs).
  (Robert Collins, #914362)

0.9.12
~~~~~~

This is a very big release.  We've made huge improvements on three fronts:
 1. Test failures are way nicer and easier to read
 2. Matchers and ``assertThat`` are much more convenient to use
 3. Correct handling of extended unicode characters

We've trimmed off the fat from the stack trace you get when tests fail, we've
cut out the bits of error messages that just didn't help, we've made it easier
to annotate mismatch failures, to compare complex objects and to match raised
exceptions.

Testing code was never this fun.

Changes
-------

* ``AfterPreproccessing`` renamed to ``AfterPreprocessing``, which is a more
  correct spelling.  Old name preserved for backwards compatibility, but is
  now deprecated.  Please stop using it.
  (Jonathan Lange, #813460)

* ``assertThat`` raises ``MismatchError`` instead of
  ``TestCase.failureException``.  ``MismatchError`` is a subclass of
  ``AssertionError``, so in most cases this change will not matter. However,
  if ``self.failureException`` has been set to a non-default value, then
  mismatches will become test errors rather than test failures.

* ``gather_details`` takes two dicts, rather than two detailed objects.
  (Jonathan Lange, #801027)

* ``MatchesRegex`` mismatch now says "<value> does not match /<regex>/" rather
  than "<regex> did not match <value>". The regular expression contains fewer
  backslashes too. (Jonathan Lange, #818079)

* Tests that run with ``AsynchronousDeferredRunTest`` now have the ``reactor``
  attribute set to the running reactor. (Jonathan Lange, #720749)

Improvements
------------

* All public matchers are now in ``testtools.matchers.__all__``.
  (Jonathan Lange, #784859)

* ``assertThat`` can actually display mismatches and matchers that contain
  extended unicode characters. (Jonathan Lange, Martin [gz], #804127)

* ``assertThat`` output is much less verbose, displaying only what the mismatch
  tells us to display. Old-style verbose output can be had by passing
  ``verbose=True`` to assertThat. (Jonathan Lange, #675323, #593190)

* ``assertThat`` accepts a message which will be used to annotate the matcher.
  This can be given as a third parameter or as a keyword parameter.
  (Robert Collins)

* Automated the Launchpad part of the release process.
  (Jonathan Lange, #623486)

* Correctly display non-ASCII unicode output on terminals that claim to have a
  unicode encoding. (Martin [gz], #804122)

* ``DocTestMatches`` correctly handles unicode output from examples, rather
  than raising an error. (Martin [gz], #764170)

* ``ErrorHolder`` and ``PlaceHolder`` added to docs. (Jonathan Lange, #816597)

* ``ExpectedException`` now matches any exception of the given type by
  default, and also allows specifying a ``Matcher`` rather than a mere regular
  expression. (Jonathan Lange, #791889)

* ``FixtureSuite`` added, allows test suites to run with a given fixture.
  (Jonathan Lange)

* Hide testtools's own stack frames when displaying tracebacks, making it
  easier for test authors to focus on their errors.
  (Jonathan Lange, Martin [gz], #788974)

* Less boilerplate displayed in test failures and errors.
  (Jonathan Lange, #660852)

* ``MatchesException`` now allows you to match exceptions against any matcher,
  rather than just regular expressions.  (Jonathan Lange, #791889)

* ``MatchesException`` now permits a tuple of types rather than a single type
  (when using the type matching mode).  (Robert Collins)

* ``MatchesStructure.byEquality`` added to make the common case of matching
  many attributes by equality much easier.  ``MatchesStructure.byMatcher``
  added in case folk want to match by things other than equality.
  (Jonathan Lange)

* New convenience assertions, ``assertIsNone`` and ``assertIsNotNone``.
  (Christian Kampka)

* New matchers:

  * ``AllMatch`` matches many values against a single matcher.
    (Jonathan Lange, #615108)

  * ``Contains``. (Robert Collins)

  * ``GreaterThan``. (Christian Kampka)

* New helper, ``safe_hasattr`` added. (Jonathan Lange)

* ``reraise`` added to ``testtools.compat``. (Jonathan Lange)


0.9.11
~~~~~~

This release brings consistent use of super for better compatibility with
multiple inheritance, fixed Python3 support, improvements in fixture and mather
outputs and a compat helper for testing libraries that deal with bytestrings.

Changes
-------

* ``TestCase`` now uses super to call base ``unittest.TestCase`` constructor,
  ``setUp`` and ``tearDown``. (Tim Cole, #771508)

* If, when calling ``useFixture`` an error occurs during fixture set up, we
  still attempt to gather details from the fixture. (Gavin Panella)


Improvements
------------

* Additional compat helper for ``BytesIO`` for libraries that build on
  testtools and are working on Python 3 porting. (Robert Collins)

* Corrected documentation for ``MatchesStructure`` in the test authors
  document.  (Jonathan Lange)

* ``LessThan`` error message now says something that is logically correct.
  (Gavin Panella, #762008)

* Multiple details from a single fixture are now kept separate, rather than
  being mooshed together. (Gavin Panella, #788182)

* Python 3 support now back in action. (Martin [gz], #688729)

* ``try_import`` and ``try_imports`` have a callback that is called whenever
  they fail to import a module.  (Martin Pool)


0.9.10
~~~~~~

The last release of testtools could not be easy_installed.  This is considered
severe enough for a re-release.

Improvements
------------

* Include ``doc/`` in the source distribution, making testtools installable
  from PyPI again (Tres Seaver, #757439)


0.9.9
~~~~~

Many, many new matchers, vastly expanded documentation, stacks of bug fixes,
better unittest2 integration.  If you've ever wanted to try out testtools but
been afraid to do so, this is the release to try.


Changes
-------

* The timestamps generated by ``TestResult`` objects when no timing data has
  been received are now datetime-with-timezone, which allows them to be
  sensibly serialised and transported. (Robert Collins, #692297)

Improvements
------------

* ``AnnotatedMismatch`` now correctly returns details.
  (Jonathan Lange, #724691)

* distutils integration for the testtools test runner. Can now use it for
  'python setup.py test'. (Christian Kampka, #693773)

* ``EndsWith`` and ``KeysEqual`` now in testtools.matchers.__all__.
  (Jonathan Lange, #692158)

* ``MatchesException`` extended to support a regular expression check against
  the str() of a raised exception.  (Jonathan Lange)

* ``MultiTestResult`` now forwards the ``time`` API. (Robert Collins, #692294)

* ``MultiTestResult`` now documented in the manual. (Jonathan Lange, #661116)

* New content helpers ``content_from_file``, ``content_from_stream`` and
  ``attach_file`` make it easier to attach file-like objects to a
  test. (Jonathan Lange, Robert Collins, #694126)

* New ``ExpectedException`` context manager to help write tests against things
  that are expected to raise exceptions. (Aaron Bentley)

* New matchers:

  * ``MatchesListwise`` matches an iterable of matchers against an iterable
    of values. (Michael Hudson-Doyle)

  * ``MatchesRegex`` matches a string against a regular expression.
    (Michael Hudson-Doyle)

  * ``MatchesStructure`` matches attributes of an object against given
    matchers.  (Michael Hudson-Doyle)

  * ``AfterPreproccessing`` matches values against a matcher after passing them
    through a callable.  (Michael Hudson-Doyle)

  * ``MatchesSetwise`` matches an iterable of matchers against an iterable of
    values, without regard to order.  (Michael Hudson-Doyle)

* ``setup.py`` can now build a snapshot when Bazaar is installed but the tree
  is not a Bazaar tree. (Jelmer Vernooij)

* Support for running tests using distutils (Christian Kampka, #726539)

* Vastly improved and extended documentation. (Jonathan Lange)

* Use unittest2 exception classes if available. (Jelmer Vernooij)


0.9.8
~~~~~

In this release we bring some very interesting improvements:

* new matchers for exceptions, sets, lists, dicts and more.

* experimental (works but the contract isn't supported) twisted reactor
  support.

* The built in runner can now list tests and filter tests (the -l and
  --load-list options).

Changes
-------

* addUnexpectedSuccess is translated to addFailure for test results that don't
  know about addUnexpectedSuccess.  Further, it fails the entire result for
  all testtools TestResults (i.e. wasSuccessful() returns False after
  addUnexpectedSuccess has been called). Note that when using a delegating
  result such as ThreadsafeForwardingResult, MultiTestResult or
  ExtendedToOriginalDecorator then the behaviour of addUnexpectedSuccess is
  determined by the delegated to result(s).
  (Jonathan Lange, Robert Collins, #654474, #683332)

* startTestRun will reset any errors on the result.  That is, wasSuccessful()
  will always return True immediately after startTestRun() is called. This
  only applies to delegated test results (ThreadsafeForwardingResult,
  MultiTestResult and ExtendedToOriginalDecorator) if the delegated to result
  is a testtools test result - we cannot reliably reset the state of unknown
  test result class instances. (Jonathan Lange, Robert Collins, #683332)

* Responsibility for running test cleanups has been moved to ``RunTest``.
  This change does not affect public APIs and can be safely ignored by test
  authors.  (Jonathan Lange, #662647)

Improvements
------------

* New matchers:

  * ``EndsWith`` which complements the existing ``StartsWith`` matcher.
    (Jonathan Lange, #669165)

  * ``MatchesException`` matches an exception class and parameters. (Robert
    Collins)

  * ``KeysEqual`` matches a dictionary with particular keys.  (Jonathan Lange)

* ``assertIsInstance`` supports a custom error message to be supplied, which
  is necessary when using ``assertDictEqual`` on Python 2.7 with a
  ``testtools.TestCase`` base class. (Jelmer Vernooij)

* Experimental support for running tests that return Deferreds.
  (Jonathan Lange, Martin [gz])

* Provide a per-test decorator, run_test_with, to specify which RunTest
  object to use for a given test.  (Jonathan Lange, #657780)

* Fix the runTest parameter of TestCase to actually work, rather than raising
  a TypeError.  (Jonathan Lange, #657760)

* Non-release snapshots of testtools will now work with buildout.
  (Jonathan Lange, #613734)

* Malformed SyntaxErrors no longer blow up the test suite.  (Martin [gz])

* ``MismatchesAll.describe`` no longer appends a trailing newline.
  (Michael Hudson-Doyle, #686790)

* New helpers for conditionally importing modules, ``try_import`` and
  ``try_imports``.  (Jonathan Lange)

* ``Raises`` added to the ``testtools.matchers`` module - matches if the
  supplied callable raises, and delegates to an optional matcher for validation
  of the exception. (Robert Collins)

* ``raises`` added to the ``testtools.matchers`` module - matches if the
  supplied callable raises and delegates to ``MatchesException`` to validate
  the exception. (Jonathan Lange)

* Tests will now pass on Python 2.6.4 : an ``Exception`` change made only in
  2.6.4 and reverted in Python 2.6.5 was causing test failures on that version.
  (Martin [gz], #689858).

* ``testtools.TestCase.useFixture`` has been added to glue with fixtures nicely.
  (Robert Collins)

* ``testtools.run`` now supports ``-l`` to list tests rather than executing
  them. This is useful for integration with external test analysis/processing
  tools like subunit and testrepository. (Robert Collins)

* ``testtools.run`` now supports ``--load-list``, which takes a file containing
  test ids, one per line, and intersects those ids with the tests found. This
  allows fine grained control of what tests are run even when the tests cannot
  be named as objects to import (e.g. due to test parameterisation via
  testscenarios). (Robert Collins)

* Update documentation to say how to use testtools.run() on Python 2.4.
  (Jonathan Lange, #501174)

* ``text_content`` conveniently converts a Python string to a Content object.
  (Jonathan Lange, James Westby)



0.9.7
~~~~~

Lots of little cleanups in this release; many small improvements to make your
testing life more pleasant.

Improvements
------------

* Cleanups can raise ``testtools.MultipleExceptions`` if they have multiple
  exceptions to report. For instance, a cleanup which is itself responsible for
  running several different internal cleanup routines might use this.

* Code duplication between assertEqual and the matcher Equals has been removed.

* In normal circumstances, a TestCase will no longer share details with clones
  of itself. (Andrew Bennetts, bug #637725)

* Less exception object cycles are generated (reduces peak memory use between
  garbage collection). (Martin [gz])

* New matchers 'DoesNotStartWith' and 'StartsWith' contributed by Canonical
  from the Launchpad project. Written by James Westby.

* Timestamps as produced by subunit protocol clients are now forwarded in the
  ThreadsafeForwardingResult so correct test durations can be reported.
  (Martin [gz], Robert Collins, #625594)

* With unittest from Python 2.7 skipped tests will now show only the reason
  rather than a serialisation of all details. (Martin [gz], #625583)

* The testtools release process is now a little better documented and a little
  smoother.  (Jonathan Lange, #623483, #623487)


0.9.6
~~~~~

Nothing major in this release, just enough small bits and pieces to make it
useful enough to upgrade to.

In particular, a serious bug in assertThat() has been fixed, it's easier to
write Matchers, there's a TestCase.patch() method for those inevitable monkey
patches and TestCase.assertEqual gives slightly nicer errors.

Improvements
------------

* 'TestCase.assertEqual' now formats errors a little more nicely, in the
  style of bzrlib.

* Added `PlaceHolder` and `ErrorHolder`, TestCase-like objects that can be
  used to add results to a `TestResult`.

* 'Mismatch' now takes optional description and details parameters, so
  custom Matchers aren't compelled to make their own subclass.

* jml added a built-in UTF8_TEXT ContentType to make it slightly easier to
  add details to test results. See bug #520044.

* Fix a bug in our built-in matchers where assertThat would blow up if any
  of them failed. All built-in mismatch objects now provide get_details().

* New 'Is' matcher, which lets you assert that a thing is identical to
  another thing.

* New 'LessThan' matcher which lets you assert that a thing is less than
  another thing.

* TestCase now has a 'patch()' method to make it easier to monkey-patching
  objects in tests. See the manual for more information. Fixes bug #310770.

* MultiTestResult methods now pass back return values from the results it
  forwards to.

0.9.5
~~~~~

This release fixes some obscure traceback formatting issues that probably
weren't affecting you but were certainly breaking our own test suite.

Changes
-------

* Jamu Kakar has updated classes in testtools.matchers and testtools.runtest
  to be new-style classes, fixing bug #611273.

Improvements
------------

* Martin[gz] fixed traceback handling to handle cases where extract_tb returns
  a source line of None. Fixes bug #611307.

* Martin[gz] fixed an unicode issue that was causing the tests to fail,
  closing bug #604187.

* testtools now handles string exceptions (although why would you want to use
  them?) and formats their tracebacks correctly. Thanks to Martin[gz] for
  fixing bug #592262.

0.9.4
~~~~~

This release overhauls the traceback formatting layer to deal with Python 2
line numbers and traceback objects often being local user encoded strings
rather than unicode objects. Test discovery has also been added and Python 3.1
is also supported. Finally, the Mismatch protocol has been extended to let
Matchers collaborate with tests in supplying detailed data about failures.

Changes
-------

* testtools.utils has been renamed to testtools.compat. Importing
  testtools.utils will now generate a deprecation warning.

Improvements
------------

* Add machinery for Python 2 to create unicode tracebacks like those used by
  Python 3. This means testtools no longer throws on encountering non-ascii
  filenames, source lines, or exception strings when displaying test results.
  Largely contributed by Martin[gz] with some tweaks from Robert Collins.

* James Westby has supplied test discovery support using the Python 2.7
  TestRunner in testtools.run. This requires the 'discover' module. This
  closes bug #250764.

* Python 3.1 is now supported, thanks to Martin[gz] for a partial patch.
  This fixes bug #592375.

* TestCase.addCleanup has had its docstring corrected about when cleanups run.

* TestCase.skip is now deprecated in favour of TestCase.skipTest, which is the
  Python2.7 spelling for skip. This closes bug #560436.

* Tests work on IronPython patch from Martin[gz] applied.

* Thanks to a patch from James Westby testtools.matchers.Mismatch can now
  supply a get_details method, which assertThat will query to provide
  additional attachments. This can be used to provide additional detail
  about the mismatch that doesn't suite being included in describe(). For
  instance, if the match process was complex, a log of the process could be
  included, permitting debugging.

* testtools.testresults.real._StringException will now answer __str__ if its
  value is unicode by encoding with UTF8, and vice versa to answer __unicode__.
  This permits subunit decoded exceptions to contain unicode and still format
  correctly.

0.9.3
~~~~~

More matchers, Python 2.4 support, faster test cloning by switching to copy
rather than deepcopy and better output when exceptions occur in cleanups are
the defining characteristics of this release.

Improvements
------------

* New matcher "Annotate" that adds a simple string message to another matcher,
  much like the option 'message' parameter to standard library assertFoo
  methods.

* New matchers "Not" and "MatchesAll". "Not" will invert another matcher, and
  "MatchesAll" that needs a successful match for all of its arguments.

* On Python 2.4, where types.FunctionType cannot be deepcopied, testtools will
  now monkeypatch copy._deepcopy_dispatch using the same trivial patch that
  added such support to Python 2.5. The monkey patch is triggered by the
  absence of FunctionType from the dispatch dict rather than a version check.
  Bug #498030.

* On windows the test 'test_now_datetime_now' should now work reliably.

* TestCase.getUniqueInteger and TestCase.getUniqueString now have docstrings.

* TestCase.getUniqueString now takes an optional prefix parameter, so you can
  now use it in circumstances that forbid strings with '.'s, and such like.

* testtools.testcase.clone_test_with_new_id now uses copy.copy, rather than
  copy.deepcopy. Tests that need a deeper copy should use the copy protocol to
  control how they are copied. Bug #498869.

* The backtrace test result output tests should now pass on windows and other
  systems where os.sep is not '/'.

* When a cleanUp or tearDown exception occurs, it is now accumulated as a new
  traceback in the test details, rather than as a separate call to addError / 
  addException. This makes testtools work better with most TestResult objects
  and fixes bug #335816.


0.9.2
~~~~~

Python 3 support, more matchers and better consistency with Python 2.7 --
you'd think that would be enough for a point release. Well, we here on the
testtools project think that you deserve more.

We've added a hook so that user code can be called just-in-time whenever there
is an exception, and we've also factored out the "run" logic of test cases so
that new outcomes can be added without fiddling with the actual flow of logic.

It might sound like small potatoes, but it's changes like these that will
bring about the end of test frameworks.


Improvements
------------

* A failure in setUp and tearDown now report as failures not as errors.

* Cleanups now run after tearDown to be consistent with Python 2.7's cleanup
  feature.

* ExtendedToOriginalDecorator now passes unrecognised attributes through
  to the decorated result object, permitting other extensions to the
  TestCase -> TestResult protocol to work.

* It is now possible to trigger code just-in-time after an exception causes
  a test outcome such as failure or skip. See the testtools MANUAL or
  ``pydoc testtools.TestCase.addOnException``. (bug #469092)

* New matcher Equals which performs a simple equality test.

* New matcher MatchesAny which looks for a match of any of its arguments.

* TestCase no longer breaks if a TestSkipped exception is raised with no
  parameters.

* TestCase.run now clones test cases before they are run and runs the clone.
  This reduces memory footprint in large test runs - state accumulated on
  test objects during their setup and execution gets freed when test case
  has finished running unless the TestResult object keeps a reference.
  NOTE: As test cloning uses deepcopy, this can potentially interfere if
  a test suite has shared state (such as the testscenarios or testresources
  projects use).  Use the __deepcopy__ hook to control the copying of such
  objects so that the shared references stay shared.

* Testtools now accepts contributions without copyright assignment under some
  circumstances. See HACKING for details.

* Testtools now provides a convenient way to run a test suite using the
  testtools result object: python -m testtools.run testspec [testspec...].

* Testtools now works on Python 3, thanks to Benjamin Peterson.

* Test execution now uses a separate class, testtools.RunTest to run single
  tests. This can be customised and extended in a more consistent fashion than
  the previous run method idiom. See pydoc for more information.

* The test doubles that testtools itself uses are now available as part of
  the testtools API in testtols.testresult.doubles.

* TracebackContent now sets utf8 as the charset encoding, rather than not
  setting one and encoding with the default encoder.

* With python2.7 testtools.TestSkipped will be the unittest.case.SkipTest
  exception class making skips compatible with code that manually raises the
  standard library exception. (bug #490109)

Changes
-------

* TestCase.getUniqueInteger is now implemented using itertools.count. Thanks
  to Benjamin Peterson for the patch. (bug #490111)


0.9.1
~~~~~

The new matcher API introduced in 0.9.0 had a small flaw where the matchee
would be evaluated twice to get a description of the mismatch. This could lead
to bugs if the act of matching caused side effects to occur in the matchee.
Since having such side effects isn't desirable, we have changed the API now
before it has become widespread.

Changes
-------

* Matcher API changed to avoid evaluating matchee twice. Please consult
  the API documentation.

* TestCase.getUniqueString now uses the test id, not the test method name,
  which works nicer with parameterised tests.

Improvements
------------

* Python2.4 is now supported again.


0.9.0
~~~~~

This release of testtools is perhaps the most interesting and exciting one
it's ever had. We've continued in bringing together the best practices of unit
testing from across a raft of different Python projects, but we've also
extended our mission to incorporating unit testing concepts from other
languages and from our own research, led by Robert Collins.

We now support skipping and expected failures. We'll make sure that you
up-call setUp and tearDown, avoiding unexpected testing weirdnesses. We're
now compatible with Python 2.5, 2.6 and 2.7 unittest library.

All in all, if you are serious about unit testing and want to get the best
thinking from the whole Python community, you should get this release.

Improvements
------------

* A new TestResult API has been added for attaching details to test outcomes.
  This API is currently experimental, but is being prepared with the intent
  of becoming an upstream Python API. For more details see pydoc
  testtools.TestResult and the TestCase addDetail / getDetails methods.

* assertThat has been added to TestCase. This new assertion supports
  a hamcrest-inspired matching protocol. See pydoc testtools.Matcher for
  details about writing matchers, and testtools.matchers for the included
  matchers. See http://code.google.com/p/hamcrest/.

* Compatible with Python 2.6 and Python 2.7

* Failing to upcall in setUp or tearDown will now cause a test failure.
  While the base methods do nothing, failing to upcall is usually a problem
  in deeper hierarchies, and checking that the root method is called is a
  simple way to catch this common bug.

* New TestResult decorator ExtendedToOriginalDecorator which handles
  downgrading extended API calls like addSkip to older result objects that
  do not support them. This is used internally to make testtools simpler but
  can also be used to simplify other code built on or for use with testtools.

* New TextTestResult supporting the extended APIs that testtools provides.

* Nose will no longer find 'runTest' tests in classes derived from
   testtools.testcase.TestCase (bug #312257).

* Supports the Python 2.7/3.1 addUnexpectedSuccess and addExpectedFailure
  TestResult methods, with a support function 'knownFailure' to let tests
  trigger these outcomes.

* When using the skip feature with TestResult objects that do not support it
  a test success will now be reported. Previously an error was reported but
  production experience has shown that this is too disruptive for projects that
  are using skips: they cannot get a clean run on down-level result objects.


.. _testtools: http://pypi.python.org/pypi/testtools