############################ testrepository release notes ############################ NEXT (In development) +++++++++++++++++++++ CHANGES ------- * setup.py testr now supports different coverage names. (Marc Abramowitz, #1298398) * Isolate the testrepository test suite from the chunking (or otherwise) behaviour of testtools' exception handlers. (Robert Collins) * pbr is now used to manage build and release tooling. (Monty Taylor, Robert Collins) * Unexpected errors reading streams from a file repository now propogate the correct exception. (Matthew Treinish, Robert Collins, #1409936) * testrepository now requires at least version 0.0.11 of the subunit Python module. (Jelmer Vernooij) 0.0.20 ++++++ IMPROVEMENTS ------------ * Tests will be reliably tagged with worker-%d. The previous tagging logic had an implicit race - the tag id was looked up via a closure which gets the state of the pos variable at the position the overall loop has advanced too, not the position when the closure was created. (Robert Collins, #1316858) 0.0.19 ++++++ CHANGES ------- * Passing --subunit to all testr commands will now consistently output subunit v2. Previously it would output v1 for stored streams and v2 for live streams. (Robert Collins) * ``run`` was outputting bad MIME types - test/plain, not text/plain. (Robert Collins) * Test filtering was failing under python3 and would only apply the filters to the first test listed by discover. (Clark Boylan, #1317607) * Tests that are enumerated but not executed will no longer reset the test timing data. Enumeration was incorrectly recording a 0 timestamp for enumerated tests. This leads to poor scheduling after an interrupted test run. (Robert Collins, #1322763) * Version 0.0.18 of subunit is now a hard dependency - the v2 protocol solves key issues in concurrency and stream handling. Users that cannot use subunit v2 can run an older testrepository, or contact upstream to work through whatever issue is blocking them. (Robert Collins) * When list-tests encounters an error, a much clearer response will now be shown. (Robert Collins, #1271133) INTERNALS --------- * The ``get_subunit_stream`` methods now return subunit v2 streams rather than v1 streams, preparing the way for storage of native v2 streams in the repository. (Robert Collins) * ``UI.output_stream`` is now tested for handling of non-utf8 bytestreams. (Robert Collins) 0.0.18 ++++++ CHANGES ------- * ``run`` now accepts ``--isolated`` as a parameter, which will cause each selected test to be run independently. This can be useful to both workaround isolation bugs and detect tests that can not be run independently. (Robert Collins) INTERNALS --------- * ``capture_ids`` in test_run now returns a list of captures, permitting tests that need to test multiple runs to do so. (Robert Collins) 0.0.17 ++++++ CHANGES ------- * Restore the ability to import testrepository.repository.memory on Python 2.6. (Robert Collins) 0.0.16 ++++++ CHANGES ------- * A new testr.conf option ``group_regex`` can be used for grouping tests so that they get run in the same backend runner. (Matthew Treinish) * Fix Python 3.* support for entrypoints; the initial code was Python3 incompatible. (Robert Collins, Clark Boylan, #1187192) * Switch to using multiprocessing to determine CPU counts. (Chris Jones, #1092276) * The cli UI now has primitive differentiation between multiple stream types. This is not yet exposed to the end user, but is sufficient to enable the load command to take interactive input without it reading from the raw subunit stream on stdin. (Robert Collins) * The scheduler can now groups tests together permitting co-dependent tests to always be scheduled onto the same backend. Note that this does not force co-dependent tests to be executed, so partial test runs (e.g. --failing) may still fail. (Matthew Treinish, Robert Collins) * When test listing fails, testr will now report an error rather than incorrectly trying to run zero tests. A test listing failure is detected by the returncode of the test listing process. (Robert Collins, #1185231) 0.0.15 ++++++ CHANGES ------- * Expects subunit v2 if the local library has v2 support in the subunit library. This should be seamless if the system under test shares the Python libraries. If it doesn't, either arrange to use ``subunit-2to1`` or upgrade the subunit libraries for the system under test. (Robert Collins) * ``--full-results`` is now a no-op, use ``--subunit`` to get unfiltered output. (Robert Collins) 0.0.14 ++++++ IMPROVEMENTS ------------ * First cut at full Python 3 support. The 'works for me' release. (Robert Collins) 0.0.13 ++++++ IMPROVEMENTS ------------ * ``setup.py testr`` was not indicating test failures via it's return code. (Monty Taylor) 0.0.12 ++++++ IMPROVEMENTS ------------ * There is now a setuptools extension provided by ``testrespository`` making it easy to invoke testr from setup.py driven workflows. (Monty Taylor, Robert Collins) INTERNALS --------- * BSD license file incorrectly claimed copyright by subunit contributors. (Monty Taylor) * .testr.conf is now shipped in the source distribution to aid folk wanting to validate that testrepository works correctly on their machine. (Robert Collins) 0.0.11 ++++++ IMPROVEMENTS ------------ * Fix another incompatability with Mac OS X - gdbm dbm modules don't support get. (Robert Collins, #1094330) 0.0.10 ++++++ IMPROVEMENTS ------------ * It's now possible to configure ``test_run_concurrency`` in ``.testr.conf`` to have concurrency defined by a callout. (Robert Collins) * Testr supports running tests in arbitrary environments. See ``Remote or isolated test environments`` in MANUAL.txt / ``testr help run`` (Robert Collins) INTERNALS --------- * TestCommand is now a fixture. This is used to ensure cached test instances are disposed of - if using the object to run or list tests, you will need to adjust your calls. (Robert Collins) * ``TestCommand`` now offers, and ``TestListingFixture`` consumes a small protocol for obtaining and releasing test execution instances. (Robert Collins) 0.0.9 +++++ IMPROVEMENTS ------------ * On OSX the ``anydbm`` module by default returns an implementation that doesn't support update(). Workaround that by falling back to a loop. (Robert Collins, #1091500) * ``testr --analyze-improvements`` now honours test regex filters and only analyzes matching tests. (Robert Collins) 0.0.8 +++++ CHANGES ------- * As a side effect of fixing bug #597060 additional arguments passed to testr run or testr list are only passed to the underlying test runner if they are preceeded by '--'. (Robert Collins, #597060) * ``testr run --failing`` will no longer run any tests at all if there are no failing tests. (Robert Collins, #904400) IMPROVEMENTS ------------ * ``AbstractArgument`` now handles the case where additional arguments are present that the argument type cannot parse, but enough have been parsed for it to be valid. This allows optional arguments to be in the middle of a grammar. (Robert Collins) * ``cli.UI`` now passed '--' down to the argument layer for handling rather than implicitly stripping it. (Robert Collins) * ``DoubledashArgument`` added to allow fine grained control over the impact of -- in command lines. (Robert Collins) * New argument type ``ExistingPathArgument`` for use when commands want to take the name of a file. (Robert Collins) * ``testr`` will now show the version. (Robert Collins) * ``testr last`` when just one test run has been run works again. (Robert Collins) * ``testr help command`` now shows the docstring for commands (Robert Collins) * ``testr --help command`` or ``testr command --help`` now shows the options for the command. (Robert Collins) * ``testr run --analyze-isolation`` will search the current failing tests for spurious failures caused by interactions with other tests. (Robert Collins, #684069) * ``testr run --until-failure`` will repeat a test run until interrupted by ctrl-C or until a failure occurs. (Robert Collins, #680995) * ``Repository.get_test_run`` now raises KeyError if asked for a missing or nonexistant test run. (Robert Collins) * Sphinx has been added to tie the documentation toghether (And it is available on testrepository.readthedocs.org). (Robert Collins) * ``StringArgument`` now rejects '--' - it should be handled by the use of a ``DoubledashArgument`` where one is expected. This is a bit awkward and does not permit passing '--' down to a child process, so further work may be needed - file a bug if this affects you. (Robert Collins) * ``test failing --subunit`` now exits 0 unless there was a problem generating the stream. This is consistent with the general processing model of subunit generators. (Robert Collins) * ``testr last`` now supports ``--subunit`` and when passed will output the stored subunit stream. Note that the exit code is always 0 when this is done (unless an exception occurs reading the stream) - subunit consumers should parse the subunit to determine success/failure. (Robert Collins) * ``testr load`` now supports passing filenames to subunit streams to load. (Robert Collins, #620386) * ``testr run`` will now fail a test run if the test process exits non-zero. As a side effect of this change, if the test program closes its stdout but does not exit, ``testr run`` will hang (waiting for the test program to exit). (Robert Collins) * ``testr run --load-list FILENAME`` will limit the tests run to the test ids supplied in the list file FILENAME. This is useful for manually specifying the tests to run, or running testr subordinate to testr (e.g. on remote machines). (Robert Collins, partial fix for #597060) * ``testr run foo`` now applies foo as a regex filter against the tests found by doing a listing of the test runners tests. Likewise ``testr list-tests foo`` will apply foo as a filter against the found tests. This makes it easy to limit the tests that will be requested for running by the backend test process - simply pass one or more regex filters into testr run. (Robert Collins, #597060) * Test tags are now shown in failures. Of particular interest for folk debgging cross-test interactions will be the worker-N tags which indicate which backend test process executed a given test. (Robert Collins) 0.0.7 +++++ CHANGES ------- * testrepository is now distributed via distribute rather than distutils, allowing installation via pip into virtualenv environments. (Robert Collins) IMPROVEMENTS ------------ * stream loading will now synthesise datestamps before demultiplexing rather than on insertion into the repository. This fixes erroneously short times being recorded on non timestamped streams. Additionally, moving the automatic addition of timestamp material in front of the demuxer has removed the skew that caused test times to be reported as longer than the stream could indicate (by the amount of time the test runner took to start outputting subunit). This time may be something we want to track later, but the prior mechanism was inconsistent between the current run and reporting on prior runs, which lead to a very confusing UI. Now it is consistent, but totally ignores that overhead. (Robert Collins, #1048126, #980950) * ``testr run`` now accepts a --concurrency option, allowing command line override of the number of workers spawned. This allows conccurency on operating systems where autodetection is not yet implemented, or just debugging problems with concurrent test suites. (Robert Collins, #957145) * ''test_id_list_default'' would prevent ''test_list_option'' being used in previous releases. For Python environments where the context to load tests from is always needed this was not an issue (and thus not uncovered). However given a test runner which wants either a global context or a list of specific tests with no global context, there was no way to achieve that with this bug. (Robert Collins, #1027042) 0.0.6 +++++ CHANGES ------- * Now relies on subunit 0.0.8 or better and testtools 0.9.15 or better. IMPROVEMENTS ------------ * Much better handling of unicode input from subunit streams. Specifically, we won't crash when we can't figure something out. (Francesco Banconi, Martin Packman, #955006) * Parallel tests now record their worker thread number as tags in tests. This makes identifying test ordering problems much easier. (Benji York, #974622) * Python2.7 changed the interface for DBM, this has been worked around. (Robert Collins, #775214, #961103) * Subunit 0.0.7 Changes its TestResultFilter implementation, requiring the subclass in testrepository.filter to be come more robust. (Robert Collins) * A horrible thinko in the testrepository test suite came to light and has been fixed. How the tests ever ran is a mystery. (Robert Collins, #881497) * ''failing'', ''run'' and ''load'' now both take a ''--subunit'' option, which displays output in raw subunit format. If ''--full-results'' is passed too, then all subunit information is displayed. (Brad Crittenden, #949950) * Setting ''filter_tags'' in ''.testr.conf'' will cause tests tagged with those tags to be hidden unless the fail/error. This requires Subunit 0.0.8. If an older version of subunit is configured, testr will return an error. (Robert Collins, #914166) * ``testr`` will drop into PDB from its command line UI if the environment variable TESTR_PDB is set. (Robert Collins) * Test partitioning now handles a corner case where multiple tests have a reported duration of 0. Previously they could all accumulate into one partition, now they split across partitions (the length of a partition is used as a tie breaker if two partitions have the same duration). (Robert Collins, #914359) * The test 'test_outputs_results_to_stdout' was sensitive to changes in testtools and has been made more generic. (Robert Collins) 0.0.5 +++++ CHANGES ------- * The testrepository test suite depends on testtools 0.9.8. (Robert Collins) * If interrupted while updating the ``failing`` list, temp files are now cleaned up - previously a carefully timed interrupt would leave the temporary failing file in place. (Robert Collins, #531665) * Local implementation of MatchesException has been removed in favour of the testtools implementation. All ``self.assertRaises`` have been migrated to this new testing interface. * ``setup.py`` will read the version number from PKG-INFO when it is running without a bzr tree : this makes it easier to snapshot without doing a release. (Jonathan Lange) * Testrepository should be more compatible with win32 environments. (Martin [gz]) * ``testr init-repo`` now has a ``--force-init`` option which when provided will cause a repository to be created just-in-time. (Jonathan Lange) * ``testr load`` and ``testr run`` now have a flag ``--partial``. When set this will cause existing failures to be preserved. When not set, doing a load will reset existing failures. The ``testr run`` flag ``--failing`` implicitly sets ``--partial`` (so that an interrupted incremental test run does not incorrectly discard a failure record). The ``--partial`` flag exists so that deleted or renamed tests do not persist forever in the database. (Robert Collins) * ``testr load`` now loads all input streams in parallel. This has no impact on the CLI as yet, but permits API users to load from parallel processes. (Robert Collins) * ``testr list-tests`` is a new command that will list the tests for a project when ``.testr.conf`` has been configured with a ``test_list_option``. (Robert Collins) * ``test run --parallel`` will parallelise by running each test command once per CPU in the machine (detection for this only implemented on Linux so far). An internally parallelising command will not benefit from this, but for many projects it will be a win either from simplicity or because getting their test runner to parallise is nontrivial. The observed duration of tests is used to inform the partitioning algorithm - so each test runner should complete at approximately the same time, minimising total runtime. (Robert Collins) * ``testr run`` no longer attempts to expand unknown variables. This permits the use of environmen variables to control the test run. For instance, ${PYTHON:-python} in the test_command setting will run the command with $PYTHON or python if $PYTHON is not set. (Robert Collins, #595295) * ``testr run`` now resets the SIGPIPE handler to default - which is what most Unix processes expect. (Robert Collins) * ``testr run`` now uses a unique file name rather than hard coding failing.list - while not as clear, this permits concurrent testr invocations, or parallel testing from within testr, to execute safely. (Robert Collins) * ``testr run`` uses an in-process load rather than reinvoking testr. This should be faster on Windows and avoids the issue with running the wrong testr when PYTHONPATH but not PATH is set. (Robert Collins, #613129) * ``testr run`` will now pass -d to the ``testr load`` invocation, so that running ``testr run -d /some/path`` will work correctly. (Robert Collins, #529698) * ``testr run`` will now pass ``-q`` down to ``testr load``. (Robert Collins, #529701) * The ``testrepository.repository.Repository`` interface now tracks test times for use in estimating test run duration and parallel test partitioning. (Robert Collins) * There are the beginnings of a samba buildfarm backend for testrepository, though it is not hooked into the UI yet, so is only useful for API users. (Jelmer Vernooij) * Updates to next-stream are done via a temporary file to reduce the chance of an empty next-stream being written to disk. (Robert Collins, #531664) * Variable expansion no longer does python \ escape expansion. (Robert Collins, #694800) * When next-stream is damaged testr will report that it is corrupt rather than reporting an invalid literal. (Robert Collins, #531663) 0.0.4 +++++ IMPROVEMENTS ------------ * ``failing`` now supports ``--list`` to list the failing tests. (Jonathan Lange) * Repository not found errors are now clearer. (Jonathan Lange, #530010) * The summary of a test run is now formatted as foo=NN rather than foo: NN, which some folk find easier to read. * The file implementation of Repository.open now performs ~ expansion. (Jonathan Lange, #529665) * Test failures and errors are now shown as we get them in 'load', 'failing' and 'last'. (Jonathan Lange, #613152) 0.0.3 +++++ IMPROVEMENTS ------------ * ``failing`` now correctly calls ``repository.get_failing`` and will this track all seen failures rather than just the latest observed failures. * New argument type ``StringArgument`` for use when a supplied argument is just a string, rather than a typed argument. * New subcommand 'failing' added. * New subcommand ``run`` added which reads a .testr.conf file to figure out how to run tests with subunit output. It then runs them and pipes into testr load. This allows simpler integration and permits a programming interface so that tools like Tribunal/Eclipe etc can refresh tests in a testrepository. ``run`` also passes arguments and options down to the child process. ``run`` can also supply test ids on the command, for test runners that want that. * The command 'last' will no longer error on a new repository.