[TIP] py.test 1.3.2: API and reporting refinements & fixes
holger at merlinux.eu
Thu Jul 8 08:31:35 PDT 2010
The pylib/py.test 1.3.2 release brings many bug fixes and some new
features. It was refined for and tested against the recently released
Python2.7 and remains compatibile to the usual armada of interpreters
(Python2.4 through to Python3.1.2, Jython and PyPy). Note that for using
distributed testing features you'll need to upgrade to the jointly released
pytest-xdist-1.4 because of some internal refactorings.
See http://pytest.org for general documentation and below for
a inlined detailed CHANGELOG.
cheers & particular thanks to Benjamin Peterson, Ronny Pfannschmidt
and all issue and patch contributors,
Changes between 1.3.1 and 1.3.2
- fix issue103: introduce py.test.raises as context manager, examples::
x = 0
1 / x
with py.test.raises(RuntimeError) as excinfo:
# you may do extra checks on excinfo.value|type|traceback here
(thanks Ronny Pfannschmidt)
- Funcarg factories can now dynamically apply a marker to a
test invocation. This is for example useful if a factory
provides parameters to a test which are expected-to-fail::
- improved error reporting on collection and import errors. This makes
use of a more general mechanism, namely that for custom test item/collect
nodes ``node.repr_failure(excinfo)`` is now uniformly called so that you can
override it to return a string error representation of your choice
which is going to be reported as a (red) string.
- introduce '--junitprefix=STR' option to prepend a prefix
to all reports in the junitxml file.
Bug fixes / Maintenance
- make tests and the ``pytest_recwarn`` plugin in particular fully compatible
to Python2.7 (if you use the ``recwarn`` funcarg warnings will be enabled so that
you can properly check for their existence in a cross-python manner).
- refine --pdb: ignore xfailed tests, unify its TB-reporting and
don't display failures again at the end.
- fix assertion interpretation with the ** operator (thanks Benjamin Peterson)
- fix issue105 assignment on the same line as a failing assertion (thanks Benjamin Peterson)
- fix issue104 proper escaping for test names in junitxml plugin (thanks anonymous)
- fix issue57 -f|--looponfail to work with xpassing tests (thanks Ronny)
- fix issue92 collectonly reporter and --pastebin (thanks Benjamin Peterson)
- fix py.code.compile(source) to generate unique filenames
- fix assertion re-interp problems on PyPy, by defering code
compilation to the (overridable) Frame.eval class. (thanks Amaury Forgeot)
- fix py.path.local.pyimport() to work with directories
- streamline py.path.local.mkdtemp implementation and usage
- don't print empty lines when showing junitxml-filename
- add optional boolean ignore_errors parameter to py.path.local.remove
- fix terminal writing on win32/python2.4
- py.process.cmdexec() now tries harder to return properly encoded unicode objects
on all python versions
- install plain py.test/py.which scripts also for Jython, this helps to
get canonical script paths in virtualenv situations
- make path.bestrelpath(path) return ".", note that when calling
X.bestrelpath the assumption is that X is a directory.
- make initial conftest discovery ignore "--" prefixed arguments
- fix resultlog plugin when used in an multicpu/multihost xdist situation
(thanks Jakub Gustak)
- perform distributed testing related reporting in the xdist-plugin
rather than having dist-related code in the generic py.test
- fix homedir detection on Windows
- ship distribute_setup.py version 0.6.13
Changes between 1.3.0 and 1.3.1
- issue91: introduce new py.test.xfail(reason) helper
to imperatively mark a test as expected to fail. Can
be used from within setup and test functions. This is
useful especially for parametrized tests when certain
configurations are expected-to-fail. In this case the
declarative approach with the @py.test.mark.xfail cannot
be used as it would mark all configurations as xfail.
- issue102: introduce new --maxfail=NUM option to stop
test runs after NUM failures. This is a generalization
of the '-x' or '--exitfirst' option which is now equivalent
to '--maxfail=1'. Both '-x' and '--maxfail' will
now also print a line near the end indicating the Interruption.
- issue89: allow py.test.mark decorators to be used on classes
(class decorators were introduced with python2.6) and
also allow to have multiple markers applied at class/module level
by specifying a list.
- improve and refine letter reporting in the progress bar:
f failed test
s skipped tests (reminder: use for dependency/platform mismatch only)
x xfailed test (test that was expected to fail)
X xpassed test (test that was expected to fail but passed)
You can use any combination of 'fsxX' with the '-r' extended
reporting option. The xfail/xpass results will show up as
skipped tests in the junitxml output - which also fixes
- make py.test.cmdline.main() return the exitstatus instead of raising
SystemExit and also allow it to be called multiple times. This of
course requires that your application and tests are properly teared
down and don't have global state.
Fixes / Maintenance
- improved traceback presentation:
- improved and unified reporting for "--tb=short" option
- Errors during test module imports are much shorter, (using --tb=short style)
- raises shows shorter more relevant tracebacks
- --fulltrace now more systematically makes traces longer / inhibits cutting
- improve support for raises and other dynamically compiled code by
manipulating python's linecache.cache instead of the previous
rather hacky way of creating custom code objects. This makes
it seemlessly work on Jython and PyPy where it previously didn't.
- fix issue96: make capturing more resilient against Control-C
interruptions (involved somewhat substantial refactoring
to the underlying capturing functionality to avoid race
- fix chaining of conditional skipif/xfail decorators - so it works now
as expected to use multiple @py.test.mark.skipif(condition) decorators,
including specific reporting which of the conditions lead to skipping.
- fix issue95: late-import zlib so that it's not required
for general py.test startup.
- fix issue94: make reporting more robust against bogus source code
(and internally be more careful when presenting unexpected byte sequences)
Changes between 1.2.1 and 1.3.0
- deprecate --report option in favour of a new shorter and easier to
remember -r option: it takes a string argument consisting of any
combination of 'xfsX' characters. They relate to the single chars
you see during the dotted progress printing and will print an extra line
per test at the end of the test run. This extra line indicates the exact
position or test ID that you directly paste to the py.test cmdline in order
to re-run a particular test.
- allow external plugins to register new hooks via the new
pytest_addhooks(pluginmanager) hook. The new release of
the pytest-xdist plugin for distributed and looponfailing
testing requires this feature.
- add a new pytest_ignore_collect(path, config) hook to allow projects and
plugins to define exclusion behaviour for their directory structure -
for example you may define in a conftest.py this method::
to prevent even a collection try of any tests in symlinked dirs.
- new pytest_pycollect_makemodule(path, parent) hook for
allowing customization of the Module collection object for a
matching test module.
- extend and refine xfail mechanism:
``@py.test.mark.xfail(run=False)`` do not run the decorated test
``@py.test.mark.xfail(reason="...")`` prints the reason string in xfail summaries
specifiying ``--runxfail`` on command line virtually ignores xfail markers
- expose (previously internal) commonly useful methods:
py.io.get_terminal_with() -> return terminal width
py.io.ansi_print(...) -> print colored/bold text on linux/win32
py.io.saferepr(obj) -> return limited representation string
- expose test outcome related exceptions as py.test.skip.Exception,
py.test.raises.Exception etc., useful mostly for plugins
doing special outcome interpretation/tweaking
- (issue85) fix junitxml plugin to handle tests with non-ascii output
- fix/refine python3 compatibility (thanks Benjamin Peterson)
- fixes for making the jython/win32 combination work, note however:
jython2.5.1/win32 does not provide a command line launcher, see
http://bugs.jython.org/issue1491 . See pylib install documentation
for how to work around.
- fixes for handling of unicode exception values and unprintable objects
- (issue87) fix unboundlocal error in assertionold code
- (issue86) improve documentation for looponfailing
- refine IO capturing: stdin-redirect pseudo-file now has a NOP close() method
- ship distribute_setup.py version 0.6.10
- added links to the new capturelog and coverage plugins
Changes between 1.2.1 and 1.2.0
- refined usage and options for "py.cleanup"::
py.cleanup # remove "*.pyc" and "*$py.class" (jython) files
py.cleanup -e .swp -e .cache # also remove files with these extensions
py.cleanup -s # remove "build" and "dist" directory next to setup.py files
py.cleanup -d # also remove empty directories
py.cleanup -a # synonym for "-s -d -e 'pip-log.txt'"
py.cleanup -n # dry run, only show what would be removed
- add a new option "py.test --funcargs" which shows available funcargs
and their help strings (docstrings on their respective factory function)
for a given test path
- display a short and concise traceback if a funcarg lookup fails
- early-load "conftest.py" files in non-dot first-level sub directories.
allows to conveniently keep and access test-related options in a ``test``
subdir and still add command line options.
- fix issue67: new super-short traceback-printing option: "--tb=line" will print a single line for each failing (python) test indicating its filename, lineno and the failure value
- fix issue78: always call python-level teardown functions even if the
according setup failed. This includes refinements for calling setup_module/class functions
which will now only be called once instead of the previous behaviour where they'd be called
multiple times if they raise an exception (including a Skipped exception). Any exception
will be re-corded and associated with all tests in the according module/class scope.
- fix issue63: assume <40 columns to be a bogus terminal width, default to 80
- fix pdb debugging to be in the correct frame on raises-related errors
- update apipkg.py to fix an issue where recursive imports might
unnecessarily break importing
- fix plugin links
Changes between 1.2 and 1.1.1
- moved dist/looponfailing from py.test core into a new
separately released pytest-xdist plugin.
- new junitxml plugin: --junitxml=path will generate a junit style xml file
which is processable e.g. by the Hudson CI system.
- new option: --genscript=path will generate a standalone py.test script
which will not need any libraries installed. thanks to Ralf Schmitt.
- new option: --ignore will prevent specified path from collection.
Can be specified multiple times.
- new option: --confcutdir=dir will make py.test only consider conftest
files that are relative to the specified dir.
- new funcarg: "pytestconfig" is the pytest config object for access
to command line args and can now be easily used in a test.
- install 'py.test' and `py.which` with a ``-$VERSION`` suffix to
disambiguate between Python3, python2.X, Jython and PyPy installed versions.
- new "pytestconfig" funcarg allows access to test config object
- new "pytest_report_header" hook can return additional lines
to be displayed at the header of a test run.
- (experimental) allow "py.test path::name1::name2::..." for pointing
to a test within a test collection directly. This might eventually
evolve as a full substitute to "-k" specifications.
- streamlined plugin loading: order is now as documented in
customize.html: setuptools, ENV, commandline, conftest.
also setuptools entry point names are turned to canonical namees ("pytest_*")
- automatically skip tests that need 'capfd' but have no os.dup
- allow pytest_generate_tests to be defined in classes as well
- deprecate usage of 'disabled' attribute in favour of pytestmark
- deprecate definition of Directory, Module, Class and Function nodes
in conftest.py files. Use pytest collect hooks instead.
- collection/item node specific runtest/collect hooks are only called exactly
on matching conftest.py files, i.e. ones which are exactly below
the filesystem path of an item
- change: the first pytest_collect_directory hook to return something
will now prevent further hooks to be called.
- change: figleaf plugin now requires --figleaf to run. Also
change its long command line options to be a bit shorter (see py.test -h).
- change: pytest doctest plugin is now enabled by default and has a
new option --doctest-glob to set a pattern for file matches.
- change: remove internal py._* helper vars, only keep py._pydir
- robustify capturing to survive if custom pytest_runtest_setup
code failed and prevented the capturing setup code from running.
- make py.test.* helpers provided by default plugins visible early -
works transparently both for pydoc and for interactive sessions
which will regularly see e.g. py.test.mark and py.test.importorskip.
- simplify internal plugin manager machinery
- simplify internal collection tree by introducing a RootCollector node
- fix assert reinterpreation that sees a call containing "keyword=..."
- fix issue66: invoke pytest_sessionstart and pytest_sessionfinish
hooks on slaves during dist-testing, report module/session teardown
- fix issue65: properly handle dist-testing if no
execnet/py lib installed remotely.
- skip some install-tests if no execnet is available
- fix docs, fix internal bin/ script generation
Changes between 1.1.1 and 1.1.0
- introduce automatic plugin registration via 'pytest11'
entrypoints via setuptools' pkg_resources.iter_entry_points
- fix py.test dist-testing to work with execnet >= 1.0.0b4
- re-introduce py.test.cmdline.main() for better backward compatibility
- svn paths: fix a bug with path.check(versioned=True) for svn paths,
allow '%' in svn paths, make svnwc.update() default to interactive mode
like in 1.0.x and add svnwc.update(interactive=False) to inhibit interaction.
- refine distributed tarball to contain test and no pyc files
- try harder to have deprecation warnings for py.compat.* accesses
report a correct location
Changes between 1.1.0 and 1.0.2
* adjust and improve docs
* remove py.rest tool and internal namespace - it was
never really advertised and can still be used with
the old release if needed. If there is interest
it could be revived into its own tool i guess.
* fix issue48 and issue59: raise an Error if the module
from an imported test file does not seem to come from
the filepath - avoids "same-name" confusion that has
been reported repeatedly
* merged Ronny's nose-compatibility hacks: now
nose-style setup_module() and setup() functions are
* introduce generalized py.test.mark function marking
* reshuffle / refine command line grouping
* deprecate parser.addgroup in favour of getgroup which creates option group
* add --report command line option that allows to control showing of skipped/xfailed sections
* generalized skipping: a new way to mark python functions with skipif or xfail
at function, class and modules level based on platform or sys-module attributes.
* extend py.test.mark decorator to allow for positional args
* introduce and test "py.cleanup -d" to remove empty directories
* fix issue #59 - robustify unittest test collection
* make bpython/help interaction work by adding an __all__ attribute
to ApiModule, cleanup initpkg
* use MIT license for pylib, add some contributors
* remove py.execnet code and substitute all usages with 'execnet' proper
* fix issue50 - cached_setup now caches more to expectations
for test functions with multiple arguments.
* merge Jarko's fixes, issue #45 and #46
* add the ability to specify a path for py.lookup to search in
* fix a funcarg cached_setup bug probably only occuring
in distributed testing and "module" scope with teardown.
* many fixes and changes for making the code base python3 compatible,
many thanks to Benjamin Peterson for helping with this.
* consolidate builtins implementation to be compatible with >=2.3,
add helpers to ease keeping 2 and 3k compatible code
* deprecate py.compat.doctest|subprocess|textwrap|optparse
* deprecate py.magic.autopath, remove py/magic directory
* move pytest assertion handling to py/code and a pytest_assertion
plugin, add "--no-assert" option, deprecate py.magic namespaces
in favour of (less) py.code ones.
* consolidate and cleanup py/code classes and files
* cleanup py/misc, move tests to bin-for-dist
* introduce delattr/delitem/delenv methods to py.test's monkeypatch funcarg
* consolidate py.log implementation, remove old approach.
* introduce py.io.TextIO and py.io.BytesIO for distinguishing between
text/unicode and byte-streams (uses underlying standard lib io.*
* make py.unittest_convert helper script available which converts "unittest.py"
style files into the simpler assert/direct-test-classes py.test/nosetests
style. The script was written by Laura Creighton.
* simplified internal localpath implementation
Changes between 1.0.1 and 1.0.2
* fixing packaging issues, triggered by fedora redhat packaging,
also added doc, examples and contrib dirs to the tarball.
* added a documentation link to the new django plugin.
Changes between 1.0.0 and 1.0.1
* added a 'pytest_nose' plugin which handles nose.SkipTest,
nose-style function/method/generator setup/teardown and
tries to report functions correctly.
* capturing of unicode writes or encoded strings to sys.stdout/err
work better, also terminalwriting was adapted and somewhat
unified between windows and linux.
* improved documentation layout and content a lot
* added a "--help-config" option to show conftest.py / ENV-var names for
all longopt cmdline options, and some special conftest.py variables.
renamed 'conf_capture' conftest setting to 'option_capture' accordingly.
* fix issue #27: better reporting on non-collectable items given on commandline
(e.g. pyc files)
* fix issue #33: added --version flag (thanks Benjamin Peterson)
* fix issue #32: adding support for "incomplete" paths to wcpath.status()
* "Test" prefixed classes are *not* collected by default anymore if they
have an __init__ method
* monkeypatch setenv() now accepts a "prepend" parameter
* improved reporting of collection error tracebacks
* simplified multicall mechanism and plugin architecture,
renamed some internal methods and argnames
Changes between 1.0.0b9 and 1.0.0
* more terse reporting try to show filesystem path relatively to current dir
* improve xfail output a bit
Changes between 1.0.0b8 and 1.0.0b9
* cleanly handle and report final teardown of test setup
* fix svn-1.6 compat issue with py.path.svnwc().versioned()
(thanks Wouter Vanden Hove)
* setup/teardown or collection problems now show as ERRORs
or with big "E"'s in the progress lines. they are reported
and counted separately.
* dist-testing: properly handle test items that get locally
collected but cannot be collected on the remote side - often
due to platform/dependency reasons
* simplified py.test.mark API - see keyword plugin documentation
* integrate better with logging: capturing now by default captures
test functions and their immediate setup/teardown in a single stream
* capsys and capfd funcargs now have a readouterr() and a close() method
(underlyingly py.io.StdCapture/FD objects are used which grew a
readouterr() method as well to return snapshots of captured out/err)
* make assert-reinterpretation work better with comparisons not
returning bools (reported with numpy from thanks maciej fijalkowski)
* reworked per-test output capturing into the pytest_iocapture.py plugin
and thus removed capturing code from config object
* item.repr_failure(excinfo) instead of item.repr_failure(excinfo, outerr)
Changes between 1.0.0b7 and 1.0.0b8
* pytest_unittest-plugin is now enabled by default
* introduced pytest_keyboardinterrupt hook and
refined pytest_sessionfinish hooked, added tests.
* workaround a buggy logging module interaction ("closing already closed
files"). Thanks to Sridhar Ratnakumar for triggering.
* if plugins use "py.test.importorskip" for importing
a dependency only a warning will be issued instead
of exiting the testing process.
* many improvements to docs:
- refined funcargs doc , use the term "factory" instead of "provider"
- added a new talk/tutorial doc page
- better download page
- better plugin docstrings
- added new plugins page and automatic doc generation script
* fixed teardown problem related to partially failing funcarg setups
(thanks MrTopf for reporting), "pytest_runtest_teardown" is now
always invoked even if the "pytest_runtest_setup" failed.
* tweaked doctest output for docstrings in py modules,
Changes between 1.0.0b3 and 1.0.0b7
* renamed py.test.xfail back to py.test.mark.xfail to avoid
two ways to decorate for xfail
* re-added py.test.mark decorator for setting keywords on functions
(it was actually documented so removing it was not nice)
* remove scope-argument from request.addfinalizer() because
request.cached_setup has the scope arg. TOOWTDI.
* perform setup finalization before reporting failures
* apply modified patches from Andreas Kloeckner to allow
test functions to have no func_code (#22) and to make
"-k" and function keywords work (#20)
* apply patch from Daniel Peolzleithner (issue #23)
* resolve issue #18, multiprocessing.Manager() and
* make __name__ == "__channelexec__" for remote_exec code
Changes between 1.0.0b1 and 1.0.0b3
* plugin classes are removed: one now defines
hooks directly in conftest.py or global pytest_*.py
* added new pytest_namespace(config) hook that allows
to inject helpers directly to the py.test.* namespace.
* documented and refined many hooks
* added new style of generative tests via
pytest_generate_tests hook that integrates
well with function arguments.
Changes between 0.9.2 and 1.0.0b1
* introduced new "funcarg" setup method,
* introduced plugin architecuture and many
new py.test plugins, see
* teardown_method is now guaranteed to get
called after a test method has run.
* new method: py.test.importorskip(mod,minversion)
will either import or call py.test.skip()
* completely revised internal py.test architecture
* new py.process.ForkedFunc object allowing to
fork execution of a function to a sub process
and getting a result back.
XXX lots of things missing here XXX
Changes between 0.9.1 and 0.9.2
* refined installation and metadata, created new setup.py,
now based on setuptools/ez_setup (thanks to Ralf Schmitt
for his support).
* improved the way of making py.* scripts available in
windows environments, they are now added to the
Scripts directory as ".cmd" files.
* py.path.svnwc.status() now is more complete and
uses xml output from the 'svn' command if available
* fix for py.path.svn* to work with svn 1.5
* fix path.relto(otherpath) method on windows to
use normcase for checking if a path is relative.
* py.test's traceback is better parseable from editors
(follows the filenames:LINENO: MSG convention)
(thanks to Osmo Salomaa)
should work more reliably now
* removed previously accidentally added
py.test.broken and py.test.notimplemented helpers.
* there now is a py.__version__ attribute
Changes between 0.9.0 and 0.9.1
This is a fairly complete list of changes between 0.9 and 0.9.1, which can
serve as a reference for developers.
* allowing + signs in py.path.svn urls 
* fixed support for Failed exceptions without excinfo in py.test 
* added support for killing processes for Windows (as well as platforms that
support os.kill) in py.misc.killproc 
* added setup/teardown for generative tests to py.test 
* added detection of FAILED TO LOAD MODULE to py.test [40703, 40738, 40739]
* fixed problem with calling .remove() on wcpaths of non-versioned files in
* fixed some import and inheritance issues in py.test [41480, 44648, 44655]
* fail to run greenlet tests when pypy is available, but without stackless
* small fixes in rsession tests 
* fixed issue with 2.5 type representations in py.test [45483, 45484]
* made that internal reporting issues displaying is done atomically in py.test
* made that non-existing files are igored by the py.lookup script 
* improved exception name creation in py.test 
* made that less threads are used in execnet [merge in 45539]
* removed lock required for atomical reporting issue displaying in py.test
* removed globals from execnet [45541, 45547]
* refactored cleanup mechanics, made that setDaemon is set to 1 to make atexit
get called in 2.5 (py.execnet) 
* fixed bug in joining threads in py.execnet's servemain 
* refactored py.test.rsession tests to not rely on exact output format anymore
* using repr() on test outcome 
* added 'Reason' classes for py.test.skip() [45648, 45649]
* killed some unnecessary sanity check in py.test.collect 
* avoid using os.tmpfile() in py.io.fdcapture because on Windows it's only
usable by Administrators 
* added support for locking and non-recursive commits to py.path.svnwc 
* locking files in py.execnet to prevent CPython from segfaulting 
* added export() method to py.path.svnurl
* fixed -d -x in py.test 
* fixed argument concatenation problem in py.path.svnwc 
* restore py.test behaviour that it exits with code 1 when there are failures
* don't fail on html files that don't have an accompanying .txt file 
* fixed 'utestconvert.py < input' 
* small fix for code indentation in py.code.source 
* fix _docgen.py documentation building 
* improved checks for source representation of code blocks in py.test 
* added support for passing authentication to py.path.svn* objects [52000,
* removed sorted() call for py.apigen tests in favour of .sort() to support
Python 2.3 
More information about the testing-in-python