pypi pytest 6.0.0rc1

latest releases: 8.3.3, 8.3.2, 8.3.1...
4 years ago

pytest 6.0.0rc1 (2020-07-08)

Breaking Changes

  • #1316: TestReport.longrepr is now always an instance of ReprExceptionInfo. Previously it was a str when a test failed with pytest.fail(..., pytrace=False).

  • #5965: symlinks are no longer resolved during collection and matching conftest.py files with test file paths.

    Resolving symlinks for the current directory and during collection was introduced as a bugfix in 3.9.0, but it actually is a new feature which had unfortunate consequences in Windows and surprising results in other platforms.

    The team decided to step back on resolving symlinks at all, planning to review this in the future with a more solid solution (see discussion in
    #6523 for details).

    This might break test suites which made use of this feature; the fix is to create a symlink
    for the entire test tree, and not only to partial files/tress as it was possible previously.

  • #6505: Testdir.run().parseoutcomes() now always returns the parsed nouns in plural form.

    Originally parseoutcomes() would always returns the nouns in plural form, but a change
    meant to improve the terminal summary by using singular form single items (1 warning or 1 error)
    caused an unintended regression by changing the keys returned by parseoutcomes().

    Now the API guarantees to always return the plural form, so calls like this:

    result = testdir.runpytest()
    result.assert_outcomes(error=1)
    

    Need to be changed to:

    result = testdir.runpytest()
    result.assert_outcomes(errors=1)
    
  • #6903: The os.dup() function is now assumed to exist. We are not aware of any
    supported Python 3 implementations which do not provide it.

  • #7040: -k no longer matches against the names of the directories outside the test session root.

    Also, pytest.Package.name is now just the name of the directory containing the package's
    __init__.py file, instead of the full path. This is consistent with how the other nodes
    are named, and also one of the reasons why -k would match against any directory containing
    the test suite.

  • #7122: Expressions given to the -m and -k options are no longer evaluated using Python's eval.
    The format supports or, and, not, parenthesis and general identifiers to match against.
    Python constants, keywords or other operators are no longer evaluated differently.

  • #7135: Pytest now uses its own TerminalWriter class instead of using the one from the py library.
    Plugins generally access this class through TerminalReporter.writer, TerminalReporter.write()
    (and similar methods), or _pytest.config.create_terminal_writer().

    The following breaking changes were made:

    • Output (write() method and others) no longer flush implicitly; the flushing behavior
      of the underlying file is respected. To flush explicitly (for example, if you
      want output to be shown before an end-of-line is printed), use write(flush=True) or
      terminal_writer.flush().
    • Explicit Windows console support was removed, delegated to the colorama library.
    • Support for writing bytes was removed.
    • The reline method and chars_on_current_line property were removed.
    • The stringio and encoding arguments was removed.
    • Support for passing a callable instead of a file was removed.
  • #7224: The item.catch_log_handler and item.catch_log_handlers attributes, set by the
    logging plugin and never meant to be public , are no longer available.

    The deprecated --no-print-logs option is removed. Use --show-capture instead.

  • #7226: Removed the unused args parameter from pytest.Function.__init__.

  • #7418: Removed the pytest_doctest_prepare_content hook specification. This hook
    hasn't been triggered by pytest for at least 10 years.

  • #7438: Some changes were made to the internal _pytest._code.source, listed here
    for the benefit of plugin authors who may be using it:

    • The deindent argument to Source() has been removed, now it is always true.
    • Support for zero or multiple arguments to Source() has been removed.
    • Support for comparing Source with an str has been removed.
    • The methods Source.isparseable() and Source.putaround() have been removed.
    • The method Source.compile() and function _pytest._code.compile() have
      been removed; use plain compile() instead.
    • The function _pytest._code.source.getsource() has been removed; use
      Source() directly instead.

Deprecations

  • #7210: The special -k '-expr' syntax to -k is deprecated. Use -k 'not expr'
    instead.

    The special -k 'expr:' syntax to -k is deprecated. Please open an issue
    if you use this and want a replacement.

  • #4049: pytest_warning_captured is deprecated in favor of the pytest_warning_recorded hook.

Features

  • #1556: pytest now supports pyproject.toml files for configuration.

    The configuration options is similar to the one available in other formats, but must be defined
    in a [tool.pytest.ini_options] table to be picked up by pytest:

    # pyproject.toml
    [tool.pytest.ini_options]
    minversion = "6.0"
    addopts = "-ra -q"
    testpaths = [
        "tests",
        "integration",
    ]
    

    More information can be found in the docs.

  • #3342: pytest now includes inline type annotations and exposes them to user programs.
    Most of the user-facing API is covered, as well as internal code.

    If you are running a type checker such as mypy on your tests, you may start
    noticing type errors indicating incorrect usage. If you run into an error that
    you believe to be incorrect, please let us know in an issue.

    The types were developed against mypy version 0.780. Versions before 0.750
    are known not to work. We recommend using the latest version. Other type
    checkers may work as well, but they are not officially verified to work by
    pytest yet.

  • #4049: Introduced a new hook named pytest_warning_recorded to convey information about warnings captured by the internal pytest warnings plugin.

    This hook is meant to replace pytest_warning_captured, which is deprecated and will be removed in a future release.

  • #6471: New command-line flags:

    • `--no-header`: disables the initial header, including platform, version, and plugins.
    • `--no-summary`: disables the final test summary, including warnings.
  • #6856: A warning is now shown when an unknown key is read from a config INI file.

    The --strict-config flag has been added to treat these warnings as errors.

  • #6906: Added --code-highlight command line option to enable/disable code highlighting in terminal output.

  • #7245: New --import-mode=importlib option that uses importlib to import test modules.

    Traditionally pytest used __import__ while changing sys.path to import test modules (which
    also changes sys.modules as a side-effect), which works but has a number of drawbacks, like requiring test modules
    that don't live in packages to have unique names (as they need to reside under a unique name in sys.modules).

    --import-mode=importlib uses more fine grained import mechanisms from importlib which don't
    require pytest to change sys.path or sys.modules at all, eliminating much of the drawbacks
    of the previous mode.

    We intend to make --import-mode=importlib the default in future versions, so users are encouraged
    to try the new mode and provide feedback (both positive or negative) in issue #7245.

    You can read more about this option in the documentation.

  • #7305: New required_plugins configuration option allows the user to specify a list of plugins, including version information, that are required for pytest to run. An error is raised if any required plugins are not found when running pytest.

Improvements

  • #4375: The pytest command now suppresses the BrokenPipeError error message that
    is printed to stderr when the output of pytest is piped and and the pipe is
    closed by the piped-to program (common examples are less and head).

  • #4391: Improved precision of test durations measurement. CallInfo items now have a new <CallInfo>.duration attribute, created using time.perf_counter(). This attribute is used to fill the <TestReport>.duration attribute, which is more accurate than the previous <CallInfo>.stop - <CallInfo>.start (as these are based on time.time()).

  • #4675: Rich comparison for dataclasses and attrs-classes is now recursive.

  • #6285: Exposed the pytest.FixtureLookupError exception which is raised by request.getfixturevalue()
    (where request is a FixtureRequest fixture) when a fixture with the given name cannot be returned.

  • #6433: If an error is encountered while formatting the message in a logging call, for
    example logging.warning("oh no!: %s: %s", "first") (a second argument is
    missing), pytest now propagates the error, likely causing the test to fail.

    Previously, such a mistake would cause an error to be printed to stderr, which
    is not displayed by default for passing tests. This change makes the mistake
    visible during testing.

    You may supress this behavior temporarily or permanently by setting
    logging.raiseExceptions = False.

  • #6817: Explicit new-lines in help texts of command-line options are preserved, allowing plugins better control
    of the help displayed to users.

  • #6940: When using the --duration option, the terminal message output is now more precise about the number and duration of hidden items.

  • #6991: Collected files are displayed after any reports from hooks, e.g. the status from --lf.

  • #7091: When fd capturing is used, through --capture=fd or the capfd and
    capfdbinary fixtures, and the file descriptor (0, 1, 2) cannot be
    duplicated, FD capturing is still performed. Previously, direct writes to the
    file descriptors would fail or be lost in this case.

  • #7119: Exit with an error if the --basetemp argument is empty, is the current working directory or is one of the parent directories.
    This is done to protect against accidental data loss, as any directory passed to this argument is cleared.

  • #7128: pytest --version now displays just the pytest version, while pytest --version --version displays more verbose information including plugins. This is more consistent with how other tools show --version.

  • #7133: caplog.set_level() <_pytest.logging.LogCaptureFixture.set_level> will now override any log_level set via the CLI or configuration file.

  • #7159: caplog.set_level() <_pytest.logging.LogCaptureFixture.set_level> and caplog.at_level() <_pytest.logging.LogCaptureFixture.at_level> no longer affect
    the level of logs that are shown in the Captured log report report section.

  • #7348: Improve recursive diff report for comparison asserts on dataclasses / attrs.

  • #7385: --junitxml now includes the exception cause in the message XML attribute for failures during setup and teardown.

    Previously:

    <error message="test setup failure">
    

    Now:

    <error message="failed on setup with &quot;ValueError: Some error during setup&quot;">
    

Bug Fixes

  • #1120: Fix issue where directories from tmpdir are not removed properly when multiple instances of pytest are running in parallel.

  • #4583: Prevent crashing and provide a user-friendly error when a marker expression (-m) invoking of eval raises any exception.

  • #4677: The path shown in the summary report for SKIPPED tests is now always relative. Previously it was sometimes absolute.

  • #5456: Fix a possible race condition when trying to remove lock files used to control access to folders
    created by tmp_path and tmpdir.

  • #6240: Fixes an issue where logging during collection step caused duplication of log
    messages to stderr.

  • #6428: Paths appearing in error messages are now correct in case the current working directory has
    changed since the start of the session.

  • #6755: Support deleting paths longer than 260 characters on windows created inside tmpdir.

  • #6871: Fix crash with captured output when using capsysbinary.

  • #6909: Revert the change introduced by #6330, which required all arguments to @pytest.mark.parametrize to be explicitly defined in the function signature.

    The intention of the original change was to remove what was expected to be an unintended/surprising behavior, but it turns out many people relied on it, so the restriction has been reverted.

  • #6910: Fix crash when plugins return an unknown stats while using the --reportlog option.

  • #6924: Ensure a unittest.IsolatedAsyncioTestCase is actually awaited.

  • #6925: Fix TerminalRepr instances to be hashable again.

  • #6947: Fix regression where functions registered with unittest.TestCase.addCleanup were not being called on test failures.

  • #6951: Allow users to still set the deprecated TerminalReporter.writer attribute.

  • #6956: Prevent pytest from printing ConftestImportFailure traceback to stdout.

  • #6991: Fix regressions with --lf filtering too much since pytest 5.4.

  • #6992: Revert "tmpdir: clean up indirection via config for factories" #6767 as it breaks pytest-xdist.

  • #7061: When a yielding fixture fails to yield a value, report a test setup error instead of crashing.

  • #7076: The path of file skipped by @pytest.mark.skip in the SKIPPED report is now relative to invocation directory. Previously it was relative to root directory.

  • #7110: Fixed regression: asyncbase.TestCase tests are executed correctly again.

  • #7126: --setup-show now doesn't raise an error when a bytes value is used as a parametrize
    parameter when Python is called with the -bb flag.

  • #7143: Fix pytest.File.from_parent so it forwards extra keyword arguments to the constructor.

  • #7145: Classes with broken __getattribute__ methods are displayed correctly during failures.

  • #7150: Prevent hiding the underlying exception when ConfTestImportFailure is raised.

  • #7180: Fix _is_setup_py for files encoded differently than locale.

  • #7215: Fix regression where running with --pdb would call unittest.TestCase.tearDown for skipped tests.

  • #7253: When using pytest.fixture on a function directly, as in pytest.fixture(func),
    if the autouse or params arguments are also passed, the function is no longer
    ignored, but is marked as a fixture.

  • #7360: Fix possibly incorrect evaluation of string expressions passed to pytest.mark.skipif and pytest.mark.xfail,
    in rare circumstances where the exact same string is used but refers to different global values.

  • #7383: Fixed exception causes all over the codebase, i.e. use raise new_exception from old_exception when wrapping an exception.

Improved Documentation

  • #7202: The development guide now links to the contributing section of the docs and RELEASING.rst on GitHub.
  • #7233: Add a note about --strict and --strict-markers and the preference for the latter one.
  • #7345: Explain indirect parametrization and markers for fixtures.

Trivial/Internal Changes

  • #7035: The originalname attribute of _pytest.python.Function now defaults to name if not
    provided explicitly, and is always set.
  • #7264: The dependency on the wcwidth package has been removed.
  • #7291: Replaced py.iniconfig with iniconfig.
  • #7295: src/_pytest/config/__init__.py now uses the warnings module to report warnings instead of sys.stderr.write.
  • #7356: Remove last internal uses of deprecated slave term from old pytest-xdist.
  • #7357: py>=1.8.2 is now required.

Don't miss a new pytest release

NewReleases is sending notifications on new releases.