Skip to content

Latest commit

 

History

History
677 lines (534 loc) · 29.1 KB

69_0.rst

File metadata and controls

677 lines (534 loc) · 29.1 KB
Raw

69.0 LTS

The Avocado team is proud to present another LTS (Long Term Stability) release: Avocado 69.0, AKA "The King's Choice", is now available!

LTS Release

For more information on what a LTS release means, please read :ref:`rfc-long-term-stability`.

Upgrading from 52.x to 69.0

Upgrading Installations

Avocado is available on a number of different repositories and installation methods. You can find the complete details in :ref:`Installing Avocado`. After looking at your installation options, please consider the following highlights about the changes in the Avocado installation:

  • Avocado fully supports both Python 2 and 3, and both can even be installed simultaneously. When using RPM packages, if you ask to have python-avocado installed, it will be provided by the Python 2 based package. If you want a Python 3 based version you must use the python3-avocado package. The same is true for plugins, which have a python2-avocado-plugins or python3-avocado-plugins prefix.
  • Avocado can now be properly installed without super user privileges. Previously one would see an error such as could not create '/etc/avocado': Permission denied when trying to do a source or PIP based installation.
  • When installing Avocado on Python "venvs", the user's base data directory is now within the venv. If you had content outside the venv, such as results or tests directories, please make sure that you either configure your data directories on the [datadir.paths] section of your configuration file, or move the data over.

Porting Tests (Test API compatibility)

If you're migration from the previous LTS version, these are the changes on the Test API that most likely will affect your test.

Note

Between non-LTS releases, the Avocado Test APIs receive a lot of effort to be kept as stable as possible. When that's not possible, a deprecation strategy is applied and breakage can occur. For guaranteed stability across longer periods of time, LTS releases such as this one should be used.

  • Support for default test parameters, given via the class level default_params dictionary has been removed. If your test contains a snippet similar to:

    default_params = {'param1': 'value1',
                  'param2': 'value2'}
def test(self):
    value1 = self.params.get('param1')
    value2 = self.params.get('param2')

    It should be rewritten to look like this:

    def test(self):
    value1 = self.params.get('param1', default='value1')
    value2 = self.params.get('param2', default='value2')

  • Support for getting parameters using the self.params.key syntax has been removed. If your test contains a snippet similar to:

    def test(self):
    value1 = self.params.key1

    It should be rewritten to look like this:

    def test(self):
    value1 = self.params.get('key1')

  • Support for the datadir test class attribute has been removed in favor of the get_data() method. If your test contains a snippet similar to:

    def test(self):
    data = os.path.join(self.datadir, 'data')

    It should be rewritten to look like this:

    def test(self):
    data = self.get_data('data')

  • Support for for srcdir test class attribute has been removed in favor of the workdir attribute. If your test contains a snippet similar to:

    def test(self):
    compiled = os.path.join(self.srcdir, 'binary')

    It should be rewritten to look like this:

    def test(self):
    compiled = os.path.join(self.workdir, 'binary')

  • The :avocado: enable and :avocado: recursive tags may not be necessary anymore, given that "recursive" is now the default loader behavior. If you test contains:

    def test(self):
   """
   :avocado: enable
   """

    Or:

    def test(self):
   """
   :avocado: recursive
   """

    Consider removing the tags completely, and check if the default loader behavior is sufficient with:

    $ avocado list your-test-file.py

  • Support for the skip method has been removed from the avocado.Test class. If your test contains a snippet similar to:

    def test(self):
    if not condition():
        self.skip("condition not suitable to keep test running")

    It should be rewritten to look like this:

    def test(self):
    if not condition():
        self.cancel("condition not suitable to keep test running")

Porting Tests (Utility API compatibility)

The changes in the utility APIs (those that live under the avocado.utils namespace are too many to present porting suggestion. Please refer to the :ref:`lts_069_utility_api_changes` section for a comprehensive list of changes, including new features your test may be able to leverage.

Changes from previous LTS

Note

This is not a collection of all changes encompassing all releases from 52.0 to 69.0. This list contains changes that are relevant to users of 52.0, when evaluating an upgrade to 69.0.

When compared to the last LTS (version 52.1), the main changes introduced by this versions are:

Test Writers

Test APIs

  • Test writers will get better protection against mistakes when trying to overwrite :class:`avocado.core.test.Test` "properties". Some of those were previously implemented using :func:`avocado.utils.data_structures.LazyProperty` which did not prevent test writers from overwriting them.
  • The avocado.Test.default_parameters mechanism for setting default parameters on tests has been removed. This was introduced quite early in the Avocado development, and allowed users to set a dictionary at the class level with keys/values that would serve as default parameter values. The recommended approach now, is to just provide default values when calling the self.params.get within a test method, such as self.params.get("key", default="default_value_for_key").
  • The __getattr__ interface for self.params has been removed. It used to allow users to use a syntax such as self.params.key when attempting to access the value for key key. The supported syntax is self.params.get("key") to achieve the same thing.
  • The support for test data files has been improved to support more specific sources of data. For instance, when a test file used to contain more than one test, all of them shared the same datadir property value, thus the same directory which contained data files. Now, tests should use the newly introduced :meth:`get_data() <avocado.core.test.TestData.get_data>` API, which will attempt to locate data files specific to the variant (if used), test name, and finally file name. For more information, please refer to the section :ref:`accessing-test-data-files`.
  • The avocado.Test.srcdir attribute has been removed, and with it, the AVOCADO_TEST_SRCDIR environment variable set by Avocado. Tests should have been modified by now to make use of the :attr:`avocado.Test.workdir` instead.
  • The avocado.Test.datadir attribute has been removed, and with it, the AVOCADO_TEST_DATADIR environment variable set by Avocado. Tests should now to make use of the :func:`avocado.Test.get_data() <avocado.core.test.TestData.get_data>` instead.
  • Switched the FileLoader discovery to 🥑 recursive by default. All tags enable, disable and recursive are still available and might help fine-tuning the class visibility.
  • The deprecated skip method, previously part of the :class:`avocado.Test` API, has been removed. To skip a test, you can still use the :func:`avocado.skip`, :func:`avocado.skipIf` and :func:`avocado.skipUnless` decorators.
  • The :class:`Avocado Test class <avocado.core.test.Test>` now exposes the :attr:`tags <avocado.core.test.Test.tags>` to the test. The test may use that information, for instance, to decide on default behavior.
  • The Avocado test loader, which does not load or execute Python source code that may contain tests for security reasons, now operates in a way much more similar to the standard Python object inheritance model. Before, classes containing tests that would not directly inherit from :class:`avocado.Test` would require a docstring statement (either :avocado: enable or :avocado: recursive). This is not necessary for most users anymore, as the recursive detection is now the default behavior.

Utility APIs

Users

  • Passing parameters to tests is now possible directly on the Avocado command line, without the use of any varianter plugin. In fact, when using variants, these parameters are (currently) ignored. To pass one parameter to a test, use -p NAME=VAL, and repeat it for other parameters.
  • The test filtering mechanism using tags now support "key:val" assignments for further categorization. See :ref:`tags_keyval` for more details.
  • The output generated by tests on stdout and stderr are now properly prefixed with [stdout] and [stderr] in the job.log. The prefix is not applied in the case of $test_result/stdout and $test_result/stderr files, as one would expect.
  • The installation of Avocado from sources has improved and moved towards a more "Pythonic" approach. Installation of files in "non-Pythonic locations" such as /etc are no longer attempted by the Python setup.py code. Configuration files, for instance, are now considered package data files of the avocado package. The end result is that installation from source works fine outside virtual environments (in addition to installations inside virtual environments). For instance, the locations of /etc (config) and /usr/libexec (libexec) files changed to live within the pkg_data (eg. /usr/lib/python2.7/site-packages/avocado/etc) by default in order to not to modify files outside the package dir, which allows user installation and also the distribution of wheel packages. GNU/Linux distributions might still modify this to better follow their conventions (eg. for RPM the original locations are used). Please refer to the output of the avocado config command to see the configuration files that are actively being used on your installation.
  • SIMPLE tests were limited to returning PASS, FAIL and WARN statuses. Now SIMPLE tests can now also return SKIP status. At the same time, SIMPLE tests were previously limited in how they would flag a WARN or SKIP from the underlying executable. This is now configurable by means of regular expressions.
  • Sysinfo collection can now be enabled on a test level basis.
  • Avocado can record the output generated from a test, which can then be used to determine if the test passed or failed. This feature is commonly known as "output check". Traditionally, users would choose to record the output from STDOUT and/or STDERR into separate streams, which would be saved into different files. Some tests suites actually put all content of STDOUT and STDERR together, and unless we record them together, it'd be impossible to record them in the right order. This version introduces the combined option to --output-check-record option, which does exactly that: it records both STDOUT and STDERR into a single stream and into a single file (named output in the test results, and output.expected in the test data directory).
  • The complete output of tests, that is the combination of STDOUT and STDERR is now also recorded in the test result directory as a file named output.
  • When the output check feature finds a mismatch between expected and actual output, will now produce a unified diff of those, instead of printing out their full content. This makes it a lot easier to read the logs and quickly spot the differences and possibly the failure cause(s).
  • The output check feature will now use the to the most specific data source location available, which is a consequence of the switch to the use of the get_data() API discussed previously. This means that two tests in a single file can generate different output, generate different stdout.expected or stderr.expected.
  • :ref:`SIMPLE <test_type_simple>` tests can also finish with SKIP OR WARN status, depending on the output produced, and the Avocado test runner configuration. It now supports patterns that span across multiple lines. For more information, refer to :ref:`test_type_simple_status`.
  • A better handling of interruption related signals, such as SIGINT and SIGTERM. Avocado will now try harder to not leave test processes that don't respond to those signals, and will itself behave better when it receives them. For a complete description refer to :ref:`signal_handlers`.
  • Improvements in the serialization of TestIDs allow test result directories to be properly stored and accessed on Windows based filesystems.
  • The deprecated jobdata/urls link to jobdata/test_references has been removed.
  • The avocado command line argument parser is now invoked before plugins are initialized, which allows the use of --config with configuration file that influence plugin behavior.
  • The test log now contains a number of metadata about the test, under the heading Test metadata:. You'll find information such as the test file name (if one exists), its workdir and its teststmpdir if one is set.
  • The test runner wil now log the test initialization (look for INIT in your test logs) in addition to the already existing start of test execution (logged as START).
  • The test profilers, which are defined by default in /etc/avocado/sysinfo/profilers, are now executed without a backing shell. While Avocado doesn't ship with examples of shell commands as profilers, or suggests users to do so, it may be that some users could be using that functionality. If that's the case, it will now be necessary to write a script that wraps you previous shell command. The reason for doing so, was to fix a bug that could leave profiler processes after the test had already finished.
  • The Human UI plugin, will now show the "reason" behind test failures, cancellations and others right along the test result status. This hopefully will give more information to users without requiring them to resort to logs every single time.
  • When installing and using Avocado in a Python virtual environment, the ubiquitous "venvs", the base data directory now respects the virtual environment. If you have are using the default data directory outside of a venv, please be aware that the updated
  • Avocado packages are now available in binary "wheel" format on PyPI. This brings faster, more convenient and reliable installs via pip. Previously, the source-only tarballs would require the source to be built on the target system, but the wheel package install is mostly an unpack of the already compiled files.
  • The legacy options --filter-only, --filter-out and --multiplex have now been removed. Please adjust your usage, replacing those options with --mux-filter-only, --mux-filter-out and --mux-yaml respectively.
  • The location of the Avocado configuration files can now be influenced by third parties by means of a new plugin.
  • The configuration files that have been effectively parsed are now displayed as part of avocado config command output.

Output Plugins

  • Including test logs in TAP plugin is disabled by default and can be enabled using --tap-include-logs.
  • The TAP result format plugin received improvements, including support for reporting Avocado tests with CANCEL status as SKIP (which is the closest status available in the TAP specification), and providing more visible warning information in the form of comments when Avocado tests finish with WARN status (while maintaining the test as a PASS, since TAP doesn't define a WARN status).
  • A new (optional) plugin is available, the "result uploader". It allows job results to be copied over to a centralized results server at the end of job execution. Please refer to :ref:`results-upload-plugin` for more information.
  • Added possibility to limit the amount of characters embedded as "system-out" in the xunit output plugin (--xunit-max-test-log-chars XX).
  • The xunit result plugin can now limit the amount of output generated by individual tests that will make into the XML based output file. This is intended for situations where tests can generate prohibitive amounts of output that can render the file too large to be reused elsewhere (such as imported by Jenkins).
  • The xunit output now names the job after the Avocado job results directory. This should make the correlation of results displayed in UIs such as Jenkins and the complete Avocado results much easier.
  • The xUnit plugin now should produce output that is more compatible with other implementations, specifically newer Jenkin's as well as Ant and Maven. The specific change was to format the time field with 3 decimal places.
  • Redundant (and deprecated) fields in the test sections of the JSON result output were removed. Now, instead of url, test and id carrying the same information, only id remains.

Test Loader Plugins

  • A new loader implementation, that reuses (and resembles) the YAML input used for the varianter yaml_to_mux plugin. It allows the definition of test suite based on a YAML file, including different variants for different tests. For more information refer to :ref:`yaml_loader`.
  • Users of the YAML test loader have now access to a few special keys that can tweak test attributes, including adding prefixes to test names. This allows users to easily differentiate among execution of the same test, but executed different configurations. For more information, look for "special keys" in the :ref:`YAML Loader plugin documentation <yaml_loader>`.
  • A :ref:`new plugin <glib-plugin>` enables users to list and execute tests based on the GLib test framework. This plugin allows individual tests inside a single binary to be listed and executed.
  • Avocado can now run list and run standard Python unittests, that is, tests written in Python that use the :mod:`unittest` library alone.
  • Support for listing and running golang tests has been introduced. Avocado can now discover tests written in Go, and if Go is properly installed, Avocado can run them.

Varianter Plugins

  • A new varianter plugin has been introduced, based on PICT. PICT is a "Pair Wise" combinatorial tool, that can generate optimal combination of parameters to tests, so that (by default) at least a unique pair of parameter values will be tested at once.
  • A new varianter plugin, the :ref:`cit-varianter-plugin`. This plugin implements a "Pair-Wise", also known as "Combinatorial Independent Testing" algorithm, in pure Python. This exciting new functionality is provided thanks to a collaboration with the Czech Technical University in Prague.
  • Users can now dump variants to a (JSON) file, and also reuse a previously created file in their future jobs execution. This allows users to avoid recomputing the variants on every job, which might bring significant speed ups in job execution or simply better control of the variants used during a job. Also notice that even when users do not manually dump a variants file to a specific location, Avocado will automatically save a suitable file at jobdata/variants.json as part of a Job results directory structure. The feature has been isolated into a varianter implementation called json_variants, that you can see with avocado plugins.

Test Runner Plugins

  • The command line options --filter-by-tags and --filter-by-tags-include-empty are now white listed for the remote runner plugin.
  • The remote runner plugin will now respect ~/.ssh/config configuration.

Complete list of changes

For a complete list of changes between the last LTS release (52.1) and this release, please check out the Avocado commit changelog.