Scheduled weekly dependency update for week 50

[pyup-bot]

Update appdirs from 1.4.3 to 1.4.4.

Changelog

1.4.4

-------------
- [PR 92] Don't import appdirs from setup.py which resolves issue 91
- [PR 100] Corrects the config directory on OSX/macOS, which resolves issue 63.

Project officially classified as Stable which is important
for inclusion in other distros such as ActivePython.

Links

• PyPI: https://pypi.org/project/appdirs
• Changelog: https://pyup.io/changelogs/appdirs/
• Repo: http://github.com/ActiveState/appdirs

Update async-generator from 1.9 to 1.10.

The bot wasn't able to find a changelog for this release. Got an idea?

Links

• PyPI: https://pypi.org/project/async-generator
• Repo: https://github.com/python-trio/async_generator

Update attrs from 18.1.0 to 20.3.0.

Changelog

20.3.0

-------------------

Backward-incompatible Changes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

- ``attr.define()``, ``attr.frozen()``, ``attr.mutable()``, and ``attr.field()`` remain **provisional**.

This release does **not** change change anything about them and they are already used widely in production though.

If you wish to use them together with mypy, you can simply drop `this plugin <https://gist.github.com/hynek/1e3844d0c99e479e716169034b5fa963file-attrs_ng_plugin-py>`_ into your project.

Feel free to provide feedback to them in the linked issue 668.

We will release the ``attrs`` namespace once we have the feeling that the APIs have properly settled.
`668 <https://github.com/python-attrs/attrs/issues/668>`_


Changes
^^^^^^^

- ``attr.s()`` now has a *field_transformer* hook that is called for all ``Attribute``\ s and returns a (modified or updated) list of ``Attribute`` instances.
``attr.asdict()`` has a *value_serializer* hook that can change the way values are converted.
Both hooks are meant to help with data (de-)serialization workflows.
`653 <https://github.com/python-attrs/attrs/issues/653>`_
- ``kw_only=True`` now works on Python 2.
`700 <https://github.com/python-attrs/attrs/issues/700>`_
- ``raise from`` now works on frozen classes on PyPy.
`703 <https://github.com/python-attrs/attrs/issues/703>`_,
`712 <https://github.com/python-attrs/attrs/issues/712>`_
- ``attr.asdict()`` and ``attr.astuple()`` now treat ``frozenset``\ s like ``set``\ s with regards to the *retain_collection_types* argument.
`704 <https://github.com/python-attrs/attrs/issues/704>`_
- The type stubs for ``attr.s()`` and ``attr.make_class()`` are not missing the *collect_by_mro* argument anymore.
`711 <https://github.com/python-attrs/attrs/issues/711>`_


----

20.2.0

-------------------

Backward-incompatible Changes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

- ``attr.define()``, ``attr.frozen()``, ``attr.mutable()``, and ``attr.field()`` remain **provisional**.

This release fixes a bunch of bugs and ergonomics but they remain mostly unchanged.

If you wish to use them together with mypy, you can simply drop `this plugin <https://gist.github.com/hynek/1e3844d0c99e479e716169034b5fa963file-attrs_ng_plugin-py>`_ into your project.

Feel free to provide feedback to them in the linked issue 668.

We will release the ``attrs`` namespace once we have the feeling that the APIs have properly settled.
`668 <https://github.com/python-attrs/attrs/issues/668>`_


Changes
^^^^^^^

- ``attr.define()`` et al now correct detect ``__eq__`` and ``__ne__``.
`671 <https://github.com/python-attrs/attrs/issues/671>`_
- ``attr.define()`` et al's hybrid behavior now also works correctly when arguments are passed.
`675 <https://github.com/python-attrs/attrs/issues/675>`_
- It's possible to define custom ``__setattr__`` methods on slotted classes again.
`681 <https://github.com/python-attrs/attrs/issues/681>`_
- In 20.1.0 we introduced the ``inherited`` attribute on the ``attr.Attribute`` class to differentiate attributes that have been inherited and those that have been defined directly on the class.

It has shown to be problematic to involve that attribute when comparing instances of ``attr.Attribute`` though, because when sub-classing, attributes from base classes are suddenly not equal to themselves in a super class.

Therefore the ``inherited`` attribute will now be ignored when hashing and comparing instances of ``attr.Attribute``.
`684 <https://github.com/python-attrs/attrs/issues/684>`_
- ``zope.interface`` is now a "soft dependency" when running the test suite; if ``zope.interface`` is not installed when running the test suite, the interface-related tests will be automatically skipped.
`685 <https://github.com/python-attrs/attrs/issues/685>`_
- The ergonomics of creating frozen classes using ``define(frozen=True)`` and sub-classing frozen classes has been improved:
you don't have to set ``on_setattr=None`` anymore.
`687 <https://github.com/python-attrs/attrs/issues/687>`_


----

20.1.0

-------------------

Backward-incompatible Changes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

- Python 3.4 is not supported anymore.
It has been unsupported by the Python core team for a while now, its PyPI downloads are negligible, and our CI provider removed it as a supported option.

It's very unlikely that ``attrs`` will break under 3.4 anytime soon, which is why we do *not* block its installation on Python 3.4.
But we don't test it anymore and will block it once someone reports breakage.
`608 <https://github.com/python-attrs/attrs/issues/608>`_


Deprecations
^^^^^^^^^^^^

- Less of a deprecation and more of a heads up: the next release of ``attrs`` will introduce an ``attrs`` namespace.
That means that you'll finally be able to run ``import attrs`` with new functions that aren't cute abbreviations and that will carry better defaults.

This should not break any of your code, because project-local packages have priority before installed ones.
If this is a problem for you for some reason, please report it to our bug tracker and we'll figure something out.

The old ``attr`` namespace isn't going anywhere and its defaults are not changing – this is a purely additive measure.
Please check out the linked issue for more details.

These new APIs have been added *provisionally* as part of 666 so you can try them out today and provide feedback.
Learn more in the `API docs <https://www.attrs.org/en/stable/api.htmlprovisional-apis>`_.
`408 <https://github.com/python-attrs/attrs/issues/408>`_


Changes
^^^^^^^

- Added ``attr.resolve_types()``.
It ensures that all forward-references and types in string form are resolved into concrete types.

You need this only if you need concrete types at runtime.
That means that if you only use types for static type checking, you do **not** need this function.
`288 <https://github.com/python-attrs/attrs/issues/288>`_,
`302 <https://github.com/python-attrs/attrs/issues/302>`_
- Added ``attr.s(collect_by_mro=False)`` argument that if set to ``True`` fixes the collection of attributes from base classes.

It's only necessary for certain cases of multiple-inheritance but is kept off for now for backward-compatibility reasons.
It will be turned on by default in the future.

As a side-effect, ``attr.Attribute`` now *always* has an ``inherited`` attribute indicating whether an attribute on a class was directly defined or inherited.
`428 <https://github.com/python-attrs/attrs/issues/428>`_,
`635 <https://github.com/python-attrs/attrs/issues/635>`_
- On Python 3, all generated methods now have a docstring explaining that they have been created by ``attrs``.
`506 <https://github.com/python-attrs/attrs/issues/506>`_
- It is now possible to prevent ``attrs`` from auto-generating the ``__setstate__`` and ``__getstate__`` methods that are required for pickling of slotted classes.

Either pass ``attr.s(getstate_setstate=False)`` or pass ``attr.s(auto_detect=True)`` and implement them yourself:
if ``attrs`` finds either of the two methods directly on the decorated class, it assumes implicitly ``getstate_setstate=False`` (and implements neither).

This option works with dict classes but should never be necessary.
`512 <https://github.com/python-attrs/attrs/issues/512>`_,
`513 <https://github.com/python-attrs/attrs/issues/513>`_,
`642 <https://github.com/python-attrs/attrs/issues/642>`_
- Fixed a ``ValueError: Cell is empty`` bug that could happen in some rare edge cases.
`590 <https://github.com/python-attrs/attrs/issues/590>`_
- ``attrs`` can now automatically detect your own implementations and infer ``init=False``, ``repr=False``, ``eq=False``, ``order=False``, and ``hash=False`` if you set ``attr.s(auto_detect=True)``.
``attrs`` will ignore inherited methods.
If the argument implies more than one method (e.g. ``eq=True`` creates both ``__eq__`` and ``__ne__``), it's enough for *one* of them to exist and ``attrs`` will create *neither*.

This feature requires Python 3.
`607 <https://github.com/python-attrs/attrs/issues/607>`_
- Added ``attr.converters.pipe()``.
The feature allows combining multiple conversion callbacks into one by piping the value through all of them, and retuning the last result.

As part of this feature, we had to relax the type information for converter callables.
`618 <https://github.com/python-attrs/attrs/issues/618>`_
- Fixed serialization behavior of non-slots classes with ``cache_hash=True``.
The hash cache will be cleared on operations which make "deep copies" of instances of classes with hash caching,
though the cache will not be cleared with shallow copies like those made by ``copy.copy()``.

Previously, ``copy.deepcopy()`` or serialization and deserialization with ``pickle`` would result in an un-initialized object.

This change also allows the creation of ``cache_hash=True`` classes with a custom ``__setstate__``,
which was previously forbidden (`494 <https://github.com/python-attrs/attrs/issues/494>`_).
`620 <https://github.com/python-attrs/attrs/issues/620>`_
- It is now possible to specify hooks that are called whenever an attribute is set **after** a class has been instantiated.

You can pass ``on_setattr`` both to ``attr.s()`` to set the default for all attributes on a class, and to ``attr.ib()`` to overwrite it for individual attributes.

``attrs`` also comes with a new module ``attr.setters`` that brings helpers that run validators, converters, or allow to freeze a subset of attributes.
`645 <https://github.com/python-attrs/attrs/issues/645>`_,
`660 <https://github.com/python-attrs/attrs/issues/660>`_
- **Provisional** APIs called ``attr.define()``, ``attr.mutable()``, and ``attr.frozen()`` have been added.

They are only available on Python 3.6 and later, and call ``attr.s()`` with different default values.

If nothing comes up, they will become the official way for creating classes in 20.2.0 (see above).

**Please note** that it may take some time until mypy – and other tools that have dedicated support for ``attrs`` – recognize these new APIs.
Please **do not** open issues on our bug tracker, there is nothing we can do about it.
`666 <https://github.com/python-attrs/attrs/issues/666>`_
- We have also provisionally added ``attr.field()`` that supplants ``attr.ib()``.
It also requires at least Python 3.6 and is keyword-only.
Other than that, it only dropped a few arguments, but changed no defaults.

As with ``attr.s()``: ``attr.ib()`` is not going anywhere.
`669 <https://github.com/python-attrs/attrs/issues/669>`_


----

19.3.0

-------------------

Changes
^^^^^^^

- Fixed ``auto_attribs`` usage when default values cannot be compared directly with ``==``, such as ``numpy`` arrays.
`585 <https://github.com/python-attrs/attrs/issues/585>`_


----

19.2.0

-------------------

Backward-incompatible Changes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

- Removed deprecated ``Attribute`` attribute ``convert`` per scheduled removal on 2019/1.
This planned deprecation is tracked in issue `307 <https://github.com/python-attrs/attrs/issues/307>`_.
`504 <https://github.com/python-attrs/attrs/issues/504>`_
- ``__lt__``, ``__le__``, ``__gt__``, and ``__ge__`` do not consider subclasses comparable anymore.

This has been deprecated since 18.2.0 and was raising a ``DeprecationWarning`` for over a year.
`570 <https://github.com/python-attrs/attrs/issues/570>`_


Deprecations
^^^^^^^^^^^^

- The ``cmp`` argument to ``attr.s()`` and ``attr.ib()`` is now deprecated.

Please use ``eq`` to add equality methods (``__eq__`` and ``__ne__``) and ``order`` to add ordering methods (``__lt__``, ``__le__``, ``__gt__``, and ``__ge__``) instead – just like with `dataclasses <https://docs.python.org/3/library/dataclasses.html>`_.

Both are effectively ``True`` by default but it's enough to set ``eq=False`` to disable both at once.
Passing ``eq=False, order=True`` explicitly will raise a ``ValueError`` though.

Since this is arguably a deeper backward-compatibility break, it will have an extended deprecation period until 2021-06-01.
After that day, the ``cmp`` argument will be removed.

``attr.Attribute`` also isn't orderable anymore.
`574 <https://github.com/python-attrs/attrs/issues/574>`_


Changes
^^^^^^^

- Updated ``attr.validators.__all__`` to include new validators added in `425`_.
`517 <https://github.com/python-attrs/attrs/issues/517>`_
- Slotted classes now use a pure Python mechanism to rewrite the ``__class__`` cell when rebuilding the class, so ``super()`` works even on environments where ``ctypes`` is not installed.
`522 <https://github.com/python-attrs/attrs/issues/522>`_
- When collecting attributes using ``attr.s(auto_attribs=True)``, attributes with a default of ``None`` are now deleted too.
`523 <https://github.com/python-attrs/attrs/issues/523>`_,
`556 <https://github.com/python-attrs/attrs/issues/556>`_
- Fixed ``attr.validators.deep_iterable()`` and ``attr.validators.deep_mapping()`` type stubs.
`533 <https://github.com/python-attrs/attrs/issues/533>`_
- ``attr.validators.is_callable()`` validator now raises an exception ``attr.exceptions.NotCallableError``, a subclass of ``TypeError``, informing the received value.
`536 <https://github.com/python-attrs/attrs/issues/536>`_
- ``attr.s(auto_exc=True)`` now generates classes that are hashable by ID, as the documentation always claimed it would.
`543 <https://github.com/python-attrs/attrs/issues/543>`_,
`563 <https://github.com/python-attrs/attrs/issues/563>`_
- Added ``attr.validators.matches_re()`` that checks string attributes whether they match a regular expression.
`552 <https://github.com/python-attrs/attrs/issues/552>`_
- Keyword-only attributes (``kw_only=True``) and attributes that are excluded from the ``attrs``'s ``__init__`` (``init=False``) now can appear before mandatory attributes.
`559 <https://github.com/python-attrs/attrs/issues/559>`_
- The fake filename for generated methods is now more stable.
It won't change when you restart the process.
`560 <https://github.com/python-attrs/attrs/issues/560>`_
- The value passed to ``attr.ib(repr=…)`` can now be either a boolean (as before) or a callable.
That callable must return a string and is then used for formatting the attribute by the generated ``__repr__()`` method.
`568 <https://github.com/python-attrs/attrs/issues/568>`_
- Added ``attr.__version_info__`` that can be used to reliably check the version of ``attrs`` and write forward- and backward-compatible code.
Please check out the `section on deprecated APIs <http://www.attrs.org/en/stable/api.htmldeprecated-apis>`_ on how to use it.
`580 <https://github.com/python-attrs/attrs/issues/580>`_

.. _`425`: https://github.com/python-attrs/attrs/issues/425


----

19.1.0

-------------------

Backward-incompatible Changes
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

- Fixed a bug where deserialized objects with ``cache_hash=True`` could have incorrect hash code values.
This change breaks classes with ``cache_hash=True`` when a custom ``__setstate__`` is present.
An exception will be thrown when applying the ``attrs`` annotation to such a class.
This limitation is tracked in issue `494 <https://github.com/python-attrs/attrs/issues/494>`_.
`482 <https://github.com/python-attrs/attrs/issues/482>`_


Changes
^^^^^^^

- Add ``is_callable``, ``deep_iterable``, and ``deep_mapping`` validators.

* ``is_callable``: validates that a value is callable
* ``deep_iterable``: Allows recursion down into an iterable,
 applying another validator to every member in the iterable
 as well as applying an optional validator to the iterable itself.
* ``deep_mapping``: Allows recursion down into the items in a mapping object,
 applying a key validator and a value validator to the key and value in every item.
 Also applies an optional validator to the mapping object itself.

You can find them in the ``attr.validators`` package.
`425`_
- Fixed stub files to prevent errors raised by mypy's ``disallow_any_generics = True`` option.
`443 <https://github.com/python-attrs/attrs/issues/443>`_
- Attributes with ``init=False`` now can follow after ``kw_only=True`` attributes.
`450 <https://github.com/python-attrs/attrs/issues/450>`_
- ``attrs`` now has first class support for defining exception classes.

If you define a class using ``attr.s(auto_exc=True)`` and subclass an exception, the class will behave like a well-behaved exception class including an appropriate ``__str__`` method, and all attributes additionally available in an ``args`` attribute.
`500 <https://github.com/python-attrs/attrs/issues/500>`_
- Clarified documentation for hashing to warn that hashable objects should be deeply immutable (in their usage, even if this is not enforced).
`503 <https://github.com/python-attrs/attrs/issues/503>`_


----

18.2.0

-------------------

Deprecations
^^^^^^^^^^^^

- Comparing subclasses using ``<``, ``>``, ``<=``, and ``>=`` is now deprecated.
The docs always claimed that instances are only compared if the types are identical, so this is a first step to conform to the docs.

Equality operators (``==`` and ``!=``) were always strict in this regard.
`394 <https://github.com/python-attrs/attrs/issues/394>`_


Changes
^^^^^^^

- ``attrs`` now ships its own `PEP 484 <https://www.python.org/dev/peps/pep-0484/>`_ type hints.
Together with `mypy <http://mypy-lang.org>`_'s ``attrs`` plugin, you've got all you need for writing statically typed code in both Python 2 and 3!

At that occasion, we've also added `narrative docs <https://www.attrs.org/en/stable/types.html>`_ about type annotations in ``attrs``.
`238 <https://github.com/python-attrs/attrs/issues/238>`_
- Added *kw_only* arguments to ``attr.ib`` and ``attr.s``, and a corresponding *kw_only* attribute to ``attr.Attribute``.
This change makes it possible to have a generated ``__init__`` with keyword-only arguments on Python 3, relaxing the required ordering of default and non-default valued attributes.
`281 <https://github.com/python-attrs/attrs/issues/281>`_,
`411 <https://github.com/python-attrs/attrs/issues/411>`_
- The test suite now runs with ``hypothesis.HealthCheck.too_slow`` disabled to prevent CI breakage on slower computers.
`364 <https://github.com/python-attrs/attrs/issues/364>`_,
`396 <https://github.com/python-attrs/attrs/issues/396>`_
- ``attr.validators.in_()`` now raises a ``ValueError`` with a useful message even if the options are a string and the value is not a string.
`383 <https://github.com/python-attrs/attrs/issues/383>`_
- ``attr.asdict()`` now properly handles deeply nested lists and dictionaries.
`395 <https://github.com/python-attrs/attrs/issues/395>`_
- Added ``attr.converters.default_if_none()`` that allows to replace ``None`` values in attributes.
For example ``attr.ib(converter=default_if_none(""))`` replaces ``None`` by empty strings.
`400 <https://github.com/python-attrs/attrs/issues/400>`_,
`414 <https://github.com/python-attrs/attrs/issues/414>`_
- Fixed a reference leak where the original class would remain live after being replaced when ``slots=True`` is set.
`407 <https://github.com/python-attrs/attrs/issues/407>`_
- Slotted classes can now be made weakly referenceable by passing ``attr.s(weakref_slot=True)``.
`420 <https://github.com/python-attrs/attrs/issues/420>`_
- Added *cache_hash* option to ``attr.s`` which causes the hash code to be computed once and stored on the object.
`426 <https://github.com/python-attrs/attrs/issues/426>`_
- Attributes can be named ``property`` and ``itemgetter`` now.
`430 <https://github.com/python-attrs/attrs/issues/430>`_
- It is now possible to override a base class' class variable using only class annotations.
`431 <https://github.com/python-attrs/attrs/issues/431>`_


----

Links

• PyPI: https://pypi.org/project/attrs
• Changelog: https://pyup.io/changelogs/attrs/
• Homepage: https://www.attrs.org/

Update cassandra-driver from 3.14.0 to 3.24.0.

The bot wasn't able to find a changelog for this release. Got an idea?

Links

• PyPI: https://pypi.org/project/cassandra-driver
• Repo: http://github.com/datastax/python-driver

Update coverage from 4.5.1 to 5.3.

Changelog

5.3

--------------------------

- The ``source`` setting has always been interpreted as either a file path or a
module, depending on which existed.  If both interpretations were valid, it
was assumed to be a file path.  The new ``source_pkgs`` setting can be used
to name a package to disambiguate this case.  Thanks, Thomas Grainger. Fixes
`issue 268`_.

- If a plugin was disabled due to an exception, we used to still try to record
its information, causing an exception, as reported in `issue 1011`_.  This is
now fixed.

.. _issue 268: https://github.com/nedbat/coveragepy/issues/268
.. _issue 1011: https://github.com/nedbat/coveragepy/issues/1011


.. _changes_521:

5.2.1

----------------------------

- The dark mode HTML report still used light colors for the context listing,
making them unreadable (`issue 1009`_).  This is now fixed.

- The time stamp on the HTML report now includes the time zone. Thanks, Xie
Yanbo (`pull request 960`_).

.. _pull request 960: https://github.com/nedbat/coveragepy/pull/960
.. _issue 1009: https://github.com/nedbat/coveragepy/issues/1009


.. _changes_52:

5.2

--------------------------

- The HTML report has been redesigned by Vince Salvino.  There is now a dark
mode, the code text is larger, and system sans serif fonts are used, in
addition to other small changes (`issue 858`_ and `pull request 931`_).

- The ``coverage report`` and ``coverage html`` commands now accept a
``--precision`` option to control the number of decimal points displayed.
Thanks, Teake Nutma (`pull request 982`_).

- The ``coverage report`` and ``coverage html`` commands now accept a
``--no-skip-covered`` option to negate ``--skip-covered``.  Thanks, Anthony
Sottile (`issue 779`_ and `pull request 932`_).

- The ``--skip-empty`` option is now available for the XML report, closing
`issue 976`_.

- The ``coverage report`` command now accepts a ``--sort`` option to specify
how to sort the results.  Thanks, Jerin Peter George (`pull request 1005`_).

- If coverage fails due to the coverage total not reaching the ``--fail-under``
value, it will now print a message making the condition clear.  Thanks,
Naveen Yadav (`pull request 977`_).

- TOML configuration files with non-ASCII characters would cause errors on
Windows (`issue 990`_).  This is now fixed.

- The output of ``--debug=trace`` now includes information about how the
``--source`` option is being interpreted, and the module names being
considered.

.. _pull request 931: https://github.com/nedbat/coveragepy/pull/931
.. _pull request 932: https://github.com/nedbat/coveragepy/pull/932
.. _pull request 977: https://github.com/nedbat/coveragepy/pull/977
.. _pull request 982: https://github.com/nedbat/coveragepy/pull/982
.. _pull request 1005: https://github.com/nedbat/coveragepy/pull/1005
.. _issue 779: https://github.com/nedbat/coveragepy/issues/779
.. _issue 858: https://github.com/nedbat/coveragepy/issues/858
.. _issue 976: https://github.com/nedbat/coveragepy/issues/976
.. _issue 990: https://github.com/nedbat/coveragepy/issues/990


.. _changes_51:

5.1

--------------------------

- The JSON report now includes counts of covered and missing branches. Thanks,
Salvatore Zagaria.

- On Python 3.8, try-finally-return reported wrong branch coverage with
decorated async functions (`issue 964`_).  This is now fixed. Thanks, Kjell
Braden.

- The :meth:`~coverage.Coverage.get_option` and
:meth:`~coverage.Coverage.set_option` methods can now manipulate the
``[paths]`` configuration setting.  Thanks to Bernát Gábor for the fix for
`issue 967`_.

.. _issue 964: https://github.com/nedbat/coveragepy/issues/964
.. _issue 967: https://github.com/nedbat/coveragepy/issues/967


.. _changes_504:

5.0.4

----------------------------

- If using the ``[run] relative_files`` setting, the XML report will use
relative files in the ``<source>`` elements indicating the location of source
code.  Closes `issue 948`_.

- The textual summary report could report missing lines with negative line
numbers on PyPy3 7.1 (`issue 943`_).  This is now fixed.

- Windows wheels for Python 3.8 were incorrectly built, but are now fixed.
(`issue 949`_)

- Updated Python 3.9 support to 3.9a4.

- HTML reports couldn't be sorted if localStorage wasn't available. This is now
fixed: sorting works even though the sorting setting isn't retained. (`issue
944`_ and `pull request 945`_). Thanks, Abdeali Kothari.

.. _issue 943: https://github.com/nedbat/coveragepy/issues/943
.. _issue 944: https://github.com/nedbat/coveragepy/issues/944
.. _pull request 945: https://github.com/nedbat/coveragepy/pull/945
.. _issue 948: https://github.com/nedbat/coveragepy/issues/948
.. _issue 949: https://github.com/nedbat/coveragepy/issues/949


.. _changes_503:

5.0.3

----------------------------

- A performance improvement in 5.0.2 didn't work for test suites that changed
directory before combining data, causing "Couldn't use data file: no such
table: meta" errors (`issue 916`_).  This is now fixed.

- Coverage could fail to run your program with some form of "ModuleNotFound" or
"ImportError" trying to import from the current directory. This would happen
if coverage had been packaged into a zip file (for example, on Windows), or
was found indirectly (for example, by pyenv-virtualenv).  A number of
different scenarios were described in `issue 862`_ which is now fixed.  Huge
thanks to Agbonze O. Jeremiah for reporting it, and Alexander Waters and
George-Cristian Bîrzan for protracted debugging sessions.

- Added the "premain" debug option.

- Added SQLite compile-time options to the "debug sys" output.

.. _issue 862: https://github.com/nedbat/coveragepy/issues/862
.. _issue 916: https://github.com/nedbat/coveragepy/issues/916


.. _changes_502:

5.0.2

----------------------------

- Programs that used multiprocessing and changed directories would fail under
coverage.  This is now fixed (`issue 890`_).  A side effect is that debug
information about the config files read now shows absolute paths to the
files.

- When running programs as modules (``coverage run -m``) with ``--source``,
some measured modules were imported before coverage starts.  This resulted in
unwanted warnings ("Already imported a file that will be measured") and a
reduction in coverage totals (`issue 909`_).  This is now fixed.

- If no data was collected, an exception about "No data to report" could happen
instead of a 0% report being created (`issue 884`_).  This is now fixed.

- The handling of source files with non-encodable file names has changed.
Previously, if a file name could not be encoded as UTF-8, an error occurred,
as described in `issue 891`_.  Now, those files will not be measured, since
their data would not be recordable.

- A new warning ("dynamic-conflict") is issued if two mechanisms are trying to
change the dynamic context.  Closes `issue 901`_.

- ``coverage run --debug=sys`` would fail with an AttributeError. This is now
fixed (`issue 907`_).

.. _issue 884: https://github.com/nedbat/coveragepy/issues/884
.. _issue 890: https://github.com/nedbat/coveragepy/issues/890
.. _issue 891: https://github.com/nedbat/coveragepy/issues/891
.. _issue 901: https://github.com/nedbat/coveragepy/issues/901
.. _issue 907: https://github.com/nedbat/coveragepy/issues/907
.. _issue 909: https://github.com/nedbat/coveragepy/issues/909


.. _changes_501:

5.0.1

----------------------------

- If a 4.x data file is the cause of a "file is not a database" error, then use
a more specific error message, "Looks like a coverage 4.x data file, are you
mixing versions of coverage?"  Helps diagnose the problems described in
`issue 886`_.

- Measurement contexts and relative file names didn't work together, as
reported in `issue 899`_ and `issue 900`_.  This is now fixed, thanks to
David Szotten.

- When using ``coverage run --concurrency=multiprocessing``, all data files
should be named with parallel-ready suffixes.  5.0 mistakenly named the main
process' file with no suffix when using ``--append``.  This is now fixed,
closing `issue 880`_.

- Fixed a problem on Windows when the current directory is changed to a
different drive (`issue 895`_).  Thanks, Olivier Grisel.

- Updated Python 3.9 support to 3.9a2.

.. _issue 880: https://github.com/nedbat/coveragepy/issues/880
.. _issue 886: https://github.com/nedbat/coveragepy/issues/886
.. _issue 895: https://github.com/nedbat/coveragepy/issues/895
.. _issue 899: https://github.com/nedbat/coveragepy/issues/899
.. _issue 900: https://github.com/nedbat/coveragepy/issues/900


.. _changes_50:

5.0

--------------------------

Nothing new beyond 5.0b2.


.. _changes_50b2:

5.0b2

----------------------------

- An experimental ``[run] relative_files`` setting tells coverage to store
relative file names in the data file. This makes it easier to run tests in
one (or many) environments, and then report in another.  It has not had much
real-world testing, so it may change in incompatible ways in the future.

- When constructing a :class:`coverage.Coverage` object, `data_file` can be
specified as None to prevent writing any data file at all.  In previous
versions, an explicit `data_file=None` argument would use the default of
".coverage". Fixes `issue 871`_.

- Python files run with ``-m`` now have ``__spec__`` defined properly.  This
fixes `issue 745`_ (about not being able to run unittest tests that spawn
subprocesses), and `issue 838`_, which described the problem directly.

- The ``[paths]`` configuration section is now ordered. If you specify more
than one list of patterns, the first one that matches will be used.  Fixes
`issue 649`_.

- The :func:`.coverage.numbits.register_sqlite_functions` function now also
registers `numbits_to_nums` for use in SQLite queries.  Thanks, Simon
Willison.

- Python 3.9a1 is supported.

- Coverage.py has a mascot: :ref:`Sleepy Snake <sleepy>`.

.. _issue 649: https://github.com/nedbat/coveragepy/issues/649
.. _issue 745: https://github.com/nedbat/coveragepy/issues/745
.. _issue 838: https://github.com/nedbat/coveragepy/issues/838
.. _issue 871: https://github.com/nedbat/coveragepy/issues/871


.. _changes_50b1:

5.0b1

----------------------------

- The HTML and textual reports now have a ``--skip-empty`` option that skips
files with no statements, notably ``__init__.py`` files.  Thanks, Reya B.

- Configuration can now be read from `TOML`_ files.  This requires installing
coverage.py with the ``[toml]`` extra.  The standard "pyproject.toml" file
will be read automatically if no other configuration file is found, with
settings in the ``[tool.coverage.]`` namespace.  Thanks to Frazer McLean for
implementation and persistence.  Finishes `issue 664`_.

- The ``[run] note`` setting has been deprecated. Using it will result in a
warning, and the note will not be written to the data file.  The
corresponding :class:`.CoverageData` methods have been removed.

- The HTML report has been reimplemented (no more table around the source
code). This allowed for a better presentation of the context information,
hopefully resolving `issue 855`_.

- Added sqlite3 module version information to ``coverage debug sys`` output.

- Asking the HTML report to show contexts (``[html] show_contexts=True`` or
``coverage html --show-contexts``) will issue a warning if there were no
contexts measured (`issue 851`_).

.. _TOML: https://github.com/toml-lang/tomlreadme
.. _issue 664: https://github.com/nedbat/coveragepy/issues/664
.. _issue 851: https://github.com/nedbat/coveragepy/issues/851
.. _issue 855: https://github.com/nedbat/coveragepy/issues/855


.. _changes_50a8:

5.0a8

----------------------------

- The :class:`.CoverageData` API has changed how queries are limited to
specific contexts.  Now you use :meth:`.CoverageData.set_query_context` to
set a single exact-match string, or :meth:`.CoverageData.set_query_contexts`
to set a list of regular expressions to match contexts.  This changes the
command-line ``--contexts`` option to use regular expressions instead of
filename-style wildcards.


.. _changes_50a7:

5.0a7

----------------------------

- Data can now be "reported" in JSON format, for programmatic use, as requested
in `issue 720`_.  The new ``coverage json`` command writes raw and summarized
data to a JSON file.  Thanks, Matt Bachmann.

- Dynamic contexts are now supported in the Python tracer, which is important
for PyPy users.  Closes `issue 846`_.

- The compact line number representation introduced in 5.0a6 is called a
"numbits."  The :mod:`coverage.numbits` module provides functions for working
with them.

- The reporting methods used to permanently apply their arguments to the
configuration of the Coverage object.  Now they no longer do.  The arguments
affect the operation of the method, but do not persist.

- A class named "test_something" no longer confuses the ``test_function``
dynamic context setting.  Fixes `issue 829`_.

- Fixed an unusual tokenizing issue with backslashes in comments.  Fixes
`issue 822`_.

- ``debug=plugin`` didn't properly support configuration or dynamic context
plugins, but now it does, closing `issue 834`_.

.. _issue 720: https://github.com/nedbat/coveragepy/issues/720
.. _issue 822: https://github.com/nedbat/coveragepy/issues/822
.. _issue 834: https://github.com/nedbat/coveragepy/issues/834
.. _issue 829: https://github.com/nedbat/coveragepy/issues/829
.. _issue 846: https://github.com/nedbat/coveragepy/issues/846


.. _changes_50a6:

5.0a6

----------------------------

- Reporting on contexts. Big thanks to Stephan Richter and Albertas Agejevas
for the contribution.

- The ``--contexts`` option is available on the ``report`` and ``html``
 commands.  It's a comma-separated list of shell-style wildcards, selecting
 the contexts to report on.  Only contexts matching one of the wildcards
 will be included in the report.

- The ``--show-contexts`` option for the ``html`` command adds context
 information to each covered line.  Hovering over the "ctx" marker at the
 end of the line reveals a list of the contexts that covered the line.

- Database changes:

- Line numbers are now stored in a much more compact way.  For each file and
 context, a single binary string is stored with a bit per line number.  This
 greatly improves memory use, but makes ad-hoc use difficult.

- Dynamic contexts with no data are no longer written to the database.

- SQLite data storage is now faster.  There's no longer a reason to keep the
 JSON data file code, so it has been removed.

- Changes to the :class:`.CoverageData` interface:

- The new :meth:`.CoverageData.dumps` method serializes the data to a string,
 and a corresponding :meth:`.CoverageData.loads` method reconstitutes this
 data.  The format of the data string is subject to change at any time, and
 so should only be used between two installations of the same version of
 coverage.py.

- The :meth:`CoverageData constructor<.CoverageData.__init__>` has a new
 argument, `no_disk` (default: False).  Setting it to True prevents writing
 any data to the disk.  This is useful for transient data objects.

- Added the classmethod :meth:`.Coverage.current` to get the latest started
Coverage instance.

- Multiprocessing support in Python 3.8 was broken, but is now fixed.  Closes
`issue 828`_.

- Error handling during reporting has changed slightly.  All reporting methods
now behave the same.  The ``--ignore-errors`` option keeps errors from
stopping the reporting, but files that couldn't parse as Python will always
be reported as warnings.  As with other warnings, you can suppress them with
the ``[run] disable_warnings`` configuration setting.

- Coverage.py no longer fails if the user program deletes its current
directory. Fixes `issue 806`_.  Thanks, Dan Hemberger.

- The scrollbar markers in the HTML report now accurately show the highlighted
lines, regardless of what categories of line are highlighted.

- The hack to accommodate ShiningPanda_ looking for an obsolete internal data
file has been removed, since ShiningPanda 0.22 fixed it four years ago.

- The deprecated `Reporter.file_reporters` property has been removed.

.. _ShiningPanda: https://wiki.jenkins.io/display/JENKINS/ShiningPanda+Plugin
.. _issue 806: https://github.com/nedbat/coveragepy/pull/806
.. _issue 828: https://github.com/nedbat/coveragepy/issues/828


.. _changes_50a5:

5.0a5

----------------------------

- Drop support for Python 3.4

- Dynamic contexts can now be set two new ways, both thanks to Justas
Sadzevičius.

- A plugin can implement a ``dynamic_context`` method to check frames for
 whether a new context should be started.  See
 :ref:`dynamic_context_plugins` for more details.

- Another tool (such as a test runner) can use the new
 :meth:`.Coverage.switch_context` method to explicitly change the context.

- The ``dynamic_context = test_function`` setting now works with Python 2
old-style classes, though it only reports the method name, not the class it
was defined on.  Closes `issue 797`_.

- ``fail_under`` values more than 100 are reported as errors.  Thanks to Mike
Fiedler for closing `issue 746`_.

- The "missing" values in the text output are now sorted by line number, so
that missing branches are reported near the other lines they affect. The
values used to show all missing lines, and then all missing branches.

- Access to the SQLite database used for data storage is now thread-safe.
Thanks, Stephan Richter. This closes `issue 702`_.

- Combining data stored in SQLite is now about twice as fast, fixing `issue
761`_.  Thanks, Stephan Richter.

- The ``filename`` attribute on :class:`.CoverageData` objects has been made
private.  You can use the ``data_filename`` method to get the actual file
name being used to store data, and the ``base_filename`` method to get the
original filename before parallelizing suffixes were added.  This is part of
fixing `issue 708`_.

- Line numbers in the HTML report now align properly with source lines, even
when Chrome's minimum font size is set, fixing `issue 748`_.  Thanks Wen Ye.

.. _issue 702: https://github.com/nedbat/coveragepy/issues/702
.. _issue 708: https://github.com/nedbat/coveragepy/issues/708
.. _issue 746: https://github.com/nedbat/coveragepy/issues/746
.. _issue 748: https://github.com/nedbat/coveragepy/issues/748
.. _issue 761: https://github.com/nedbat/coveragepy/issues/761
.. _issue 797: https://github.com/nedbat/coveragepy/issues/797


.. _changes_50a4:

5.0a4

----------------------------

- You can specify the command line to run your program with the ``[run]
command_line`` configuration setting, as requested in `issue 695`_.

- Coverage will create directories as needed for the data file if they don't
exist, closing `issue 721`_.

- The ``coverage run`` command has always adjusted the first entry in sys.path,
to properly emulate how Python runs your program.  Now this adjustment is
skipped if sys.path[0] is already different than Python's default.  This
fixes `issue 715`_.

- Improvements to context support:

- The "no such table: meta" error is fixed.: `issue 716`_.

- Combining data files is now much faster.

- Python 3.8 (as of today!) passes all tests.

.. _issue 695: https://github.com/nedbat/coveragepy/issues/695
.. _issue 715: https://github.com/nedbat/coveragepy/issues/715
.. _issue 716: https://github.com/nedbat/coveragepy/issues/716
.. _issue 721: https://github.com/nedbat/coveragepy/issues/721


.. _changes_50a3:

5.0a3

----------------------------

- Context support: static contexts let you specify a label for a coverage run,
which is recorded in the data, and retained when you combine files.  See
:ref:`contexts` for more information.

- Dynamic contexts: specifying ``[run] dynamic_context = test_function`` in the
config file will record the test function name as a dynamic context during
execution.  This is the core of "Who Tests What" (`issue 170`_).  Things to
note:

- There is no reporting support yet.  Use SQLite to query the .coverage file
 for information.  Ideas are welcome about how reporting could be extended
 to use this data.

- There's a noticeable slow-down before any test is run.

- Data files will now be roughly N times larger, where N is the number of
 tests you have.  Combining data files is therefore also N times slower.

- No other values for ``dynamic_context`` are recognized yet.  Let me know
 what else would be useful.  I'd like to use a pytest plugin to get better
 information directly from pytest, for example.

.. _issue 170: https://github.com/nedbat/coveragepy/issues/170

- Environment variable substitution in configuration files now supports two
syntaxes for controlling the behavior of undefined variables: if ``VARNAME``
is not defined, ``${VARNAME?}`` will raise an error, and ``${VARNAME-default
value}`` will use "default value".

- Partial support for Python 3.8, which has not yet released an alpha. Fixes
`issue 707`_ and `issue 714`_.

.. _issue 707: https://github.com/nedbat/coveragepy/issues/707
.. _issue 714: https://github.com/nedbat/coveragepy/issues/714


.. _changes_50a2:

5.0a2

----------------------------

- Coverage's data storage has changed.  In version 4.x, .coverage files were
basically JSON.  Now, they are SQLite databases.  This means the data file
can be created earlier than it used to.  A large amount of code was
refactored to support this change.

- Because the data file is created differently than previous releases, you
 may need ``parallel=true`` where you didn't before.

- The old data format is still available (for now) by setting the environment
 variable COVERAGE_STORAGE=json. Please tell me if you think you need to
 keep the JSON format.

- The database schema is guaranteed to change in the future, to support new
 features.  I'm looking for opinions about making the schema part of the
 public API to coverage.py or not.

- Development moved from `Bitbucket`_ to `GitHub`_.

- HTML files no longer have trailing and extra whitespace.

- The sort order in the HTML report is stored in local storage rather than
cookies, closing `issue 611`_.  Thanks, Federico Bond.

- pickle2json, for converting v3 data files to v4 data files, has been removed.

.. _Bitbucket: https://bitbucket.org
.. _GitHub: https://github.com/nedbat/coveragepy

.. _issue 611: https://github.com/nedbat/coveragepy/issues/611


.. _changes_50a1:

5.0a1

----------------------------

- Coverage.py no longer supports Python 2.6 or 3.3.

- The location of the configuration file can now be specified with a
``COVERAGE_RCFILE`` environment variable, as requested in `issue 650`_.

- Namespace packages are supported on Python 3.7, where they used to cause
TypeErrors about path being None. Fixes `issue 700`_.

- A new warning (``already-imported``) is issued if measurable files have
already been imported before coverage.py started measurement.  See
:ref:`cmd_warnings` for more information.

- Running coverage many times for small runs in a single process should be
faster, closing `issue 625`_.  Thanks, David MacIver.

- Large HTML report pages load faster.  Thanks, Pankaj Pandey.

.. _issue 625: https://github.com/nedbat/coveragepy/issues/625
.. _issue 650: https://github.com/nedbat/coveragepy/issues/650
.. _issue 700: https://github.com/nedbat/coveragepy/issues/700


.. _changes_454:

4.5.4

----------------------------

- Multiprocessing support in Python 3.8 was broken, but is now fixed.  Closes
`issue 828`_.

.. _issue 828: https://github.com/nedbat/coveragepy/issues/828


.. _changes_453:

4.5.3

----------------------------

- Only packaging metadata changes.


.. _changes_452:

4.5.2

----------------------------

- Namespace packages are supported on Python 3.7, where they used to cause
TypeErrors about path being None. Fixes `issue 700`_.

- Python 3.8 (as of today!) passes all tests.  Fixes `issue 707`_ and
`issue 714`_.

- Development moved from `Bitbucket`_ to `GitHub`_.

.. _issue 700: https://github.com/nedbat/coveragepy/issues/700
.. _issue 707: https://github.com/nedbat/coveragepy/issues/707
.. _issue 714: https://github.com/nedbat/coveragepy/issues/714

.. _Bitbucket: https://bitbucket.org
.. _GitHub: https://github.com/nedbat/coveragepy


.. _changes_451:

Links

• PyPI: https://pypi.org/project/coverage
• Changelog: https://pyup.io/changelogs/coverage/
• Repo: https://github.com/nedbat/coveragepy

Update Cython from 0.28.4 to 0.29.21.

Changelog

0.29.21

====================

Bugs fixed
----------

* Fix a regression in 0.29.20 where ``__div__`` failed to be found in extension types.
(Github issue 3688)

* Fix a regression in 0.29.20 where a call inside of a finally clause could fail to compile.
Patch by David Woods.  (Github issue 3712)

* Zero-sized buffers could fail to validate as C/Fortran-contiguous.
Patch by Clemens Hofreither.  (Github issue 2093)

* ``exec()`` did not allow recent Python syntax features in Py3.8+ due to
https://bugs.python.org/issue35975.
(Github issue 3695)

* Binding staticmethods of Cython functions were not behaving like Python methods in Py3.
Patch by Jeroen Demeyer and Michał Górny.  (Github issue 3106)

* Pythran calls to NumPy methods no longer generate useless method lookup code.

* The ``PyUnicode_GET_LENGTH()`` macro was missing from the ``cpython.*`` declarations.
Patch by Thomas Caswell.  (Github issue 3692)

* The deprecated ``PyUnicode_*()`` C-API functions are no longer used, except for Unicode
strings that contain lone surrogates.  Unicode strings that contain non-BMP characters
or surrogate pairs now generate different C code on 16-bit Python 2.x Unicode deployments
(such as MS-Windows).  Generating the C code on Python 3.x is recommended in this case.
Original patches by Inada Naoki and Victor Stinner.  (Github issues 3677, 3721, 3697)

* Some template parameters were missing from the C++ ``std::unordered_map`` declaration.
Patch by will.  (Github issue 3685)

* Several internal code generation issues regarding temporary variables were resolved.
(Github issue 3708)

0.29.20

====================

Bugs fixed
----------

* Nested try-except statements with multiple ``return`` statements could crash
due to incorrect deletion of the ``except as`` target variable.
(Github issue 3666)

* The ``classmethod`` decorator no longer rejects unknown input from other decorators.
Patch by David Woods.  (Github issue 3660)

* Fused types could leak into unrelated usages.
Patch by David Woods.  (Github issue 3642)

* Now uses ``Py_SET_SIZE()`` and ``Py_SET_REFCNT()`` in Py3.9+ to avoid low-level
write access to these object fields.
Patch by Victor Stinner.  (Github issue 3639)

* The built-in ``abs()`` function could lead to undefined behaviour when used on
the negative-most value of a signed C integer type.
Patch by Serge Guelton.  (Github issue 1911)

* Usages of ``sizeof()`` and ``typeid()`` on uninitialised variables no longer
produce a warning.
Patch by Celelibi.  (Github issue 3575)

* The C++ ``typeid()`` function was allowed in C mode.
Patch by Celelibi.  (Github issue 3637)

* The error position reported for errors found in f-strings was misleading.
(Github issue 3674)

* The new ``c_api_binop_methods`` directive was added for forward compatibility, but can
only be set to True (the current default value).  It can be disabled in Cython 3.0.

0.29.19

====================

Bugs fixed
----------

* A typo in Windows specific code in 0.29.18 was fixed that broke "libc.math".
(Github issue 3622)

* A platform specific test failure in 0.29.18 was fixed.
Patch by smutch.  (Github issue 3620)

0.29.18

====================

Bugs fixed
----------

* Exception position reporting could run into race conditions on threaded code.
It now uses function-local variables again.

* Error handling early in the module init code could lead to a crash.

* Error handling in ``cython.array`` creation was improved to avoid calling
C-API functions with an error held.

* A memory corruption was fixed when garbage collection was triggered during calls
to ``PyType_Ready()`` of extension type subclasses.
(Github issue 3603)

* Memory view slicing generated unused error handling code which could negatively
impact the C compiler optimisations for parallel OpenMP code etc.  Also, it is
now helped by static branch hints.
(Github issue 2987)

* Cython's built-in OpenMP functions were not translated inside of call arguments.
Original patch by Celelibi and David Woods.  (Github issue 3594)

* Complex buffer item types of structs of arrays could fail to validate.
Patch by Leo and smutch.  (Github issue 1407)

* Decorators were not allowed on nested `async def` functions.
(Github issue 1462)

* C-tuples could use invalid C struct casting.
Patch by MegaIng.  (Github issue 3038)

* Optimised ``%d`` string formatting into f-strings failed on float values.
(Github issue 3092)

* Optimised aligned string formatting (``%05s``, ``%-5s``) failed.
(Github issue 3476)

* When importing the old Cython ``build_ext`` integration with distutils, the
additional command line arguments leaked into the regular command.
Patch by Kamekameha.  (Github issue 2209)

* When using the ``CYTHON_NO_PYINIT_EXPORT`` option in C++, the module init function
was not declared as ``extern "C"``.
(Github issue 3414)

* Three missing timedelta access macros were added in ``cpython.datetime``.

* The signature of the NumPy C-API function ``PyArray_SearchSorted()`` was fixed.
Patch by Brock Mendel.  (Github issue 3606)

0.29.17

====================

Features added
--------------

* ``std::move()`` is now available from ``libcpp.utility``.
Patch by Omer Ozarslan.  (Github issue 2169)

* The ``cython.binding`` decorator is available in Python code.
(Github issue 3505)

Bugs fixed
----------

* Creating an empty unicode slice with large bounds could crash.
Patch by Sam Sneddon.  (Github issue 3531)

* Decoding an empty bytes/char* slice with large bounds could crash.
Patch by Sam Sneddon.  (Github issue 3534)

* Re-importing a Cython extension no longer raises the error
"``__reduce_cython__ not found``".
(Github issue 3545)

* Unused C-tuples could generate incorrect code in 0.29.16.
Patch by Kirk Meyer.  (Github issue 3543)

* Creating a fused function attached it to the garbage collector before it
was fully initialised, thus risking crashes in rare failure cases.
Original patch by achernomorov.  (Github issue 3215)

* Temporary buffer indexing variables were not released and could show up in
C compiler warnings, e.g. in generators.
Patch by David Woods.  (Github issues 3430, 3522)

* The compilation cache in ``cython.inline("…")`` failed to take the language
level into account.
Patch by will-ca.  (Github issue 3419)

* The deprecated ``PyUnicode_GET_SIZE()`` function is no longer used in Py3.

0.29.16

====================

Bugs fixed
----------

* Temporary internal variables in nested prange loops could leak into other
threads.  Patch by Frank Schlimbach.  (Github issue 3348)

* Default arguments on fused functions could crash.
Patch by David Woods.  (Github issue 3370)

* C-tuples declared in ``.pxd`` files could generate incomplete C code.
Patch by Kirk Meyer.  (Github issue 1427)

* Fused functions were not always detected and optimised as Cython
implemented functions.
Patch by David Woods.  (Github issue 3384)

* Valid Python object concatenation of (iterable) strings to non-strings
could fail with an exception.
Patch by David Woods.  (Github issue 3433)

* Using C functions as temporary values lead to invalid C code.
Original patch by David Woods.  (Github issue 3418)

* Fix an unhandled C++ exception in comparisons.
Patch by David Woods.  (Github issue 3361)

* Fix deprecated import of "imp" module.
Patch by Matti Picus.  (Github issue 3350)

* Fix compatibility with Pythran 0.9.6 and later.
Patch by Serge Guelton.  (Github issue 3308)

* The ``_Py_PyAtExit()`` function in ``cpython.pylifecycle`` was misdeclared.
Patch by Zackery Spytz.  (Github issue 3382)

* Several missing declarations in ``cpython.*`` were added.
Patches by Zackery Spytz.  (Github issue 3452, 3421, 3411, 3402)

* A declaration for ``libc.math.fpclassify()`` was added.
Patch by Zackery Spytz.  (Github issue 2514)

* Avoid "undeclared" warning about automatically generated pickle methods.
Patch by David Woods.  (Github issue 3353)

* Avoid C compiler warning about unreachable code in ``prange()``.

* Some C compiler warnings in PyPy were resolved.
Patch by Matti Picus.  (Github issue 3437)

0.29.15

====================

Bugs fixed
----------

* Crash when returning a temporary Python object from an async-def function.
(Github issue 3337)

* Crash when using ``**kwargs`` in generators.
Patch by David Woods.  (Github issue 3265)

* Double reference free in ``__class__`` cell handling for ``super()`` calls.
(Github issue 3246)

* Compile error when using ``*args`` as Python class bases.
(Github issue 3338)

* Import failure in IPython 7.11.
(Github issue 3297)

* Fixed C name collision in the auto-pickle code.
Patch by ThePrez.  (Github issue 3238)

* Deprecated import failed in Python 3.9.
(Github issue 3266)

0.29.14

====================

Bugs fixed
----------

* The generated code failed to initialise the ``tp_print`` slot in CPython 3.8.
Patches by Pablo Galindo and Orivej Desh.  (Github issues 3171, 3201)

* ``?`` for ``bool`` was missing from the supported NumPy dtypes.
Patch by Max Klein.  (Github issue 2675)

* ``await`` was not allowed inside of f-strings.
Patch by Dmitro Getz.  (Github issue 2877)

* Coverage analysis failed for projects where the code resides in separate
source sub-directories.
Patch by Antonio Valentino.  (Github issue 1985)

* An incorrect compiler warning was fixed in automatic C++ string conversions.
Patch by Gerion Entrup.  (Github issue 3108)

* Error reports in the Jupyter notebook showed unhelpful stack traces.
Patch by Matthew Edwards (Github issue 3196).

* ``Python.h`` is now also included explicitly from ``public`` header files.
(Github issue 3133).

* Distutils builds with ``--parallel`` did not work when using Cython's
deprecated ``build_ext`` command.
Patch by Alphadelta14 (Github issue 3187).

Other changes
-------------

* The ``PyMemoryView_*()`` C-API is available in ``cpython.memoryview``.
Patch by Nathan Manville.  (Github issue 2541)

0.29.13

====================

Bugs fixed
----------

* A reference leak for ``None`` was fixed when converting a memoryview
to a Python object.  (Github issue 3023)

* The declaration of ``PyGILState_STATE`` in ``cpython.pystate`` was unusable.
Patch by Kirill Smelkov.  (Github issue 2997)

Other changes
-------------

* The declarations in ``posix.mman`` were extended.
Patches by Kirill Smelkov.  (Github issues 2893, 2894, 3012)

0.29.12

====================

Bugs fixed
----------

* Fix compile error in CPython 3.8b2 regarding the ``PyCode_New()`` signature.
(Github issue 3031)

* Fix a C compiler warning about a missing ``int`` downcast.
(Github issue 3028)

* Fix reported error positions of undefined builtins and constants.
Patch by Orivej Desh.  (Github issue 3030)

* A 32 bit issue in the Pythran support was resolved.
Patch by Serge Guelton.  (Github issue 3032)

0.29.11

====================

Bugs fixed
----------

* Fix compile error in CPython 3.8b2 regarding the ``PyCode_New()`` signature.
Patch by Nick Coghlan. (Github issue 3009)

* Invalid C code generated for lambda functions in cdef methods.
Patch by Josh Tobin.  (Github issue 2967)

* Support slice handling in newer Pythran versions.
Patch by Serge Guelton.  (Github issue 2989)

* A reference leak in power-of-2 calculation was fixed.
Patch by Sebastian Berg.  (Github issue 3022)

* The search order for include files was changed. Previously it was
``include_directories``, ``Cython/Includes``, ``sys.path``. Now it is
``include_directories``, ``sys.path``, ``Cython/Includes``. This was done to
allow third-party ``*.pxd`` files to override the ones in Cython.
Original patch by Matti Picus.  (Github issue 2905)

* Setting ``language_level=2`` in a file did not work if ``language_level=3``
was enabled globally before.
Patch by Jeroen Demeyer.  (Github issue 2791)

0.29.10

====================

Bugs fixed
----------

* Fix compile errors in CPython 3.8b1 due to the new "tp_vectorcall" slots.
(Github issue 2976)

0.29.9

===================

Bugs fixed
----------

* Fix a crash regression in 0.29.8 when creating code objects fails.

* Remove an incorrect cast when using true-division in C++ operations.
(Github issue 1950)

0.29.8

===================

Bugs fixed
----------

* C compile errors with CPython 3.8 were resolved.
Patch by Marcel Plch.  (Github issue 2938)

* Python tuple constants that compare equal but have different item
types could incorrectly be merged into a single constant.
(Github issue 2919)

* Non-ASCII characters in unprefixed strings could crash the compiler when
used with language level ``3str``.

* Starred expressions in %-formatting tuples could fail to compile for
unicode strings.  (Github issue 2939)

* Passing Python class references through ``cython.inline()`` was broken.
(Github issue 2936)

0.29.7

===================

Bugs fixed
----------

* Crash when the shared Cython config module gets unloaded and another Cython
module reports an exceptions.  Cython now makes sure it keeps an owned reference
to the module.
(Github issue 2885)

* Resolved a C89 compilation problem when enabling the fast-gil sharing feature.

* Coverage reporting did not include the signature line of ``cdef`` functions.
(Github issue 1461)

* Casting a GIL-requiring function into a nogil function now issues a warning.
(Github issue 2879)

* Generators and coroutines were missing their return type annotation.
(Github issue 2884)

0.29.6

===================

Bugs fixed
----------

* Fix a crash when accessing the ``__kwdefaults__`` special attribute of
fused functions.  (Github issue 1470)

* Fix the parsing of buffer format strings that contain numeric sizes, which
could lead to incorrect input rejections.  (Github issue 2845)

* Avoid a C pragma in old gcc versions that was only added in GCC 4.6.
Patch by Michael Anselmi.  (Github issue 2838)

* Auto-encoding of Unicode strings to UTF-8 C/C++ strings failed in Python 3,
even though the default encoding there is UTF-8.
(Github issue 2819)

0.29.5

===================

Bugs fixed
----------

* Crash when defining a Python subclass of an extension type and repeatedly calling
a cpdef method on it.  (Github issue 2823)

* Compiler crash when ``prange()`` loops appear inside of with-statements.
(Github issue 2780)

* Some C compiler warnings were resolved.
Patches by Christoph Gohlke.  (Github issues 2815, 2816, 2817, 2822)

* Python conversion of C++ enums failed in 0.29.
Patch by Orivej Desh.  (Github issue 2767)

0.29.4

===================

Bugs fixed
----------

* Division of numeric constants by a runtime value of 0 could fail to raise a
``ZeroDivisionError``.  (Github issue 2820)

0.29.3

===================

Bugs fixed
----------

* Some C code for memoryviews was generated in a non-deterministic order.
Patch by Martijn van Steenbergen.  (Github issue 2779)

* C89 compatibility was accidentally lost since 0.28.
Patches by gastineau and true-pasky.  (Github issues 2778, 2801)

* A C compiler cast warning was resolved.
Patch by Michael Buesch.  (Github issue 2774)

* An compilation failure with complex numbers under MSVC++ was resolved.
(Github issue 2797)

* Coverage reporting could fail when modules were moved around after the build.
Patch by Wenjun Si.  (Github issue 2776)

0.29.2

===================

Bugs fixed
----------

* The code generated for deduplicated constants leaked some references.
(Github issue 2750)

* The declaration of ``sigismember()`` in ``libc.signal`` was corrected.
(Github issue 2756)

* Crashes in compiler and test runner were fixed.
(Github issue 2736, 2755)

* A C compiler warning about an invalid safety check was resolved.
(Github issue 2731)

0.29.1

===================

Bugs fixed
----------

* Extensions compiled with MinGW-64 under Windows could misinterpret integer
objects larger than 15 bit and return incorrect results.
(Github issue 2670)

* Cython no longer requires the source to be writable when copying its data
into a memory view slice.
Patch by An