This test is unstable and currently prevents to make any new change
since the test always fails on Travis CI.

Skip the test to get more time to fix it.

socketserver.ForkingMixIn.server_close() and
socketserver.ThreadingMixIn.server_close() now waits until all child
processes and non-daemonic threads complete.

The GIL is no longer created "on demand" to fix a race condition when
PyGILState_Ensure() is called in a non-Python thread.

To mitigate the situation when the buildbot is under load
and is unable to send/receive data fast enough:

* reduce the size of the payload
* set a generous timeout for socket ops

Add os.posix_spawn to wrap the low level POSIX API of the same name.

Contributed by Pablo Galindo.

Document the error return of PyLong_As* APIs.

A frequent Python C API usage error is neglecting to check the return
value and/or PyErr_Occurred().

This reverts commit 47c0b1f7.

Definition order of kwonly params is now guaranteed preserved.

I've run some tests on 3.7 asyncio and it appears that too many
things assume that run_in_executor returns a Future.

Ensure that ``truncate()`` preserves the file position (as reported by ``tell()``) after writes longer than the buffer size. 

It will be triggered when propagating an exception.

* bpo-32659: Solaris "stat" should support "st_fstype"

* Add 'versionadded'

Better account for single-line compound statements and
semi-colon separated statements when suggesting
Py3 replacements for Py2 print statements.

Initial patch by Nitish Chandra.

Updating the C API docs was missed when the per-opcode
tracing & profiling support was initially added.

The arguments to a generator function which is declared as a
contextmanager are stored inside the context manager, and
thus are kept alive, even when it is used as a regular context
manager, and not as a function decorator (where it needs
the original arguments to recreate the generator on each
call).

This is a possible unnecessary memory leak, so this changes
contextmanager.__enter__ to release the saved arguments,
as that method being called means that particular CM instance
isn't going to need to recreate the underlying generator.

Patch by Martin Teichmann.

Add PEP 557 Data CLasses to What's New.

Class authors no longer need to specify repr=False if they want to provide a custom __repr__ for dataclasses. The same thing applies for the other dunder methods that the dataclass decorator adds. If dataclass finds that a dunder methods is defined in the class, it will not overwrite it.

On mac, sendfile throws ENOTCONN on a repeated sendfile call if
the connection is closed. Normalize it to behave like other systems.