This syncs to rev 7b43ada77821d23e55e3a4b35f6055a59b9e1ad7 there.

Summary:

- Add typing.DefaultDict (as a generic variant of collections.defaultdict).

- Use collections.Reversible if it exists (only relevant for Python 3.6).

- Revamped generic class behavior to conform to updated PEP 484.

- Improve speed of Generic.__new__.

- Make sure __init__ is called for new Generic instances. Fix issue #26391.

- Refactor async support to be compatible with 3.2, 3.3, 3.4.

- Remove 'io' and 're' from __all__ (they still exist, just not
  included by "import *"). Fix issue #26234.

- Change @overload -- you can now use it outside stubs (you still
  cannot call the decorated function though).

* Group functions into six new subsections
* Document the underlying Readline function or variable accessed
* get_history_length() returns the history file limit
* clear_history() is conditionally compiled in
* Clarify zero and one bases for history item indexes
* parse_and_bind() uses its argument directly as an init line
* Change "command line" to "line buffer" for consistency
* read_init_file() also executes the file
* read_history_file() replaces the previous history
* write_history_file() overwrites any existing file
* Differentiate history file lines from history list items, which could be
  multi-line
* Add more information about completion, also addressing Issue #10796
* libedit (Editline) may be used on any platform; detection is OS X specific

Patch by Robert Bachmann and SilentGhost.

Patch by Ashley Anderson.

Also fix links to the “tzinfo” class and attributes.

Issue #26509: In fatal error handlers, don't log ConnectionAbortedError which
occur on Windows.

Fix GH#325: Allow to pass None as a timeout value to disable timeout logic.

Change written by Andrew Svetlov and merged by Guido van Rossum.

On Python 3.3, use aliases:

* PyMem_RawMalloc = PyMem_Malloc
* PyMem_RawFree = PyMem_Free

These aliases are not need in Python 3.5, but this change makes synchronization
of code base simpler.

Also change BufferedReader.writable() and BufferedWriter.readable() to always
return False.

Affected classes are generic sequence iterators, iterators of str, bytes,
bytearray, list, tuple, set, frozenset, dict, OrderedDict, corresponding
views and os.scandir() iterator.

Reported by Matthew Cole on docs@p.o.

Use the default timeout of 30 seconds to avoid blocking forever.

* Fix test_os.BytesWalkTests on Windows
* Mimick better the reference os.DirEntry on Windows
* _DummyDirEntry now caches os.stat() result
* _DummyDirEntry constructor now tries to get os.stat()
* Fix os._DummyDirEntry.is_symlink(), don't follow symbolic links:
  use os.stat(path, follow_symlinks=False).

_Call is a subclass of tuple (changeset 3603bae63c13 only works
for classes) so we need to implement __ne__ ourselves.

Patch by Andrew Plummer.

If you pass -1, the callable can take any number of arguments.

Added tests to verify the behavior.

Initial patch by Cédric Krier.

The behaviour of the close method is tested by ClosedCurTests
in Lib/sqlite3/test/dbapi.py.

This is basically a backport of https://github.com/ghaering/pysqlite/issues/73.

Fix a warning on Windows 64-bit.

Write the module name rather than <module> in the error message, if module has
no __file__ attribute (ex: package).

* multiprocessing: open file with closefd=False to avoid ResourceWarning
* _test_multiprocessing: open file with O_EXCL to detect bugs in tests (if a
  previous test forgot to remove TESTFN)
* test_sys_exit(): remove TESTFN after each loop iteration

Initial patch written by Serhiy Storchaka.

Use -Wd comment line option to log the ResourceWarning.

Issue #26161: Use Py_uintptr_t instead of void* for atomic pointers in
pyatomic.h. Use atomic_uintptr_t when <stdatomic.h> is used.

Using void* causes compilation warnings depending on which implementation of
atomic types is used.