Backout 62658d9d8926 (issue #10241): it causes a crash at shutdown when deallocating a Tkapp object.

members are integers.

members are integers.

getgroups(2) on OS X 10.8.  Original patch by Mateusz Lenik.

(Done as separate patch from trunk as the code has diverged quite a bit.)

With old SQLite versions, _sqlite3_result_error() sets a new Python exception,
so don't restore the previous exception.

while adding the traceback, because it may call indirectly a Python function
and Python functions must not be called with an exception set.

In the case of the issue #18609, _ctypes_add_traceback() called the iso8859-1
decoder which is implemented in Python. Python has a ISO-8859-1 codec
implemented in C. It is not used because PyUnicode_Decode() only uses the C
codec for other names (aliases) of this codec ("latin-1", "latin1" and
"iso-8859-1").

when removing references to temporary build files.  This is not an elegant
solution but it does isolate changes to just build-installer.py.

We must only lower the soft limit, since lowering the hard limit makes it impossible to raise it back at the end.
(this could prevent core dumps from being generated when the test suite crashes)

Patch by Neil Schemenauer, minimally modified.

(it was wrongly classified in the pending calls test case)

It creates a refleak in subinterpreters, as atexit callbacks aren't triggered at their end.

when extensions are registered. Patch by Paul Moore.

  version again on Darwin.

a drive root on Windows. Original patch by Atsuo Ishimoto.

'mode' docs fix: the file must always be opened in binary in Python3.

Bug in Wave_write.close: when the close method calls the check that the header
exists and it raises an error, the _file attribute never gets set to None, so
the next close tries to close the file again and we get an ignored traceback
in the __del__ method.  The fix is to set _file to None in a finally clause.
This represents a behavior change...in theory a program could be checking for
the error on close and then doing a recovery action on the still open file and
closing it again.  But this change will only go into 3.4, so I think that
behavior change is acceptable given that it would be pretty weird and unlikely
logic to begin with.

    warning: 'utime.mtime_ns' may be used uninitialized in this function [-Wmaybe-uninitialized]

    warning: 'utime.mtime_ns' may be used uninitialized in this function [-Wmaybe-uninitialized]

Issue #18214: Improve finalization of Python modules to avoid setting their globals to None, in most cases.

Issue #15699: The readline module now uses PEP 3121-style module initialization, so as to reclaim allocated resources (Python callbacks) at shutdown.
Original patch by Robin Schreiber.