manifest in DLLs to refer to the root copy of the CRT.

Delete ALLUSERS property merged from CRT merge module, so that per-user installations become possible again.

growning:-)

within the standard library.  These modules use PyFile_AsFile and later
release the GIL while operating on the previously returned FILE*.

close() will now raise an IOError if any operations on the file object
are currently in progress in other threads.

Most code was written by Antoine Pitrou (pitrou).  Additional testing,
documentation and test suite cleanup done by me (gregory.p.smith).

Fixes issue 815646 and 595601 (as well as many other bugs and
references to this problem dating back to the dawn of Python).

HandlerBException is ignored, and fix one such problem, where it was thrown
during the __del__ method of the previous Popen object.

We may want to find a better way of printing verbose information so it's not
spammy when the test passes.

Don't run kill_python as part of the build process.  Change the buildbots so they have to call it explicitly instead.

Adds _winreg.DisableReflectionKey, EnableReflectionKey, QueryReflectionKey,
KEY_WOW64_64KEY and KEY_WOW64_32KEY.

directory that is not the source directory (ie, one created using
'/path/to/source/configure'.) Leaves this test very slightly degraded in
that particular case, compared to the build-in-sourcedir case, but that case
isn't a particularly strong test either: neither test the actual path that
will be used after installing. There isn't a particularly good way to test
this, and a poor test beats a failing test.

The problem is that when trying to do the second insert, sqlite seems to sleep
for a very long time.  Here is the output from strace:

  read(6, "SQLite format 3\0\4\0\1\1\0@  \0\0\0\1\0\0\0\0"..., 1024) = 1024
  nanosleep({4294, 966296000},  <unfinished ...>

I don't know which version this was fixed in, but 3.2.1 definitely fails.

Revert r62152 (Issue #2550).  Being able to observe the results of all the buildbots was certainly useful.  All of the platforms that have some form of BSD lineage -- FreeBSD, OS X, Solaris and Tru64 -- all pass the test.  Windows and Linux, on the other hand, don't.  Windows I knew about, Linux was a surprise.  Knowing this, I believe a more appropriate fix will revolve around test_support.bind_socket() -- this method needs to return a port that nothing in the system has bound already.  The best way to do this may just be to rely on ephemeral ports, rather than having the user specify a desired port, then fall back to four random ports, then try 0.

Removed libffi.pc.in because it is not needed for ctypes.

Issue 2550: extend test_socket.py to test SO_REUSEADDR semantics when bind() is called on identical (host, port) combinations in two separate sockets.  This should raise an EADDRINUSE socket.error in all cases, irrespective of whether or not SO_REUSEADDR is set on the sockets.  However, with Windows, when SO_REUSEADDR is set on the sockets, no error is thrown (an error is thrown when the option isn't set), which results in an extremely wedged python process whenever accept() is called on either of the bound sockets.  I'm committing this test now to observe if it's only Windows that has this behaviour (via the buildbots).  Note: this WILL break all Windows buildbots for now; once I've observed the results on other platforms, I'll revert, then start looking into a patch.

directory to the directory in which the setup.py script lived (which made
__file__ wrong)

fixed, with test that the script is run in the current directory of the caller

is used as compiler.