r69209 caused the fp of HTTPResponse objects to be a io.BufferedReader instead of a socket.SocketIO.  This moves the underlying socket.socket object from r.fp._sock to r.fp.raw._sock.  Since _sock is an internal object, this should be ok. The change is for the testsuite only, to assert socket state.

........
  r69252 | brett.cannon | 2009-02-02 20:58:29 -0800 (Mon, 02 Feb 2009) | 3 lines

  Make importlib a package. This allows using svn:externals in the sandbox to
  package up the code for separate distribution.
........

........
  r69242 | raymond.hettinger | 2009-02-02 21:37:03 -0600 (Mon, 02 Feb 2009) | 1 line

  Register decimals as numbers.Number
........

........
  r69237 | raymond.hettinger | 2009-02-02 20:23:19 -0600 (Mon, 02 Feb 2009) | 1 line

  Validate that __length_hint__ returns a usable result.
........

........
  r68532 | kristjan.jonsson | 2009-01-11 10:23:37 -0600 (Sun, 11 Jan 2009) | 1 line

  Issue 4879: Allow buffering for HTTPResponse
........
  r69161 | gregory.p.smith | 2009-01-31 18:24:21 -0600 (Sat, 31 Jan 2009) | 2 lines

  wording for for issue4903.
........
  r69227 | raymond.hettinger | 2009-02-02 15:50:13 -0600 (Mon, 02 Feb 2009) | 1 line

  Issue 1242657: list(obj) can swallow KeyboardInterrupt.
........

to renaming of tp_compare.

svn+ssh://pythondev/python/trunk

........
  r69217 | guilherme.polo | 2009-02-02 19:08:32 -0200 (Mon, 02 Feb 2009) | 1 line

  Fix for issue #1581476
........
  r69219 | guilherme.polo | 2009-02-02 19:14:14 -0200 (Mon, 02 Feb 2009) | 1 line

  NEWS entry for issue #1581476
........

of cmpfunc.

type of tp_compare in a separate commit, for ease of reversion
should things go wrong.

Add some notes for importlib to package up what is in Python 2.7 as well as modules that can probably get replaced by importlib at some point.

an ``svn revert`` (gee, maybe more atomic commits by use of something like a
DVCS would help with this? Wish someone would get on that ...), I am checking
in an untested, work-in-progress meta path finder to handle sys.path.

Enable buffering for HTTPResponse's fp.  read() behaves identically for buffered and non-buffered IO.  read(n) also won't block if n bytes are availble on the socket.  There is therefore no reason not to use buffering.  The reason 2.x disables buffering by default, that some clients may be accessing the underlying socket directly and so bypass the buffering buffer, doesn't apply in 3.x with its redesigned IO library.
See issue 4448 and issue 4879

svn+ssh://pythondev@svn.python.org/python/trunk

........
  r69205 | martin.v.loewis | 2009-02-02 15:23:16 +0100 (Mo, 02 Feb 2009) | 1 line

  Issue #4494: Fix build with Py_NO_ENABLE_SHARED on Windows.
........

svn+ssh://pythondev@svn.python.org/python/trunk

........
  r69200 | hirokazu.yamamoto | 2009-02-02 17:01:17 +0900 | 1 line

  Set native svn:eol-style property for text files.
........

sure that accepting the empty string for top-level modules is okay.

svn+ssh://pythondev/python/trunk

........
  r69195 | guilherme.polo | 2009-02-01 22:38:54 -0200 (Sun, 01 Feb 2009) | 3 lines

  Use a single Tcl interpreter through all these tests, this may help some
  failing buildbots.
........

from object.h; don't inherit tp_compare slot on subclasses; and raise TypeError
when initializing a type that has a nonzero tp_compare slot.  Fix up
comparison-related comments in object.c and code.h.

PyObject_Cmp, PyObject_Compare, and various support functions.

Objects.

get_data().

importlib.test.support to importlib.test.source.util.

importlib.test.abc.LoaderTests.

properly. Discovered by also moving the loader tests over to
importlib.test.abc.LoaderTests.