zodbpickle 0.3

Loading/Storing Python 2 Strings

In all their wisdom, the Python developers have decided that Python 2 str instances should be loaded as Python 3 str objects (i.e. unicode strings). Patches were proposed in Python issue 6784 but were never applied. This code base contains those patches.

Example 1: Loading Python 2 pickles on Python 3

$ python2
>>> import pickle
>>> pickle.dumps('\xff', protocol=0)
"S'\\xff'\np0\n."
>>> pickle.dumps('\xff', protocol=1)
'U\x01\xffq\x00.'
>>> pickle.dumps('\xff', protocol=2)
'\x80\x02U\x01\xffq\x00.'

$ python3
>>> from zodbpickle import pickle
>>> pickle.loads(b"S'\\xff'\np0\n.", encoding='bytes')
b'\xff'
>>> pickle.loads(b'U\x01\xffq\x00.', encoding='bytes')
b'\xff'
>>> pickle.loads(b'\x80\x02U\x01\xffq\x00.', encoding='bytes')
b'\xff'

Example 2: Loading Python 3 pickles on Python 2

$ python3
>>> from zodbpickle import pickle
>>> pickle.dumps(b"\xff", protocol=0)
b'c_codecs\nencode\np0\n(V\xff\np1\nVlatin1\np2\ntp3\nRp4\n.'
>>> pickle.dumps(b"\xff", protocol=1)
b'c_codecs\nencode\nq\x00(X\x02\x00\x00\x00\xc3\xbfq\x01X\x06\x00\x00\x00latin1q\x02tq\x03Rq\x04.'
>>> pickle.dumps(b"\xff", protocol=2)
b'\x80\x02c_codecs\nencode\nq\x00X\x02\x00\x00\x00\xc3\xbfq\x01X\x06\x00\x00\x00latin1q\x02\x86q\x03Rq\x04.'

$ python2
>>> import pickle
>>> pickle.loads('c_codecs\nencode\np0\n(V\xff\np1\nVlatin1\np2\ntp3\nRp4\n.')
'\xff'
>>> pickle.loads('c_codecs\nencode\nq\x00(X\x02\x00\x00\x00\xc3\xbfq\x01X\x06\x00\x00\x00latin1q\x02tq\x03Rq\x04.')
'\xff'
>>> pickle.loads('\x80\x02c_codecs\nencode\nq\x00X\x02\x00\x00\x00\xc3\xbfq\x01X\x06\x00\x00\x00latin1q\x02\x86q\x03Rq\x04.')
'\xff'

Example 3: everything breaks down

$ python2
>>> class Foo(object):
...     def __init__(self):
...         self.x = 'hello'
...
>>> import pickle
>>> pickle.dumps(Foo(), protocol=0)
"ccopy_reg\n_reconstructor\np0\n(c__main__\nFoo\np1\nc__builtin__\nobject\np2\nNtp3\nRp4\n(dp5\nS'x'\np6\nS'hello'\np7\nsb."
>>> pickle.dumps(Foo(), protocol=1)
'ccopy_reg\n_reconstructor\nq\x00(c__main__\nFoo\nq\x01c__builtin__\nobject\nq\x02Ntq\x03Rq\x04}q\x05U\x01xq\x06U\x05helloq\x07sb.'
>>> pickle.dumps(Foo(), protocol=2)
'\x80\x02c__main__\nFoo\nq\x00)\x81q\x01}q\x02U\x01xq\x03U\x05helloq\x04sb.'

$ python3
>>> from zodbpickle import pickle
>>> class Foo(object): pass
...
>>> foo = pickle.loads("ccopy_reg\n_reconstructor\np0\n(c__main__\nFoo\np1\nc__builtin__\nobject\np2\nNtp3\nRp4\n(dp5\nS'x'\np6\nS'hello'\np7\nsb.", encoding='bytes')
>>> foo.x
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: 'Foo' object has no attribute 'x'

wait what?

>>> foo.__dict__
{b'x': b'hello'}

oooh. So we use encoding='ASCII' (the default) and errors='bytes' and hope it works:

>>> foo = pickle.loads("ccopy_reg\n_reconstructor\np0\n(c__main__\nFoo\np1\nc__builtin__\nobject\np2\nNtp3\nRp4\n(dp5\nS'x'\np6\nS'hello'\np7\nsb.", errors='bytes')
>>> foo.x
'hello'

falling back to bytes if necessary

>>> pickle.loads(b'\x80\x02U\x01\xffq\x00.', errors='bytes')
b'\xff'

Support for noload()

The ZODB uses cPickle’s noload() method to retrieve all persistent references from a pickle without loading any objects. This feature was removed from Python 3’s pickle. Unfortuantely, this unnecessarily fills the pickle cache.

This module provides a noload() method again.

0.3 (2013-03-18)

• Added noload code to Python 3.2 version of Unpickler. As with the Python 3.3 version, this code remains untested.

• Added bytes_as_strings option to the Python 3.2 version of Pickler, dump, and dumps.

0.2 (2013-03-05)

• Added bytes_as_strings option to Pickler, dump, and dumps.

• Incomplete support for Python 3.2:

  • Move _pickle.c -> _pickle_33.c.

  • Clone Python 3.2.3’s _pickle.c -> _pickle_32.c and apply the same patch.

  • Choose between them at build time based on sys.version_info.

  • Disable some tests of 3.3-only features.

  • Missing: implementation of noload() in _pickle_32.c.

  • Missing: implementation of bytes_as_strings=True in _pickle_32.c.

0.1.0 (2013-02-27)

• Initial release of Python 3.3’s pickle with the patches of Python issue 6784 applied.

• Added support for errors="bytes".