Author danielsh
Recipients Arfrever, danielsh, einarfd, eli.bendersky, ezio.melotti, flox, georg.brandl, jcea, larry, python-dev, santoso.wijaya, skrah
Date 2013-01-01.22:56:48
SpamBayes Score -1.0
Marked as misclassified Yes
Message-id <20130101225642.GB3402@lp-shahaf.local>
In-reply-to <1357055639.81.0.739252265035.issue16076@psf.upfronthosting.co.za>
Content
Eli Bendersky wrote on Tue, Jan 01, 2013 at 15:54:00 +0000:
> Why did you change the class name, by the way, I don't think it's
> a valid change at least for 3.3 in terms of backwards compatibility.
> 

With unmodified tip of 3.4:

    >>> import pickle, xml.etree.ElementTree as ET
    >>> pickle.dumps(ET.Element('foo'))
    Traceback (most recent call last):
      File "<stdin>", line 1, in <module>
    _pickle.PicklingError: Can't pickle <class 'Element'>: attribute lookup builtins.Element failed

I added the "_elementtree" to the tp_name in order to bypass the above
error.  Module-qualified names were in use elsewhere (including by
_elementtree._element_iterator) so it seemed reasonable.  I'll defer to
you about compatibility implications of this change.

> Regarding that compatibility, and even easier idea would be for the
> C pickle to return the same __dict__ implicitly gathered from the
> Python version, and then only one version of the unpickle is required.

That makes sense.  But going forward it might be even better to define
an explicit __reduce__/__getstate__ for the Python version too, so if
the instance dict grows new members, they won't get serialised
unintentionally.  (This consideration is also why C code in the latest
patch makes some effort to notice unknown args in the dict.)
History
Date User Action Args
2013-01-01 22:56:48danielshsetrecipients: + danielsh, georg.brandl, jcea, larry, ezio.melotti, Arfrever, eli.bendersky, skrah, flox, santoso.wijaya, python-dev, einarfd
2013-01-01 22:56:48danielshlinkissue16076 messages
2013-01-01 22:56:48danielshcreate