homepage

The following script, minidom_test.py,
from xml.dom import minidom
data = b'''
<rss xmlns:atom="http://www.w3.org/2005/Atom" version="2.0">
 <channel>
 <link>https://example.com/blog/</link>
 <atom:link href="https://example.com/rss2/" rel="self"></atom:link>
 <item>
 <link>https://example.com/blog/1/</link>
 </item>
 </channel>
</rss>'''
doc = minidom.parseString(data)
for link in doc.getElementsByTagName('link'):
 print(link._attrs)
produces different results in Python 3.2 and 3.3:
vinay@eta-oneiric64:~/projects/scratch$ python3.2 minidom_test.py 
{}
{}
vinay@eta-oneiric64:~/projects/scratch$ python3.3 minidom_test.py 
None
None

That would be caused by Martin’s change in 5d27a32ebbcc. I don’t see the docs marking _attrs public, so I think this is not a bug.

The error which prompted this issue was not caused by external code peeking into the internals - it was just my toy test script which did that. The error came up in Django testing:
======================================================================
ERROR: test_secure_urls (regressiontests.syndication.tests.SyndicationFeedTest)
----------------------------------------------------------------------
Traceback (most recent call last):
 File "/Users/vinay/projects/django3/tests/regressiontests/syndication/tests.py", line 255, in test_secure_urls
 if link.getAttribute('rel') == 'self':
 File "/Library/Frameworks/Python.framework/Versions/3.3/lib/python3.3/xml/dom/minidom.py", line 727, in getAttribute
 return self._attrs[attname].value
TypeError: 'NoneType' object is not subscriptable
So, perhaps there is a missing _ensure_attributes() call somewhere.

Upon inspection, _ensure_attributes() is only called in _get_attributes(), setAttributeNode() and _set_attribute_node(). The last of these is only called by setAttributeNode(), and it would appear that setAttributeNode() is only called if there *are* attributes.
_get_attributes() is only called by writexml().
So, calls to public methods getAttribute(), getAttributeNS() and a few others are likely to fail because, if there are no attributes in the XML being processed, _ensure_attributes() wouldn't get called.
It looks like a bug to me! ;-)

Agreed :) Needs unit tests that use the public attributes.

Instead of calling _ensure_attributes (which creates the attribute dictionaries), getAttribute should return "" immediately if there are no attributes at all, to avoid creating those dictionaries.
I'll do a patch shortly, unless somebody beats me.

New changeset 73c76466cf44 by Martin v. Löwis in branch 'default':
Issue #14168: Check for presence of _attrs before accessing it.
http://hg.python.org/cpython/rev/73c76466cf44