When trying to parse the string a&b, the parser raises an UnboundLocalError:

{{{
>>> from html.parser import HTMLParser
>>> p = HTMLParser()
>>> p.feed('a&b')
>>> p.close()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python3.3/html/parser.py", line 149, in close
    self.goahead(1)
  File "/usr/lib/python3.3/html/parser.py", line 252, in goahead
    if k <= i:
UnboundLocalError: local variable 'k' referenced before assignment
}}}

Granted, the HTML is invalid, but this error looks like it might have been an oversight.

Thanks for the report. Yes, that's in a complicated bit of error recovery code, and clearly you found a path through it that doesn't have a corresponding test :)

Just adding a patch here with a few unit tests to demonstrate the issue, comments here are welcome. This is my first patch, I believe I have put the tests in the correct place.

It appears the problem only occurs if there is an incomplete XML entity where a sequence of valid characters (for an XML entity's name) lead to the end-of-file.

The test case for "a&b " passes, as it detects the space as an illegal character for the entity name.

Thanks for the patch Thomas!
Starting from your work I made an updated patch that fixes the bug, but at the same time the tests revealed another possible issue.
In case of invalid character references, HTMLParser still calls handle_entityref instead of reporting them as 'data'. Not sure what the preferable behavior should be though, but anyway this is a separate issue.

New changeset 9cb90c1a1a46 by Ezio Melotti in branch '3.3':
bpo-17802: Fix an UnboundLocalError in html.parser. Initial tests by Thomas Barlow.
http://hg.python.org/cpython/rev/9cb90c1a1a46

New changeset 20be90a3a714 by Ezio Melotti in branch 'default':
bpo-17802: merge with 3.3.
http://hg.python.org/cpython/rev/20be90a3a714

Fixed, thanks for the report!