"." expression in Stream.select/Path.select botches end tags

[Kyle Alan Hale <kylealanhale@…>]

Try this out:

from genshi.template import MarkupTemplate
from genshi.core import Stream

def filter(stream):
    for event in stream:
        print event
        yield event

template = MarkupTemplate('''<rhyme xmlns:py="http://genshi.edgewall.org/">
    <py:match path="*[@type]">
        ${select('.') | filter}
    </py:match>
    
    <lines>
        <first type="one">fish</first>
        <second type="two">fish</second>
        <third type="red">fish</third>
        <fourth type="blue">fish</fourth>
    </lines>
</rhyme>
''')
stream = template.generate(filter=filter)
print stream.render('xml')

As a summary, this uses a py:match to find all elements with an attribute named type, selects the whole found node, and runs it through a filter to print each event in the selected stream to stdout. When I run it, I get the following output:

('START', (QName('first'), Attrs([(QName('type'), u'one')])), (None, 7, 8))
('TEXT', u'fish', (None, 7, 26))
('END', QName('first'), (None, 7, 30))
('START', (QName('second'), Attrs([(QName('type'), u'two')])), (None, 8, 8))
('TEXT', u'fish', (None, 8, 27))
('END', QName('first'), (None, 7, 30))
('START', (QName('third'), Attrs([(QName('type'), u'red')])), (None, 9, 8))
('TEXT', u'fish', (None, 9, 26))
('END', QName('first'), (None, 7, 30))
('START', (QName('fourth'), Attrs([(QName('type'), u'blue')])), (None, 10, 8))
('TEXT', u'fish', (None, 10, 28))
('END', QName('first'), (None, 7, 30))
<rhyme>
    <lines>
        <first type="one">fish</first>
        <second type="two">fish</first>
        <third type="red">fish</first>
        <fourth type="blue">fish</first>
    </lines>
</rhyme>

You'll see that the output keeps the ending tag of the first child element of <lines> for the following three lines. If you remove the child content of each element, the generated markup is correct, but the printed events still show incorrect END events.

This is not caused by the filtering; if you remove the filter and keep the child content, the same invalid markup is generated. The same behavior occurs no matter how the stream is rendered (html, xhtml, whatever).

Also, the same behavior occurs when the expression is changed to the equivalent "//*".

[hodgestar]

Fix selected tail.

[anonymous]

The attached patch fixes the problem for me without breaking the test suite. It looks like a simple mistake. The matching code has to work around Python generators not being able to return values when they finish iterating by passing in a list append function to catch the last tag of the matched XML. A single local list was being used to catch all closing tags so the list grew indefinitely.

[Kyle Alan Hale <kylealanhale@…>]

That seems to solve the issue for both my test example and the original code in which I discovered the issue. Well done.

[anonymous]

Fix committed to py3k branch in r1146.

[kylealanhale]

This is not an enhancement; it is a defect. If it will not be incorporated into a 0.6.x release, fine, but 0.8 seems like a far shot for an easily-patched defect.

[hodgestar]

Not sure why that got changed to 'enhancement'. Note that I haven't committed to trunk only because I don't have commit rights for it.

[hodgestar]

Fix committed to trunk by asmodai in r1149. Awaiting merging to 0.6.x branch.

[hodgestar]

Fix committed to 0.6.x branch in r1152.