Regression: multiple py:match elements cause Exception 'list index out of range'

[Felix Schwarz <felix.schwarz@…>]

We have a Genshi template that uses several stacked py:match expressions which worked with 0.5.1. Using the latest 0.6 trunk I get an exception when rendering this template with Genshi:

...
File "trac/web/chrome.py", line 832, in _strip_accesskeys
  for kind, data, pos in stream:
File "genshi/core.py", line 288, in _ensure
  for event in stream:
File "trac/web/chrome.py", line 821, in _generate
  for kind, data, pos in stream:
File "genshi/template/base.py", line 592, in _include
  for event in stream:
File "genshi/template/markup.py", line 344, in _match
  if test(event, namespaces, ctxt) is True:
File "genshi/path.py", line 129, in _test
  pos_queue = deque([(pos, cou, []) for pos, cou in stack[-1]])

[Felix Schwarz <felix.schwarz@…>]

Testcase

[Felix Schwarz <felix.schwarz@…>]

The problem exists since r1020 when soc2008-xpath was merged into trunk.

[Felix Schwarz <felix.schwarz@…>]

preliminary patch

[Felix Schwarz <felix.schwarz@…>]

Ok, I attached some kind of patch - however I don't even pretend to understand what's happening there so probably it's the wrong way to fix the issue. However it works for me.

[Felix Schwarz <felix.schwarz@…>]

Actually my 'fix' is faulty, as this test case fails after my proposed modification

[cmlenz]

Proposed fix

[cmlenz]

I've added a patch that fixes both test cases in my testing. I would very much appreciate if more people could try this patch and let me know whether it breaks things elsewhere.

[cmlenz]

Patch applied in [1097].

[rblank]

The Trac test suite passes fine here with the patch (Linux, Python 2.6.4). I get one failure in the Genshi test suite, but it must be unrelated, as it is also present without the patch:

======================================================================
FAIL: Doctest: genshi.template.eval.Undefined
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/usr/lib/python2.6/doctest.py", line 2145, in runTest
    raise self.failureException(self.format_failure(new.getvalue()))
AssertionError: Failed doctest test for genshi.template.eval.Undefined
  File "/home/joe/src/genshi/trunk/genshi/template/eval.py", line 219, in Undefined

----------------------------------------------------------------------
File "/home/joe/src/genshi/trunk/genshi/template/eval.py", line 230, in genshi.template.eval.Undefined
Failed example:
    list(foo)
Exception raised:
    Traceback (most recent call last):
      File "/usr/lib/python2.6/doctest.py", line 1241, in __run
        compileflags, 1) in test.globs
      File "<doctest genshi.template.eval.Undefined[2]>", line 1, in <module>
        list(foo)
      File "/home/joe/src/genshi/trunk/genshi/template/eval.py", line 277, in _die
        raise UndefinedError(self._name, self._owner)
    UndefinedError: "foo" not defined


----------------------------------------------------------------------
Ran 814 tests in 1.603s

FAILED (failures=1)

[rblank]

Proposed fix for the failing test case:

genshi/template/eval.py

@@ -259 +259 @@
         self._name = name
         self._owner = owner
 
+    def __getattr__(self, name):
+        if name == '__length_hint__':
+            raise AttributeError()
+        self._die()
+
     def __iter__(self):
         return iter([])
 
@@ -275 +280 @@
         """Raise an `UndefinedError`."""
         __traceback_hide__ = True
         raise UndefinedError(self._name, self._owner)
-    __call__ = __getattr__ = __getitem__ = _die
+    __call__ = __getitem__ = _die
 
 
 class LookupBase(object):

[rblank]

Oh, this was #324. Sorry for the noise.

[cmlenz]

So the patch from #324 works for you? It's a bit simpler. Don't have 2.6.2 for testing available right now.

[rblank]

Yes, the simple fix works for me.

[Felix Schwarz <felix.schwarz@…>]

Thank you very much for the fix - however I found another issue which looks very similar. However we can work around the issue so for us it's not too important. Should I reopen the ticket or create a new one?

[cmlenz]

Replying to Felix Schwarz <felix.schwarz@…>:

Thank you very much for the fix - however I found another issue which looks very similar. However we can work around the issue so for us it's not too important. Should I reopen the ticket or create a new one?

Hmm, I don't think you can match <div id="fnord" /> with the XPath pattern div//*[@id='fnord']. That will only look for elements with the "fnord" ID in all of the ancestors of a div, not the div itself. Probably that was a bug in the old XPath implementation.

[osimons]

[1097] involved breaking change for basic match handling, breaking simple matching rules. Test and output below is from running rev 1096, and it can easily be verified not to work with current Genshi:

In [1]: from genshi.template import MarkupTemplate

In [2]: print(MarkupTemplate("""<?xml version="1.0"?>
   ...: <root xmlns:py="http://genshi.edgewall.org/">
   ...:   <py:match path="foo/bar">
   ...:     <zzzzz/>
   ...:   </py:match>
   ...:   <foo>
   ...:     <bar/>
   ...:     <bar/>
   ...:   </foo>
   ...:   <bar/>
   ...: </root>""").generate())
<?xml version="1.0"?>
<root>
  <foo>
    <zzzzz/>
    <zzzzz/>
  </foo>
  <bar/>
</root>

Current wrong output is:

<root>
  <foo>
    <zzzzz/>
    <bar/>
  </foo>
    <zzzzz/>
</root>

This should likely provide the basis for a new def test_match_multiple_times3(self) test case...

[osimons]

I forgot: Credit for discovery & test goes to Ozzi Lee reporting this to IRC + mailing list.

[mmitar@…]

Is there a workaround?

[hodgestar]

It looks like it was an out-by-error in the one-line fix committed in r1097. The current match template also needs to have test(tail[0]) run so that it can update its state. Fix (with new test -- thanks osimons and Ozzi Lee!) committed in r1170 and merged to 0.6.x in r1171.