Re: [xml] [bnwest rice edu: xml mailing list ...]

[Daniel Veillard <veillard redhat com>]

On Thu, Sep 21, 2006 at 05:52:21PM -0500, Ross J. Reedstrom wrote:
> Forwarding for a colleague who can't seem to get his email subscription
> to work.

  I didn't see any bounce, very strange, try again !

> He's starting with a quote of an email from Daniel earlier up
> the list, when we ran over the missing python implemention of the range
> types.

  okay

> ----- Forwarded message from "Brian N. West" <bnwest rice edu> -----
>
> Delivery-date: Thu, 21 Sep 2006 15:51:52 -0500
> Date: Thu, 21 Sep 2006 15:44:46 -0500
> From: "Brian N. West" <bnwest rice edu>
> To: "Ross J. Reedstrom" <reedstrm rice edu>
> Subject: xml mailing list ...
>
> Daniel Veillard said:
> >   Hum, yeah, that must be the binding missing support from the extra
> > XPath types (XPATH_POINT, XPATH_RANGE and XPATH_LOCATIONSET), I guess
> > nobody tried to really use them from Python, you're in unknown land.
> >   Probably libxml_xmlXPathObjectPtrWrap() should be augmented to
> > handle them (python/types.c around line 397), and convert them to Python
> > type. I don't know what would be the best way to represent a point,
> > maybe a tuple made of a node and an index.
> >   I would be glad to get patches and some example to run the regression 
> tests
> > from python/test/ !
> >
> >   Thanks in advance :-)
>
>
> We made the following changes to libxml_xmlXPathObjectPtrWrap():
>
> case XPATH_POINT:
>
> We create a 'point' tuple with contains the node and an index:
>
> node = libxml_xmlNodePtrWrap(obj->user);
> indexIntoNode = PyInt_FromLong((long) obj->index);
>
> ret = PyTuple_New(2);
> PyTuple_SetItem(ret, 0, node);
> PyTuple_SetItem(ret, 1, indexIntoNode);
>
> case XPATH_RANGE:
>
> We create a list (PyList) with two point tuples (PyTuples),
> or one point tuple if range is collapsed.
>
> case XPATH_LOCATIONSET:
>
> We create a list (PyList) with an entry per XPath Object contained in the 
> location set.  We call libxml_xmlXPathObjectPtrWrap() recursively for each
> XPath Object.
>
> for our python test of interest:
>
> xp='xpointer(string-range(id("para1"),"Geometric",9,1))'
>
> libxml_xmlXPathObjectPtrWrap() gets called with a Location Set with one 
> XPath Object which is a Range. We build :
>
> a PyList (Location Set) which contain one PyList (Range)
> which contains two point PyTuples.

  Okay, that sound fine.

> At some point after libxml_xmlXPathObjectPtrWrap() is returned, a call is 
> made to libxml_type() where we experience a memory access error.  Within a 
> call is made to
>
> PyArg_ParseTuple(args, (char *) "O:last", &obj)
>
> followed by
>
> cur = PyxmlNode_Get(obj)
>
> which leaves us with a 'cur' of value 1, instead of a valid memory pointer.
> Deferencing cur causes the memory access error.

  Hum, you need to chase where this is really coming from, it's hard to infer
from the stack trace since whatever called libxml_type() is not shown.
It is likely to be in the wrapper taking the low level python object 
to convert it to a instance of the classes in libxml2.py . Check
xmlCore.get_type, add debug there if necessary. Check the wrapper when
doing the reentrancy in the .py code when returning from C.

> It is a mystery to us why libxml_type() is being called.
>
> Any thoughts?

  Something is being seen as an xmlCore (i.e. a node) and that is used
when building a node instance from a low level wrapper object. Add
debug around get_type()

Daniel

-- 
Red Hat Virtualization group http://redhat.com/virtualization/
Daniel Veillard      | virtualization library  http://libvirt.org/
veillard redhat com  | libxml GNOME XML XSLT toolkit  http://xmlsoft.org/
http://veillard.com/ | Rpmfind RPM search engine  http://rpmfind.net/