Libxml2 versions and namespaces

[Fred Kiefer]

I just commit some changes to the NSXML* classes which try to do more 
compile time checking for available functions. The version numbers I use 
there are based on this file: http://xmlsoft.org/symbols.xml
I think we need to agree on a minimal version of libxml2 that we want to 
support. I would like to suggest 2.7.8, but we could also agree on 2.4.30.

As I explained in my reply to Riccardo, these checks may not be 
sufficient, if the header files don't match the installed library. If 
this turns out to be a real issue, we may need to add more configure 
tests for specific libxml2 features.

When I did this change I was also right in the middle of an attempt to 
add support for namespaces. The code still passes all current test cases 
for me. (But not my local tests for namespaces :-()
This may not be that case on other systems, please report back whether I 
introduced new issues with other versions of libxml2.

Namespaces turn out to be rather complicated to add to the NSXML 
classes. There we are able to create attributes and elements with a 
prefixed name even when there isn't a namespace definition for that 
prefix. In that case we need to store a temporary namespace definition 
somewhere and later replace it with the real one. I have a few ideas 
here, but implementing them turns out to be rather complicated. And we 
also need to support the case where we have the href but no prefix...

This also led me to rethink the memory management for namespace 
NSXMLNodes. The same libxml2 node for a namespace could be used in 
different places in an XML tree, we don't have a chance to take over the 
memory management here, as we did with normal nodes that are only 
allowed in one place in the tree. I think the best we can do here is 
copy the namespace node each time we move between the libxml2 world and 
our NSXML world. Otherwise we could end up referencing freed nodes or 
leaking nodes. I haven't implemented that, but this bit seems simple and 
the slow down for not caching the object for a namespace node should be 
minimal.

Fred