Missing option to change namespace mappings

[Wichert Akkerman]

This is essentially a summary of a thread on the mailinglist: http://codespeak.net/pipermail/lxml-dev/2010-March/005329.html

For some XML processing tools I need to be able to change the namespace mapping, specifically to insert a new entry into the namespace map. My test code looked like this:

NS="http://xml.zope.org/namespaces/i18n"
tree=lxml.etree.parse(input)
root=tree.getroot()
count=1
if "i18n" not in root.nsmap:
     root.nsmap["i18n"]=NS
for el in root.iter():
     if "{%s}translate" % NS in el.attrib:
         continue
     if hasText(el):
         el.attrib["{%s}translate" % NS]="string%d" % count
         count+=1
print lxml.etree.tostring(tree)

This did not work since root.nsmap ignores any changes you make in it. An alternative implementation suggested by Simon Wiles created a new root element with the right nsmap and moved all children over, but approach looses data (specifcally the docinfo).

The missing lxml feature to make this possible is a way to change the namespace mapping.

[Bruno Narciso (brunonar)]

This is critical to convert XML-like versions as GPX 1.0 to 1.1.

[scoder (scoder)]

> This is critical to convert XML-like versions as GPX 1.0 to 1.1.

Oh, I'd certainly take patches that implement this.

[Bruno Narciso (brunonar)]

Oh, sorry, the version that I'm using is 3.3.1 (pip install). Just now I saw that there is the version 3.4.
Thanks.

[Lennart Regebro (regebro-gmail)]

I think etree.register_namespace() might be the solution to this.

[scoder (scoder)]

For a global mapping/redefinition of a prefix, yes, I agree. Not sure if this is meant here.

[scoder (scoder)]

This is probably very easy to do, so here is a proposal:
- copy the "_Attrib" class in etree.pyx to a new class "_NsMap"
- replace the attribute based implementation with one that looks up namespaces (see also the nsmap property)
- on read access, return all namespaces that are defined on the element or its ancestors
- on write access, add an entry on the owning element itself
- return a new instance from the "nsmap" property of _Element (not _ReadOnlyElement)
- add tests.

[Jens Troeger (jens.troeger)]

I keep receiving XHTML documents with incorrect namespaces, and I guess I have two choices:

 1. treat the file as plain text and search/replace to fix the namespaces; or
 2. parse the file into an element tree and fix the actual namespaces.

In order to go with my preferred solution, I think this issue would need to be implemented. Is this feature on the current roadmap at all?