The namespace stuff seems to work, and it's definitely an improvement on the status quo, so in it goes.

1 files changed, 56 insertions, 6 deletions

diff --git a/bs4/builder/_lxml.py b/bs4/builder/_lxml.py
index cc3cb86..e5e30d4 100644
--- a/bs4/builder/_lxml.py
+++ b/bs4/builder/_lxml.py
@@ -5,7 +5,7 @@ __all__ = [
 
 import collections
 from lxml import etree
-from bs4.element import Comment, Doctype
+from bs4.element import Comment, Doctype, NamespacedAttribute
 from bs4.builder import (
     FAST,
     HTML,
@@ -42,6 +42,15 @@ class LXMLTreeBuilderForXML(TreeBuilder):
             parser = parser(target=self, strip_cdata=False)
         self.parser = parser
         self.soup = None
+        self.nsmaps = None
+
+    def _getNsTag(self, tag):
+        # Split the namespace URL out of a fully-qualified lxml tag
+        # name. Copied from lxml's src/lxml/sax.py.
+        if tag[0] == '{':
+            return tuple(tag[1:].split('}', 1))
+        else:
+            return (None, tag)
 
     def prepare_markup(self, markup, user_specified_encoding=None,
                        document_declared_encoding=None):
@@ -63,15 +72,56 @@ class LXMLTreeBuilderForXML(TreeBuilder):
         self.parser.close()
 
     def close(self):
-        pass
-
-    def start(self, name, attrs):
-        self.soup.handle_starttag(name, attrs)
+        self.nsmaps = None
+
+    def start(self, name, attrs, nsmap={}):
+        nsprefix = None
+        # Invert each namespace map as it comes in.
+        if len(nsmap) == 0 and self.nsmaps != None:
+            # There are no new namespaces for this tag, but namespaces
+            # are in play, so we need a separate tag stack to know
+            # when they end.
+            self.nsmaps.append(None)
+        elif len(nsmap) > 0:
+            # A new namespace mapping has come into play.
+            if self.nsmaps is None:
+                self.nsmaps = []
+            inverted_nsmap = dict((value, key) for key, value in nsmap.items())
+            self.nsmaps.append(inverted_nsmap)
+            # Also treat the namespace mapping as a set of attributes on the
+            # tag, so we can recreate it later.
+            attrs = attrs.copy()
+            for prefix, namespace in nsmap.items():
+                attribute = NamespacedAttribute(
+                    "xmlns", prefix, "http://www.w3.org/2000/xmlns/")
+                attrs[attribute] = namespace
+        namespace, name = self._getNsTag(name)
+        if namespace is not None:
+            for inverted_nsmap in reversed(self.nsmaps):
+                if inverted_nsmap is not None and namespace in inverted_nsmap:
+                    nsprefix = inverted_nsmap[namespace]
+                    break
+        self.soup.handle_starttag(name, namespace, nsprefix, attrs)
 
     def end(self, name):
         self.soup.endData()
         completed_tag = self.soup.tagStack[-1]
-        self.soup.handle_endtag(name)
+        namespace, name = self._getNsTag(name)
+        nsprefix = None
+        if namespace is not None:
+            for inverted_nsmap in reversed(self.nsmaps):
+                if inverted_nsmap is not None and namespace in inverted_nsmap:
+                    nsprefix = inverted_nsmap[namespace]
+                    break
+        self.soup.handle_endtag(name, nsprefix)
+        if self.nsmaps != None:
+            # This tag, or one of its parents, introduced a namespace
+            # mapping, so pop it off the stack.
+            self.nsmaps.pop()
+            if len(self.nsmaps) == 0:
+                # Namespaces are no longer in play, so don't bother keeping
+                # track of the namespace stack.
+                self.nsmaps = None
 
     def pi(self, target, data):
         pass