Fix html5lib#37: Preserve order of attributes on serialization.

This doesn't do anything about the fact that none of our
treebuilders preserve attribute order: it merely avoids the
serializer reordering them from the order it receives them in.

This changes the serializer tests to use an OrderedDict to get
alphabetical order so they continue to meet their expectations.

Author: gsnedders

html5lib/serializer/htmlserializer.py

@@ -226,7 +226,7 @@ def serialize(self, treewalker, encoding=None):
                     in_cdata = True
                 elif in_cdata:
                     self.serializeError(_("Unexpected child element of a CDATA element"))
-                for (attr_namespace, attr_name), attr_value in sorted(token["data"].items()):
+                for (attr_namespace, attr_name), attr_value in token["data"].items():
                     # TODO: Add namespace support here
                     k = attr_name
                     v = attr_value

html5lib/tests/test_serializer.py

@@ -1,6 +1,7 @@
 from __future__ import absolute_import, division, unicode_literals
 
 import unittest
+from collections import OrderedDict
 from .support import get_data_files
 
 try:
@@ -73,9 +74,11 @@ def _convertAttrib(self, attribs):
         """html5lib tree-walkers use a dict of (namespace, name): value for
         attributes, but JSON cannot represent this. Convert from the format
         in the serializer tests (a list of dicts with "namespace", "name",
-        and "value" as keys) to html5lib's tree-walker format."""
-        attrs = {}
-        for attrib in attribs:
+        and "value" as keys) to html5lib's tree-walker format. Tests expect
+        attributes to be ordered alphabetically, so use an OrderedDict to
+        ensure this."""
+        attrs = OrderedDict()
+        for attrib in sorted(attribs, key=lambda x: (x["namespace"], x["name"])):
             name = (attrib["namespace"], attrib["name"])
             assert(name not in attrs)
             attrs[name] = attrib["value"]