[FIX] translate: fix xml_translate to use an HTML parser as a fallback

When terms are translated in the website, their values are sent as HTML (with
text entities escaped).  The function `xml_translate` is used to sanitize those
values, and sometimes crashes because of non-closed entities like `<br>`.  Fix
it by using an HTML parser as a fallback.

This is a partial revert of 6c8e53a087.

openerp/addons/base/tests/test_translate.py

@@ -50,13 +50,6 @@ class TranslationToolsTestCase(unittest.TestCase):
         self.assertEquals(result, source)
         self.assertItemsEqual(terms, [source])
 
-    def test_translate_xml_text_entity_unescaped(self):
-        """ Test xml_translate() on plain text with HTML entities. """
-        terms = []
-        source = "Blah blah blah"
-        with self.assertRaises(XMLSyntaxError):
-            xml_translate(terms.append, source)
-
     def test_translate_xml_text_entity(self):
         """ Test xml_translate() on plain text with HTML escaped entities. """
         terms = []

openerp/tools/translate.py

@@ -271,10 +271,11 @@ def xml_translate(callback, value):
         trans.process(root)
         return trans.get_done()
     except etree.ParseError:
-        # fallback in case it is a translated term...
-        root = etree.fromstring("<div>%s</div>" % encode(value))
-        trans.process(root)
-        return trans.get_done()[5:-6]       # remove tags <div> and </div>
+        # fallback for translated terms: use an HTML parser and wrap the term
+        wrapped = "<div>%s</div>" % encode(value)
+        root = etree.fromstring(wrapped, etree.HTMLParser(encoding='utf-8'))
+        trans.process(root[0][0])               # html > body > div
+        return trans.get_done()[5:-6]           # remove tags <div> and </div>
 
 def html_translate(callback, value):
     """ Translate an HTML value (string), using `callback` for translating text