from urllib.parse import unquote_plus as compat_urllib_parse_unquote_plus
except ImportError: # Python 2
_asciire = (compat_urllib_parse._asciire if hasattr(compat_urllib_parse, '_asciire')
- else re.compile('([\x00-\x7f]+)'))
+ else re.compile(r'([\x00-\x7f]+)'))
# HACK: The following are the correct unquote_to_bytes, unquote and unquote_plus
# implementations from cpython 3.4.3's stdlib. Python 2's version
el.text = el.text.decode('utf-8')
return doc
+if hasattr(etree, 'register_namespace'):
+ compat_etree_register_namespace = etree.register_namespace
+else:
+ def compat_etree_register_namespace(prefix, uri):
+ """Register a namespace prefix.
+ The registry is global, and any existing mapping for either the
+ given prefix or the namespace URI will be removed.
+ *prefix* is the namespace prefix, *uri* is a namespace uri. Tags and
+ attributes in this namespace will be serialized with prefix if possible.
+ ValueError is raised if prefix is reserved or is invalid.
+ """
+ if re.match(r"ns\d+$", prefix):
+ raise ValueError("Prefix format reserved for internal use")
+ for k, v in list(etree._namespace_map.items()):
+ if k == uri or v == prefix:
+ del etree._namespace_map[k]
+ etree._namespace_map[uri] = prefix
+
if sys.version_info < (2, 7):
# Here comes the crazy part: In 2.6, if the xpath is a unicode,
# .//node does not match if a node is a direct child of . !
'compat_cookiejar',
'compat_cookies',
'compat_etree_fromstring',
+ 'compat_etree_register_namespace',
'compat_expanduser',
'compat_get_terminal_size',
'compat_getenv',