It ignores the missing semicolon issue (and note also that IE can cope
with even a missing space, like "tr&eacutes mal", so you'll see that
in the wild).  Probably it could be adapted to handle that (possibly
the presumably-slower htmllib-based recipe on the python.org wiki
already does handle that, not sure).

import htmlentitydefs
import re
import unittest

def unescape_charref(ref):
    name = ref[2:-1]
    base = 10
    if name.startswith("x"):
        name = name[1:]
        base = 16
    return unichr(int(name, base))

def replace_entities(match):
    ent = match.group()
    if ent[1] == "#":
        return unescape_charref(ent)

    repl = htmlentitydefs.name2codepoint.get(ent[1:-1])
    if repl is not None:
        repl = unichr(repl)
    else:
        repl = ent
    return repl

def unescape(data):
    return re.sub(r"&#?[A-Za-z0-9]+?;", replace_entities, data)

class UnescapeTests(unittest.TestCase):

    def test_unescape_charref(self):
        self.assertEqual(unescape_charref(u"&"), u"&")
        self.assertEqual(unescape_charref(u"—"), u"\N{EM DASH}")
        self.assertEqual(unescape_charref(u"—"), u"\N{EM DASH}")

    def test_unescape(self):
        self.assertEqual(
            unescape(u"& < — — —"),
            u"& < %s %s %s" % tuple(u"\N{EM DASH}"*3)
            )
        self.assertEqual(unescape(u"&a&amp;"), u"&a&")
        self.assertEqual(unescape(u"a&amp;"), u"a&")
        self.assertEqual(unescape(u"&nonexistent;"), u"&nonexistent;")

unittest.main()