Report the lxml bug upstream

src/calibre/test_build.py

@@ -167,7 +167,7 @@ class BuildTest(unittest.TestCase):
         html.fromstring('<p>\U0001f63a')
         from calibre.utils.xml_parse import safe_xml_fromstring
         bad = '\U0001f468' * 8192
-        safe_xml_fromstring(f'<p x="\U0001f600">\U0001f63a&#x1f63a;{bad}</p>')
+        safe_xml_fromstring(f'<p>{bad}</p>')
 
     def test_certgen(self):
         from calibre.utils.certgen import create_key_pair

src/calibre/utils/xml_parse.py

@@ -30,6 +30,7 @@ def safe_xml_fromstring(string_or_bytes, recover=True):
     except etree.XMLSyntaxError:
         # this happens on windows where if string_or_bytes is unicode and
         # contains non-BMP chars lxml chokes
+        # https://bugs.launchpad.net/lxml/+bug/2125756
         if sys.platform != 'win32' or not isinstance(string_or_bytes, str):
             raise
         ans = fs(string_or_bytes.encode('utf-8'), parser=create_parser(True, encoding='utf-8'))