SL-18330: Use xml.etree.ElementTree.fromstring() when we can.

Apparently, wrapping an incoming bytes object in io.BytesIO and passing it to
xml.etree.ElementTree.parse() is slower than passing the bytes object directly
to fromstring(). Detect the BytesIO case and call fromstring() with contents.

Author: nat-goodspeed

llsd/serde_xml.py

@@ -1,4 +1,5 @@
 import base64
+import io
 import re
 import types
 
@@ -264,8 +265,15 @@ def parse_xml_nohdr(baseparser):
     # before XML declaration. Since we explicitly test support for that case,
     # skip initial whitespace.
     baseparser.matchseq(b'')
+    stream = baseparser.remainder()
     try:
-        element = _parse(baseparser.remainder()).getroot()
+        if isinstance(stream, io.BytesIO):
+            # Empirically, fromstring() seems faster than _parse(). If passed
+            # a BytesIO, extract its contents and skip to BytesIO read pos.
+            element = fromstring(stream.getvalue()[stream.tell():])
+        else:
+            # Not a BytesIO, parse the stream
+            element = _parse(stream).getroot()
     except ElementTreeError as err:
         raise LLSDParseError(*err.args)