Adding a handler to detect the unlikely edge case where a message may

have an improperly quoted boundary that can cause the python library to
fail to reproduce the original message with msg.as_bytes().  See:
https://bugs.python.org/issue43818 and
https://github.com/OfflineIMAP/offlineimap3/issues/62

offlineimap/folder/Maildir.py

@@ -225,6 +225,44 @@ class MaildirFolder(BaseFolder):
                         retval[uid] = date_excludees[uid]
         return retval
 
+    def _quote_boundary_fix(self, raw_msg_bytes):
+        """Modify a raw message to quote the boundary separator for multipart messages.
+
+        This function quotes only the first occurrence of the boundary field in
+        the email header, and quotes any boundary value.  Improperly quoted
+        boundary fields can give the internal python email library issues.
+
+        :returns: The raw byte stream containing the quoted boundary
+        """
+        # Use re.split to extract just the header, and search for the boundary in
+        # the context-type header and extract just the boundary and characters per
+        # RFC 2046 ( see https://tools.ietf.org/html/rfc2046#section-5.1.1 )
+        # We don't cap the length to 70 characters, because we are just trying to
+        # soft fix this message to resolve the python library looking for properly
+        # quoted boundaries.
+        try: boundary_field = \
+            re.search(b"content-type:.*(boundary=[\"]?[A-Za-z0-9'()+_,-./:=? ]*[\"]?)",
+            re.split(b'[\r]?\n[\r]?\n',raw_msg_bytes)[0],re.IGNORECASE).group(1)
+        except AttributeError:
+            # No match
+            return raw_msg_bytes
+        # get the boundary field, and strip off any trailing ws (against RFC rules, leading ws is OK)
+        # if it was already quoted, well then there was nothing to fix
+        boundary, value = boundary_field.split(b'=',1)
+        value = value.rstrip()
+        # ord(b'"') == 34
+        if value[0] == value[-1] == 34:
+            # Sanity Check - Do not requote if already quoted.
+            # A quoted boundary was the end goal so return the original
+            #
+            # No need to worry about if the original email did something like:
+            # boundary="ahahah  " as the email library will trim the ws for us
+            return raw_msg_bytes
+        else:
+            new_field = b''.join([boundary,b'="',value,b'"'])
+            return(raw_msg_bytes.replace(boundary_field,new_field,1))
+
+
     # Interface from BaseFolder
     def quickchanged(self, statusfolder):
         """Returns True if the Maildir has changed
@@ -260,6 +298,18 @@ class MaildirFolder(BaseFolder):
         filepath = os.path.join(self.getfullname(), filename)
         fd = open(filepath, 'rb')
         retval = self.parser['8bit'].parse(fd)
+        try:
+            _ = retval.as_bytes(policy=self.policy['8bit'])
+        except UnicodeEncodeError as err:
+            if any(isinstance(defect, NoBoundaryInMultipartDefect) for defect in retval.defects):
+                # (Hopefully) Rare instance where multipart boundary is not
+                # properly quoted.  Solve by fixing the boundary and parsing
+                fd.seek(0)
+                _buffer = fd.read()
+                retval = self.parser['8bit'].parsebytes(_quote_boundary_fix(_buffer))
+            else:
+                # Unknown issue which is causing failure of as_bytes()
+                ui.warn("Message has defects preventing it from being processed!")
         fd.close()
         return retval