It seems that we're getting something that isn't one of [i,d,l,e,0-9+] when looking up the delimiter. I can't really think of a proper cause for that. Maybe we're reading the wrong number of bytes at some point? Could it be a Unicode problem? Bencode specifies string data very loosely when compared to the mostly ascii rest of the specification.

Could be the end of data problem still I guess, as you're not running that patch. I guess my answer is ¯\(ツ)/¯