This is good idea, but there is some issues... StanfordCoreNLP.raw_parse is also used as python package. If it called as python package and an error occurated, it is expected to raise a exception. Perhaps it will exposes only the parse function in the jsonrpc interface, and some people who want to use the code as python package will use the raw_parse function as well. So would you write handling error to the parse function?
Yeah I found that issue too, since if VERBOSE is False we don't wait for the timeouts. The problem is that the timeouts depends on which models we are loading. I "fixed" it by removing the timeout of 3 in corenlp.py line 365 so now it should load even if VERBOSE is False (although we don't know if it fails to load, this is not an ideal solution but works for now).