Fabian Gundlach

Yes, sorry! I've added a check for "tornado version >= 5.0" now.

I was getting errors like this when opening the AWS web page: > Traceback (most recent call last): > File "/usr/lib/python3.7/site-packages/gevent/pywsgi.py", line 975, in handle_one_response > self.run_application() > File "/usr/lib/python3.7/site-packages/gevent/pywsgi.py",...

In case that helps, here's another reference to the AnyThreadEventLoopPolicy fix: https://github.com/tornadoweb/tornado/blob/branch5.1/tornado/wsgi.py#L186 I'm guessing only tornado is using asyncio, and for some reason needs help setting it up. :)

Your suggestion `manager < fifo | solution > fifo` would still work, except for the fact that cms currently passes the communication pipes to the manager as command line arguments...

> @fagu I don't understand why you need two extra pipes for the grader to communicate with the TaskType. Take the approach I just presented and have the TaskType kill...

@stefano-maggiolo Yes, I'm trying to support self-sufficient submissions that use stdin/stdout _directly_, not through a grader header file. (And the reason I'm providing the main function in the manager is...

@lerks I see. The idea wasn't to give more points to a solution that segfaults after the _correct_ answer, but to give a meaningful verdict to a solution that makes...

@lerks It feels a bit awkward to make that distinction, but the manager could instead treat the following two cases differently: **Illegal queries:** Immediately kill the solution. **Correct(/wrong?) answers:** Don't...

@andreyv Assuming SIGPIPE in the manager (i.e., solution didn't read everything) means "wrong answer" seems a bit problematic to me. In principle, shouldn't it be allowed to guess the correct...

@lerks I think a manager should in any case only report "correct answer" once it's reached the end of the solution's output: Extra output by the solution should also be...