Contribute a new gevent-compliant event loop (see issue #112)

[mguijarr]

The gevent-compliant event loop is activated by passing gevent=True
to the embed method of repl module.
The event loop code is heavily inspired from the prompt-toolkit POSIX
event loop, adapted to gevent. More details in file contrib/gevent_eventloop.py

[jonathanslenders]

Thanks for this pull request, I really appreciate it.

I'm not merging it however. After I discovered some issues, I took some time to debug the issue and found that it can be solved like in this change in prompt-toolkit: prompt-toolkit/python-prompt-toolkit@ac26b02

Does that work for you?

[mguijarr]

@jonathanslenders thanks a lot for having considered my PR and to take a look at the problem with gevent.

Your proposal does fix the issue with monkey.patch_all(), however it doesn't help with the other issue I found (see comments in #112) when threads are not monkey patched, ie. not replaced by gevent greenlets.

It is a common idiom I think to not patch threads, for example if someone wants to do some heavy computation in threads but still wants to use gevent for non-blocking I/O.

So, even with the new code the following consumes all file descriptors (look at lsof -a -p PID | wc -l output each time you hit Enter key):

from gevent import monkey; monkey.patch_all(thread=False)

from prompt_toolkit.shortcuts import prompt, create_eventloop

def dummy_inputhook(*a):
    pass

while True:
  eventloop = create_eventloop(inputhook=dummy_inputhook)
  # Ask for input.
  answer = prompt(u'Give me some input: ', eventloop=eventloop)
  print('You said: %s' % answer)

(in reality I did my tests with ptpython in REPL, I guess the code above with the while does the same to reproduce the problem I saw)

[JohnLunzer]

@jonathanslenders, was a decision made on which route to take on this? It's not clear, was issue #112 solved via the prompt-toolkit commit you mentioned?