Provide Device implementations

[tmc]

Current devices will block in the c implementation, this should be documented and a gevent-compatible implementation would be nice to offer. (Probably using cython for performance.)

[s0undt3ch]

My green devices approach is not good?

[tmc]

It is, this is just a reminder for me.
A cython implementation might be good for speed though.

[zacharyvoase]

I have a plan to work on this very soon, by creating a GIL-less thread in Cython to run the device and then using the self-pipe trick to synchronize the completion of that thread with gevent in the main thread. It should be as simple as calling os.pipe() and using those file descriptors.

[zacharyvoase]

OK, here's an example of applying the self-pipe trick to zmq devices: https://gist.github.com/1568277

Given the changes in my pull request for #22, and the gdevice library in that gist, you can run the gdevice_example file and it should print out the 500 messages as they're being load-balanced across 15 workers. Once it's done you can just hit Ctrl-C and the program exits neatly.

Note it's also pretty simple to wrap stuff like context.term(); indeed, any blocking calls which release the GIL can be wrapped with @selfpiped.

[tmc]

Interesting approach Zachary, I like the simplicity. This would be unix only due to the fcntl use though, right?

[zacharyvoase]

It would, but to be honest I'm not sure if that O_NONBLOCK option is even necessary—normally you only need it if you're actively reading and writing data. The code works fine without it.

[zacharyvoase]

See http://github.com/zacharyvoase/gevent-selfpipe for that solution, packaged and pip-installable.

[larsks]

Did anything happen with this proposal? Just got bit by this myself; I've written a simple Python forwarder instead, but having this in the gevent_zeromq module would be nice.