prototype to call jedi completion server in subprocess

[dholth]

Call jedi in a subprocess, executed with python on PATH.

This is a very messy prototype of calling Jedi in a subprocess, with less code running in the actual vim-embedded Python. PR for discussion only. I hope it is enough to demonstrate that doing a process split is not a big deal. Sorry that the vendored attrs library adds noise to the diff.

I wanted this to be able to use embedded-Python-3-only vim with Python 2 and importantly PyPy projects. I should probably just recompile vim myself, but I tried this.

The way it works is that jedi_client.py mocks the impressively small jedi API needed by jedi-vim. When you make a call it sends a line of JSON to the server, just a subprocess, which calls real jedi and writes a single line of JSON as the response. Just completions and signatures implemented so far.

If this was developed further then jedi-vim could stay the same, except that a configuration flag imports from jedi_client instead of jedi.

[blueyed]

Thanks!

But as stated in #623 (comment) I consider this to be out of scope for this plugin, and would assume that @davidhalter thinks similar.
There are frameworks for doing more advanced completion (using a server architecture) already (deoplete and YouCompleteMe at least).

[dholth]

I do know about those other projects, particularly I'm a fan of YCM, but it had stopped working in my vim. The other option is switching to a different editor, probably a good idea, but not convenient.

Even when YCM is running it is still nice to be able to run a very simple (jedi-vim) plugin, especially for pypy, and to be able to have the few jedi-vim features that aren't in YCM.

This idea might have more in common with jedi (not jedi-vim) davidhalter/jedi#385 and make more sense there.

[davidhalter]

I agree with @blueyed here. There are other plugins that do this if you really need it now.

I think long-term a client/server architecture should be part of Jedi and not jedi-vim. So I'm closing this one. davidhalter/jedi#385

[davidhalter]

Thanks for your effort, though! I'm really not against these kind of big pull requests. It's just that the issue I mentioned (davidhalter/jedi#385) has been on my mind for a long time. I think we should really go with that.