Supporting multiple language backends

rudimk commented 9 years ago

Seeing that rodeo uses a Jupyter/IPython kernel behind the scenes, are there any plans to support languages like R & Julia, which already have kernels built for Jupyter/IPython? If not, would love some pointers on how one could plug them in - although it should be easy enough, will involve modifying kernel.py by the looks of it.

glamp commented 9 years ago

Yes I've thought about this and I think it's a direction the project will ultimately go. The main issue is that I can't figure out how to get the IPython kernel to return values for autocompletion. Right now we're using jedi https://github.com/yhat/rodeo/blob/master/rodeo/kernel.py#L23-L41https://github.com/yhat/rodeo/blob/master/rodeo/kernel.py#L158-L163 which obviously only works with Python.

If we can get the autocompletion working, the UI changes are fairly trivial.

roshannair commented 9 years ago

@glamp

You need to call client.complete and read msgs on the shell_channel. This should more or less work with ipython 3+

def complete(self, code):
        s = self.client.complete(code)
        msg = self.client.get_shell_msg(timeout=0.1)
        results = [{"text":"", "dtype":"---"}]
        for completion in msg['content']['matches']:
                result = {
                        "text": completion,
                        "dtype": "---"
                }
                if code.endswith("."):
                        result["dtype"] = "function"
                else:
                        result["dtype"] = "session variable" # type(globals().get(code)).__name__
        results.append(result)
        return results

If you plan to support older ipython versions, then this may help - https://github.com/ivanov/vim-ipython/blob/master/ftplugin/python/vim_ipython.py#L302

roshannair commented 9 years ago

Hmm... I put together some really poor code that gets autocomplete going via ipython kernel in rodeo. However, I now realize that this requires the same sort of checking for msg_ids and kernel states as implemented in

_run_code()

Basically, its everything in the codeblock above, plus checking that msg_ids match and also setting the output to the same json structure as the output of _run_code()

It might be better to do this in _run_code.

@glamp What do you think?

glamp commented 9 years ago

if it works then do a PR!

jankatins commented 9 years ago

For a pointer re starting non-python kernels: You can use the MultiKernelManager to start one of the available kernels. The available kernels can be seen via the KernelSpecManager (see here: https://github.com/JanSchulz/knitpy/blob/master/knitpy/knitpy.py#L102 and the following commented line)

How to start a new kernel via the kernel manager: https://github.com/JanSchulz/knitpy/blob/master/knitpy/knitpy.py#L640

-> if the current kernel startup would be replaced with one via MultiKernelManager, then adding a new kernel is just a few lines of code and the UI to choose one.

rudimk commented 9 years ago

Wow. Let's do this! I'll try and push out something from my end too.

glamp commented 9 years ago

now using ipython kernel for autocomplete 610fd4fb0582bd9adeb4903bc3a7745cab90cd86