changes for Manipulate

[poke1024]

resubmitted, though Travis still fails because these changes depend on the corresponding changes in Mathics

[sn6uv]

I got an error when trying to use Manipulate:

(using this branch of imathics and mathics/Mathics/pull/362)

(<class 'RuntimeError'>, RuntimeError('Comms cannot be opened without a kernel and a comm_manager attached to that kernel.',), <traceback object at 0x7ff12ed08308>).

Edit: Just noticed this warning message (could be related):

(venv) angus@varch> jupyter notebook
/home/angus/venv/lib/python3.5/site-packages/widgetsnbextension/__init__.py:30: UserWarning: To use the jupyter-js-widgets nbextension, you'll need to update
    the Jupyter notebook to version 4.2 or later.
  the Jupyter notebook to version 4.2 or later.""")

It's a bit strange since I am running 4.2:

(venv) angus@varch> jupyter notebook --version
4.2.0

[poke1024]

I'm using jupyter notebook version 4.1.0. I'll investigate...

[sn6uv]

I'm wondering if we should add ipywidgets to the setup.py install_requires.

[poke1024]

I just removed ipywidgets from my installation and it basically still starts up without errors; but I'm guessing it's some kind of configuration thing (module versions, ...).

I didn't manage to install Jupyter 4.2, I probably have to wait for the pip package which will be a few days I guess.

I looked at it, but I don't get it. The exception seems to be thrown from ipykernel/comm/comm.py:

        comm_manager = getattr(self.kernel, 'comm_manager', None)
        if comm_manager is None:
            raise RuntimeError("Comms cannot be opened without a kernel "
                        "and a comm_manager attached to that kernel.")

self.kernel of course is a MathicsKernel in our case. Since one of the things in the changed MathicsKernel constructor is to call self.establish_comm_manager() which which will do a self.comm_manager = CommManager(). So how can kernel.comm_manager end up None?

In ipykernel/ipykernel.py the ipython guys install their CommManager, and they do it in the same way (i.e. there really shouldn't be any ipykernel open calls before the Kernel constructor has finished) :

https://github.com/ipython/ipykernel/blob/4.x/ipykernel/ipkernel.py

[poke1024]

I have Jupyter 4.2 now, turns out conda has the latest package. I cannot reproduce it though.

[sn6uv]

I tried debugging this and quickly realised I was on the wrong Imathics branch. Sorry about that. I'm not sure what that warning is, it seems unrelated.

[poke1024]

I'm quite glad to hear that actually. :-)

[sn6uv]

Got it working, It looks great!

Had some issues with virtualenv/jupter-nbextension. In addition to installing ipywidgets you have to run:

jupyter nbextension enable --py widgetsnbextension --sys-prefix

This is similar to how you have to install the imathics kernels manually (although there are some hacks to workaround this). I wonder if it makes sense to collect this initialisation stuff up somewhere.

[poke1024]

Are you on a developer install of ipywidgets? Then it seems to be the expected behaviour to have to manually enable the widgetsextenstion:

https://github.com/ipython/ipywidgets/blob/master/docs/source/dev_install.md

Using a vanilla pip install of ipywidgets on the other hand seems to enable it automatically. So, it's probably a thing only on development builds.

Still, there must be a way to check for the existence of widgetsnbextension, but I didn't yet find a way to do that.

[sn6uv]

I installed ipywidgets with pip, maybe it was because of the virtualenv.

[sn6uv]

I'm happy to to merge this after mathics/Mathics#366.