Survival kit for jupyter-notebooks devs in Zed

[rambip]

First, I have to say I absolutely love zed.
The choice Zed has made to support python repl is quite smart. Unfortunately I have run into a lot of troubles using the jupyter notebooks from the python ecosystem in Zed.

Here are some tips I have learnt on the way. They are still some pain points, and I think a guide to transit from jupyter notebook to zed would be very useful.

If it is not the right place to post this king of thing, please indicate where here I can do it.

Converting notebooks

They are very few good tools to convert jupyter notebooks to zed cells. In particular, it is very hard to keep them in sync (I needed this feature to convert them painlessly to share the visualizations with others). It was so much pain that I created one such tool: https://github.com/rambip/nbfree

Html and svg output

Very often, tools use svg, html and even javascript to render visualisations. Some times, it's overkill, but some times it's very useful.
My advice if you want to see a html output in zed is to add the following in the first cell of your file:

import webview
from multiprocessing import Process
import IPython


def display_webview(x):
    if not hasattr(x, "_repr_mimebundle_"):
        return
    data = x._repr_mimebundle_()
    if "text/html" in data:
        webview.windows = []
        webview.create_window("Html", html=data["text/html"])
        Process(target=lambda: webview.start()).start()


html_formatter = IPython.get_ipython().display_formatter.formatters["text/html"]
html_formatter.for_type(object, display_webview)

note that you can do the same for svg or any other type of output.

Magics

In jupyter notebook, magic commands can be very useful. For example, if you want to autoreload all libraries each time you run a cell, you can do

%load_ext autoreload
%autoreload 2

Well, you can do the same in zed. The only issue is that your language server will complain about the % sign. You can also use this syntax:

ipython = IPython.get_ipython()
ipython.run_line_magic("load_ext", "autoreload")
ipython.run_line_magic("autoreload", "2")

remote kernells

I'm currently working on this one. I would love to be able to run a jupyter notebook on a slurm cluster, to have my local developmements tools but a remote ipython instance on a server with gpus.
The documentation for that is not great at all.

Other ideas

If you have any other ideas, please share them here !

As a source of inspiration, I recommend reading this great survery about computational notebooks in general: https://cseweb.ucsd.edu/~lerner/papers/grounded_jupyter_icse25.pdf

[bbb651]

If you're working on notebooks in git repositories with other people a lot you could setup filter git attributes to automatically convert them back and forth from to the zed cells format