Specify jupyterlab --notebook-dir before lauch

[fmaussion]

Issue moved from jupyterhub/binderhub#764

I'd like jupyterlab to have its "home" in a folder within the repository, not at the repository root. It will make a nicer user experience if all the repository config files aren't available in the jupterlab directory tree.

Note that starting from a specific notebook within the repository already makes it much nicer by starting at the right place in the directory tree, but I'd like users not to even be able to see or access the repository root folder.

I've tried to tweak the start script as @jhamman suggests but it doesn't work as I'm not sure what command exactly is triggered by repo2docker when an url with lab is given.

I also tried the following in the start script, without success:

#!/bin/bash

jupyter() { command "jupyter $@ --notebook-dir ./notebooks"; }

exec "$@"

Thanks!

[jhamman]

@fmaussion - I chatted with @yuvipanda about this last week. He shared this example: https://github.com/yuvipanda/jupyterlab/blob/pure-notebook/binder/start

Based on @yuvipanda's example, you may try:

#!/usr/bin/env python3
import sys
import shutil
import os


# Add path to notebook dir
args = sys.argv + ['--notebook-dir ./notebooks']


print(args, flush=True)
os.execv(shutil.which('python3'), args)

Now, I think this should be easier. I wonder if there is a change in repo2docker that help here. One idea that comes to mind would be to append an environment variable (e.g. ${EXEC_ARGS}) to the launch command. So, in this case, @fmaussion would have defined the ${EXEC_ARGS} variable as "--notebook-dir ./notebooks" in postBuild and would not need the start script.

[krinsman]

@jhamman I probably didn't implement the aforementioned fix correctly, but when I tried to, it didn't work:
berkeley-dsep-infra/datahub@d862945

Correct me if I'm wrong, but would any fix involve adding some fancy Jinja formatting to the line here: https://github.com/jupyter/repo2docker/blob/master/repo2docker/buildpacks/base.py#L194

[betatim]

You are using the image with a "plain" JupyterHub not BinderHub. Maybe that difference is what broke the example. I'd check the command that your JupyterHub uses to launch the pods.