[IPython-dev] [jupyter] Re: IHaskell on Jupyter
Nicholas Bollweg
nick.bollweg at gmail.com
Mon Mar 30 11:05:43 EDT 2015
Sounding good.
I had a look at docker-demo-images
<https://github.com/jupyter/docker-demo-images>. It seems very monolithic
right now, and was very hard to get to build completely, though this is
likely due to the ongoing github vs nation state issues.
My first question is whether, before asking people to start hacking on it,
to modularize it a bit more... basically, have everyone that wants a kernel
have one place where they PR, and have an agreed-upon structure:
- Dockerfile
...
ADD kernels /srv/kernels
RUN find /srv/kernels -name install.sh -exec bash {} \;
...
USER joyvan
...
RUN cp -r /srv/kernels
- kernels
- Julia
- install.sh
- Welcome.ipynb
- R
- install.sh
- Welcome.ipynb
The larger question, though is: why have one image to rule them all?
This base image, is already many gigabytes, and only has the three kernels,
and a (heaping) handful of notebooks. As we add more kernels and notebooks,
it will only get slower to build and more fragile w/r/t dependencies. If
we're already requiring a kernel author to understand docker (which they
should, presumably), perhaps an approach where we allow kernel authors to
provider a link to a Docker image, preferrably as automated official
builds, we be more beneficial, as then the images themselves would be
usable outside of try/tmpnb.
I don't know what this would take, and whether dockerspawner could handle
it anyway.
Thoughts?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/ipython-dev/attachments/20150330/9f77e901/attachment.html>
More information about the IPython-dev
mailing list