As it appears that `defaults` has added `blas` without a feature, the
`conda` solver appears to be prioritizing the removal of features over
channel priority. As a result, we leverage `MatchSpec` here to pin the
`blas` package to `conda-forge` and the `openblas` feature. This should
be respected by `conda` and thus avoid switching to the incompatible
`blas` package from `defaults`.
Upgrade Ubuntu to latest LTS (Bionic 18.04), including:
- Change the IMAGE used for base-notebook;
- Keep Mesos install from Mesosphere's Xenial repository, since they
don't yet have a Bionic repository;
- gnupg needs to be installed for apt-key to work;
- --force-yes option to apt-get has been removed;
- Package libav-tools is replaced by ffmpeg;
- Update Let's Encrypt example script to use 18.04.
Update jupyterlab from 0.31 to 0.32.
Update the following extensions to matching versions:
- jupyterlab/hub-extension
- jupyter-widget/jupyterlab-manager
- jupyterlab_bokeh
The `cd` command switches to the user's home directory (away from the directory we want to delete)
Then we remove the facets directory using the full path
As conda-forge does not use the `nomkl` package, but instead uses the
`blas` package to select BLAS versions, install the `blas` package and
select the `openblas` implementation.
- any files the user should be able to write should have group `user-permissions` with `g+rwX`
- remove `chown` from start.sh because it is no longer needed
- add `fix-permissions` script for setting the user-writable permissions on a path
- user-permissions group as GID 10000 (is there a reason for it to have a different value?)
- containers can set group with `--group-add user-writable` if they want to run with a different uid/gid
(without -u root -e NB_UID -e NB_GID, which make this unnecessary)
- mplimporthook should be obsolete with matplotlib 2.0
- curlrc seems to be obsolete now (updates to base image?)
- set WORKDIR to $HOME instead of $HOME/work (leave work there for compatibility)
