Install Charliecloud binaries
with pip
!
Instead of following the (fairly simple) [install instructions]
(https://hpc.github.io/charliecloud/install.html#manual-build-and-install),
you can instead install Charliecloud into your virtualenv
or
conda
environment with:
pip install charliecloud-bin
This only workd on Linux, and you'll get binary wheels.
Installation puts the appropriate binaries in your $PATH automatically (assuming you are in a venv or conda env), so you can go follow the tutorial or use it in any other way you wish.
Currently, the versioning scheme is <Charliecloud version>a<somenumber>
,
where I increment the second number as I try out releases.
- Because you can
- Unlike
apt
,dnf
, etc. users can usepip
without root. - If you can use
conda
for this why notpip
? - When building other python projects that want to use Charliecloud, it is much easier to depend on a python package like this than expect users to have a local install of a version we like.
(4) is really the real reason.
Yes I do. This package is distributed as a wheel that claims to be conformant to manylinux1. It claims to work with the glibc that shipped in CentOS 5, which was released in 2007. It very most definitely does not!
It's a convenient lie to be able to get on PyPI. The binaries would error out on runtime. The alternative would be to have an sdist that errors out on build time. Possibly that's better? But I'm lazy and this works. Maybe I should switch.
Glad you asked!
- Their documentation is pretty good
- This talk is also pretty good!