Conda install much bigger than pip

[DavidNemeskey]

I installed conda via Miniconda, especially because I wanted to keep the footprint small. The I installed Jupyter via conda install jupyter, and

• the command downloaded 75 packages, about 300M
• the conda directory now takes up about 2G (1.7G if the tarballs are cleaned)

I also installed Jupyter via pip into a virtualv to see if the situation there is any better. Any indeed, it is: only 42 packages were downloaded, and the size of the whole environment is only 145M.

I am just wondering if there was a way to have a "minimal" package of Jupyter for conda as well?

[takluyver]

What packages were in the list?

If you're after the Jupyter notebook, you can install the package notebook instead of jupyter (the names for these are the same in pip and conda). That will leave out the Qt console, which on conda will probably also leave out Qt, which is a pretty big dependency.

[DavidNemeskey]

I can confirm that with notebook, the number and size of the dependencies is much less. Still a bit more than with pip install jupyter, though.

Sticking with conda, these are the packages installed by conda install jupyter and notebook, respectively:

• conda_install_jupyter_packages.txt
• conda_install_notebook_packages.txt

Grepping them for "MB" reveals that the worst offender is qt at 87M, but there other differences as well:

• cryptography-vectors-2.3 at 30.5M
• icu-58.2 at 22.5M
• python-3.7.0 at 32M

It is a bit strange for me to see python updated as well. Normally I would like to control the version of the environment, and not let any random package update it for me.

The list of packages I ended up with:

• conda_install_jupyter_list.txt
• conda_install_notebook_list.txt
• pip_install_jupyter_list.txt

[ivanov]

from the response to #718 and the jupyter discourse thread, it is looking like we will be removing qtconsole dependency from the jupyter metapackage which should help a lot with how large a conda installation of jupyter ends up being.