Why does pip install requirements.txt install extra packages?

Question

I have a repository with an inflated requirements.tx that I'd like to clean up. Using pipreqs I've set my requirements.txt to be a minimal set of packages need for my repository. To test this, I setup a virtualenv to install the packages and then run all my unit tests to make sure they're satisfactory.

virtualenv temp_venv --no-site-packages
source temp_venv/bin/activate
pip install -r requirements.txt

Which runs fine, but I see that a whole bunch of extra packages are collected and installed. Why? Are these identified as needed by required packages, and thus installed? If so, should I then include them in the requirements.txt?

Answer 1

Yes. The packages are dependencies of your dependencies​.

But no, you should not specify them directly. Automatic tools know to download dependencies recursively and it would significantly add to maintenance overhead.

Answer 2

This might have been because of the dependencies of your written libs in requirements.txt . For ex: if you have written scipy as requirement numpy will also be installed because scipy is dependent on numpy .

Answer 3

Well, for me the above answers were not the case. Pip install was installing extra packages not in requirements.txt . The solution was:

Run conda create -n venv_name and conda activate venv_name, where venv_name is the name of your virtual environment.

Run conda install pip. This will install pip to your venv directory.

Then run pip install -r requeriments.txt

The above answer was adapted from here: Using Pip to install packages to Anaconda Environment