Consider using setup-python Github Action

[jafingerhut]

From antoninbas:

"You should consider using https://github.com/actions/setup-python. This way you can control which version of python to test with (you can add multiple versions to the test matrix as well). I'm assuming Python-based projects which use Github Actions for testing typically use this core action to set up the environment."

[fruffy]

@chreekat

[chreekat]

A (very) brief read makes me think: (1) setup-python doesn't support uv, (2) uv probably does everything we would need anyway.

So perhaps we should consider using https://github.com/astral-sh/setup-uv instead.

[jafingerhut]

I have heard of uv, but not familiar with it.

Currently I use the Python venv virtual environment package when installing open source P4 development tools on an Ubuntu Linux system. Reasons:

• Ubuntu 22.04 and earlier LTS releases I used Ubuntu apt-installed packages for Python packages, but as of Ubuntu 24.04 and later this is deprecated, so I stopped using that approach.

• Python venv enables you to create a personal directory, without requiring root permission, in any directory on the file system you choose, and have 100% control of what Python packages are installed there. I believe there is an option when using such a venv that either (a) those are the only Python packages visible to Python import statements, or (b) Python import first looks in the venv directory, and if it does not find it there, it then checks system-wide Python package installation directories. I believe I have only ever used option (a), as it is very predictable which packages you get -- only the ones installed in the venv. If it ain't there, you can't use it.

Here are some nice-to-have properties of whatever method we support for installing p4lang Python packages:

(1) We can create a package that can be published to pypi.org, and installed by others from there, that is definitely a good thing to provide. Can uv create such things?

(2) If someone wants to install the version of the Python package from Github source code in the p4lang repository directly, that is a plus, especially for developers who are testing different combinations of Python packages when debugging things. I do not know what files are required in the repository to enable this, but for example right now I can do the following command on this repository and it works:

pip3 install git+https://github.com/p4lang/ptf@main

(3) This might have the same requirements as (2), but I do not know. Currently you can make a local clone of the p4lang/ptf repository, and use pip to install it from the current contents of that directory, like so:

git clone https://github.com/p4lang/ptf
cd ptf
# Make whatever local changes you want here.  Again, useful for quick compile-edit-pip-install-debug cycles
pip3 install .

If we can preserve all of (1) through (3), and it is well documented and clear how to keep those things working as Python package dependencies of ptf and other p4lang Python packages change over time, then whatever does those things seems fine to me.

[jafingerhut]

Reading at least the first couple of paragraphs of uv's documentation, I find that it is intended to be used as a replacement for pip.

I suspect that means that supporting and testing all of the pip install methods mentioned in my previous comment, and also uv, might be twice the work of figuring out how to configure dependencies and setup/install scripts.

I don't truly have a strong preference for pip vs. uv myself, but there are at least multiple people today using pip to install several different p4lang Python packages. If using uv means pip install of these packages will no longer work (or is likely to be bit-rotted over time as uv config/setup files are updated and tested, but pip config/setup files are no longer), that should be factored in to the decision.

[fruffy]

Reading at least the first couple of paragraphs of uv's documentation, I find that it is intended to be used as a replacement for pip.

I suspect that means that supporting and testing all of the pip install methods mentioned in my previous comment, and also uv, might be twice the work of figuring out how to configure dependencies and setup/install scripts.

I don't truly have a strong preference for pip vs. uv myself, but there are at least multiple people today using pip to install several different p4lang Python packages. If using uv means pip install of these packages will no longer work (or is likely to be bit-rotted over time as uv config/setup files are updated and tested, but pip config/setup files are no longer), that should be factored in to the decision.

If you still would like to use pip workflows you can use uv pip install.

uv is a strict improvement over pip and probably one of the best things that happened to the Python ecosystem in a while, we should definitely make use of it.