After creating a virtual environment using conda, installing sas_kernel fails to generate a kernel.

[angelfish0808]

I'm bothering you again. Previously, I installed sas_kernel using the downloaded source code via virtualvenv. Today, I tested installing sas_kernel using conda from MiniForge to create a virtual environment. sas_kernel installed successfully from Python 3.9 to 3.14, but the kernel failed to generate. I changed the path of the kernel generated by the virtualvenv installation to the path of the Python virtual environment installed via conda, and I added the three JAR files and sascfg_personal.py correctly, but I still couldn't call the soda interface. Is the inability to generate the kernel after installing sas_kernel via conda a bug?

[angelfish0808]

To add to that, pip was used to install sas_kernel via virtualvenv.

[tomweber-sas]

Hey @angelfish0808, I don't really have enough info to know what the problems is, but I can tell you this. The SAS_Kernel isn't a 'usual' Python package (is is 'actually' one, but ...). It is a Jupyter Extension, and that's the only thing it is able to be used as. In contrast, SASPy is a usual Python package and can be installed and used in any python situation.

My first guess is that Jupyter doesn't support running it's extensions from things in arbitrary python env's as opposed to the main 'env'. Or maybe you have to do something special to get it to work that way. I really have no idea here, as I've not tried anything you're describing.

SASPy can be in any environment since it's just pure Python, not a Juptyer only thing. Obviously, the SAS_Kernel requires SASPy to be able to work, but they are independent and, since saspy IS just python, Python will find it in the environment you're trying to use, but I don't know how Juptyer (or if) supports those env's for it's purposes.

Well, a quick google says that you can have extensions in envs' but you have to configure jupyter right for it to find them.

Also, I would separate saspy from sas_kernel. You have to have saspy configured and working first, as the kernel isn't useful without that. So, you can verify saspy works independently for is the kernel is working. That way you're not mixing two different problems at the same time. Though I expect you've already done that and it's not an issue.

[tomweber-sas]

Also, @angelfish0808, I'm going to retire at the end of the year and have been trying to transition my work over to a great couple of developers. Adding them to this so they get exposure too; and maybe track it down!
@pecarmon-sas @DMcGee-SAS
Phil, you were using env's in your environment. Can you try this out and see if it works or you or not? Maybe provide some details to what you see?
Thanks!
Tom

[angelfish0808]

@tomweber-sas In a virtual environment, using conda to install (conda install -n sas -c conda-forge sas_kernel) works fine regardless of Python version (3.9-3.14), but fails to generate the kernel. However, in a virtual environment, using pip to install the downloaded sas_kernel source file works perfectly and generates the kernel. So I'm wondering if conda is missing some dependencies during the installation.

[angelfish0808]

@tomweber-sas Congratulations on your retirement! Now that you're retired, you should relax and enjoy life!

[tomweber-sas]

Ah, thanks you so much @angelfish0808, I appreciate it! I'm very excited and will enjoy it a lot!

Now I see more clearly what you're describing, thanks for the clarification. I'm not sure about the conda side of the fence for SAS_Kernel. I build SASPy for both PyPI and Conda-Forge (which would where it would com from for conda install). I'm not actually sure of the state of SAS_kernel for the Conda side of the fence. PyPI is up to date, so as you describe, that worked fine. I'm not sure I've used conda to install it before.

Seeing this is specific to the Conda install of the kernel, that points us in the right direction. We will look into that to see what we find. I do see it's the same version at both PyPI and Conda-forge, as well as the githuub repo; the last time this was touched seems to be 3 years ago. It's certainly possible something's changed with how things work in Conda since then (or not). But we'll have to investigate to see what we find. This sounds like an exciting endeavor for @pecarmon-sas and @DMcGee-SAS :)

Thanks again!
Tom

[pecarmon-sas]

Hey @angelfish0808 !

I did some testing with this and the kernel install appears to be relying on the deprecated 'setup.py install' approach to register the kernel with Jupyter. Older versions of pip will still support this installation method, but newer pip versions and conda don't. For now, if you are using a pip version that still supports the installation of the kernel, that is your best approach to getting the kernel registered with Jupyter.

I'm looking into how to manually register a Jupyter kernel (as a workaround) as well as looking at if this could be resolved by converting the project to use more modern installation methods.

Thanks for alerting us to this issue!

[angelfish0808]

@pecarmon-sas Thanks for your reply. Using pip install sas_kernel stopped working a long time ago. Whether in a local or virtual environment, pip install C:\Users\angelifish\Downloads\sas_kernel-2.4.13.zip fails to install locally, but it installs successfully in virtualenv and conda virtual environments, automatically generating the kernel. Using conda install -n sas -c conda-forge sas_kernel installs successfully in a conda virtual environment, but it fails to generate the kernel.