Microsoft could ship its own Python distro within Excel. The biggest problem is that historically Guido / the Steering Council have avoided being involved in how py libs are distribuited, so there are many ways to do everything. But if you do have an opinion and means to enforce it, mantaining a Python environment can be a quite smooth experience.
A complication here is how large the Python distro this appears to be using is. I don't know if you've ever set up a container with all of anaconda, scikit, statsmodels, pandas, Matplotlib, seaborn, and more but it gets pretty big. A lot of people would complain about the bloat if Excel installed it by default.
Another complication is they claim the container isn't just "a docker container", but for increased security isolation (they don't want a repeat of VBA malware) it's a mini-VM focused to run on top of Hyper-V (the Windows Hypervisor) itself. That's a really complicated install process on the average machine (like installing WSL2) that sometimes involves flipping entire Windows Features on, so also something unlikely to be a smooth experience out of the box for Excel.
It might be neat if they made that an optional install and let users have offline support, but it sounded like they wanted to focus on online and collaborative UX first.
Yeah my day job involves deploying Python containers to K8s so I get where you are coming from. GB sized images are not uncommon. However:
1 - That could be an optional component, behind a "Install Python for Excel " button.
2 - You need to install Python to code in Python anyway, with or without Excel.
3 - Bloat is the norm nowadays and I'm not sure whether users care. A clean Visual Studio install takes 10 GB of disk or something. Office itself takes several GB as well.
4 - Not sure why Docker would be needed. Using Python in Windows is fine nowadays. There are caveats with libs that are very reliant on POSIX (Airflow comes to mind,) but, again, if you control your distro you can limit the libs users can install.
1) I agree, which is I why I did end my comment that I think it would be nice to see an optional install offered for offline usage.
4) I don't think they are using Docker, but they are using containers. (They mention running directly on top of Hyper-V so they might not even be considered by some developers as containers in the Docker sense but closer to "mini" VMs.) The container they are using for security/sandboxing reasons. The Excel team likely still has nightmares from the worst years of VBA macro malware and this Python system seems built specifically to sandbox all the running Python code to an execution environment (container/VM) that is easy to refresh/rebuild/recycle/tear-down. I agree that Python on Windows is usually great and most of the Python ecosystem is great at cross-platform support, but I also understand why Excel would want to sandbox any Python it runs to containers that are isolated from the host machine/user and it can easily reboot.
Python data science containers are on the order of GBs.
> And who said it must be installed by default?
I did end my comment with thoughts that I think it would be nice to have as an optional install. Mentioning why it isn't likely a default install is relevant because they want this Python feature to be accessible to everyone and they want people to collaborate and the UX for an optional install isn't what they were looking for and would complicate things. (That's not an argument against also supporting the optional install, but a reasoning for why the optional install is a harder feature to support and doesn't make the "version 1" cut.)
> Python data science containers are on the order of GBs.
Yes. And SSDs are on the order of 100s and 1000s of GBs.
> Mentioning why it isn't likely a default install is relevant because they want this Python feature to be accessible to everyone and they want people to collaborate and the UX for an optional install isn't what they were looking for and would complicate things.
How do you know these things?
It won't be accessible to everyone until it's installable. Collaboration is not cloud exclusive. The UX isn't what they were looking for is circular.
Doing everything in the cloud massively simplifies deployment and support.