Configure python by default

[larsvilhuber]

The default image currently does not have Python installed. Install it, and configure (or allow to be configured) the image.

[larsvilhuber]

Such an image will be 300MB (uncompressed) larger! Shall we do this?

[larsvilhuber]

Let's solve this by providing a Dockerfile that shows people how to add this functionality instead. If there is enough demand, we could add an additional tag that has python as part of the tag line, and a different image.

[larsvilhuber]

@michaelstepner what do you think?

[michaelstepner]

@larsvilhuber is it 300MB larger if you simply copy the latest uv binaries into the docker image and provide a 1-line command for the user to install python using uv if desired? See: https://docs.astral.sh/uv/guides/integration/docker/

[larsvilhuber]

@michaelstepner very interesting idea. I should try uv. My one concern is: how does uv handle python packages that might require some compilation? The reason it is 300mb on Ubuntu is bc installing pip also installs a bunch of dependencies, mostly compilers, make, etc. Does uv not need them ever?

There are a few other things that are broadly useful, but not universally known or appreciated.

We might put together an "opiniated" or "fancy" derivative of the barebones docker containers instead.

[michaelstepner]

I think keeping it simple and then facilitating the creation of "fancy" or composite docker containers is probably a good approach!

uv does a bunch of creative things to make python installation insanely fast—including not supporting certain types of package installation. My guess is that it will work for most packages and most installations, but require additional dependencies to be installed at run time in certain scenarios. Would require more investigation to be precise about this.

[larsvilhuber]

I will keep this open. In principle, nothing currently prevents somebody from using uv within fancy Stata code on containers: just shell out, download "uv", and run it. That being said, that seems extraordinarily fancy, compared to the typical usage of any Stata user I have seen. Users that fancy would just build their own container...