A New Stack

Posted

I’m tired of teaching tools that are older than most of the people learning them, butif I teach something that isn’t (yet) widely used my learners won’t be able to collaborate with their colleagues, and most new tools will never catch on. That said, people don’t seem to be excited by the outline I’ve sketched for Research Software Design by Example, so I can either add it to my ever-growing list of books I’ve started but not finished or make it more exciting by using a new set of tools. I can’t reach too far ahead if I do the latter, which makes tool selection a balancing act. I’d be grateful for feedback on the list below.

Old New Because Modified
Bash Nushell risky, but Fish feels too conservative
Git sadly, still Git I really wish Gitless had taken off…
Make Just finally, a better build tool I can believe in 2024-04-19
SQLite DuckDB a better choice for data science?
Python still Python despite the mess that is Python packaging
Pandas Polars better ergonomics and performance
Seaborn Plotly I like it so much that I joined the company :-)
Jupyter Marimo also risky, but it really does feel like a step forward
LaTeX ??? maybe Typst but see below
Metaflow the most Pythonic option available
DVC preferred to Git-LFS
Ark the best little Python SSG I’ve found
VSCode mostly for the debugger
Docker as ubiquitous as Git but much less user-hostile

A few notes:

What’s missing: