askchapo

The Free / Non-Free demarcation is fairly well explored territory (and Debian is pretty solid on this front, despite falling short of FSF approval). Another thing that's worth consideration and doesn't get as much discussion is the organizational structure of a distribution.

Organizationally, there are roughly two types of Linux distributions. There are distributions which are created as a product or by-product of some company, who's goal is ultimately to make money (through support contracts or hardware sales typically). These are your Ubuntus, Red Hats, Fedoras, Pop_OSs, SteamOSes, etc. Then there are distributions which are maintained directly by a public collective who's sole purpose and raison d'être is the maintenance of the distribution. These are your Debians, Gentoos, ArchLinuxes, Guixs, etc.

As far as these collective organizations go, they vary a great deal in robustness. You could call a Github repository with a pair of maintainers and a dozen or so people reading the issue tracker an "organization." On the other end of this spectrum, there are distributions like Debian and Gentoo which have incredibly robust organizations with constitutions, bylaws, committees, elections, referenda, etc. Then there are a lot of distros developed on an ad-hoc basis somewhere in-between these two examples.

Having a charter and elections doesn't automatically make an institution good by any means, but governance structures like these have a big impact on the direction these distributions take, and what they are willing to compromise on. The ones which are organized publicly by members of the community rather than by the whims of some software company have done a good job keeping to their principles, and these distributions are among those of the greatest longevity.

Any experience with Fedora? I ask because apparently it's the one Torvalds uses and because there are a few science-based Fedora variants.

I think Debian is "standard" enough to be a good choice for scientific applications. Debian was an early advocate of Reproducable Builds and has waged an arduous campaign towards ensuring as many of its packages are in compliance as possible. Determinism is a very appealing characteristic for scientists. There also is a Debian Science initiative with subgroups focused on maintaining packages for various disciplines. I haven't used Debian for anything really sciency, but it's renowned stability, long release cycle, and aforementioned initiatives make it a pretty good choice.

Another distribution which is worth consideration is Guix. Guix is quite a bit more complex to get started with, but it extends the idea of reproducible builds beyond individual packages, to encompass installation and configuration of all software on the system. The idea with Guix is that you put together a recipe for what software goes in your system and how it's configured, and from that recipe it will produce an identical system each time. This allows you to build up and tear down environments at will in a completely deterministic and automated fashion, rather than manually installing, configuring, and updating software as a sequence of manual, potentially error-prone actions. Guix is very similar to the (more popular) NixOS distribution, which would be a good choice for the same reasons, but for whatever reason I've seen a higher concentration of science people around Guix.

Personally I use Gentoo. Gentoo is probably not a great choice for scientific applications due to the sheer amount of variables introduced by Portage. None the less, it packages a number of nice programs for things like CAD, Finite Element Method, Computational Fluid Analysis, Astronomy, GIS, amateur radio, etc. So, you should be able to get going on just about any mature distribution. You just might take particular interest in Debian or Guix if you are primarily using your machine for scientific applications.

Red Star OS

I mentioned the DPRK one in my post my comrade