They could have used gentoo with their own binary package server. Maximum flexibility, including all compiler flags.

I'm a Gentoo dev and work at CERN since a while. Of course I jokingly suggested that to my colleagues. I think it could actually work pretty well, though. Many physicists use Gentoo.

We run most things on Alma Linux, but there's also RedHat in the mix.

A good intersection of CERN and Gentoo is EESSI. They use CVMFS, which is a FUSE filesystem used for software distribution from CERN, and Gentoo as part of their software stack. See links below for more info:

- https://www.eessi.io/docs/filesystem_layer/

- https://www.eessi.io/docs/compatibility_layer/

	▲	ThiefMaster- a day ago \| parent \| next [-]
		Not a Gentoo dev, but also at CERN. I've been using (headless, I'm not going down the GUI toolkit building rabbit-hole again) Gentoo for the last 15 years of development (both at home and at work). Great OS :)
	▲	yjftsjthsd-h 13 hours ago \| parent \| prev [-]
		Yeah, I'm really not sure that's a joke; binary packages, optionally combined with a prefix install, seem like a great fit.

▲

xorcist 9 hours ago | parent | prev | next [-]

Honest question, what would be different from doing the same with Debian packages? Building packages from source is trivial in Debian. It is literally one command, not much different from installing binary packages. Hosting internal repositories is completely expected and well tested.

Personally I have used Gentoo for many years, though never professionally. Its secure extensions were very well tested and documented back in the day, before SELinux pretty much took over. The kernel build system was terrific.

But never once did I consider setting compiler flags an important feature. The person who knows what flags are best used with an individual package, that -Os is more important than -O3 for this particular loop, is either the packager (who knows the system architecture) or the original developer (who wrote the code). Never the end user.

Debian is comparably a more mature system. The downside is perhaps that its packages are a bit more complex than they should be. Functionality-wise it can do anything you could wish for. But if you are a big institution with a lot of competent people in house, go right ahead. You can't possibly go wrong with a well tested system and the knowledge to improve the bits you need.

▲

tosti a day ago | parent | prev | next [-]

I had the exact same thought! One guy could maintain their OS. No forced compiler flags, internal binhost, everything else same as what they had.

	▲	M95D a day ago \| parent [-]
		There's even a thing called GRS (Gentoo Reference System) - a predefined Gentoo configuration that predictibly generates identical Gentoo installations. It's probably a chore to set up (that is, more than an ordinary Gentoo), but I expect it would make it very easy to maintain hundreds of installations based on that GRS.

▲

jmclnx a day ago | parent | prev [-]

I believe NetBSD would be better for these machines. NetBSD supports old systems quite well and have a release cycle that in cases are longer than Debian.

But, CERN may have some proprietary applications on those systems that force them to use a specific setup which is hard to replicate in other distros.

	▲	amadio a day ago \| parent [-]
		Most software at CERN is open source. But you are right, for things which are quite specific to CERN, like the control systems of the accelerator, there's also some proprietary code. Same for hardware actually, CERN develops open hardware too. https://opensource.web.cern.ch/