Anyone knows of similar deprecation policies in other malleable software projects?

How can we conciliate malleability and some form of stability for user config and extensions?

I think #Emacs itself is the gold standard here: long deprecation periods and clear documentation. (Third-party Emacs packages are often not doing as good a job these days!)

In #Guix we’re discussing a deprecation policy to serve both as a guideline for contributors and as a “contract” with users:
https://issues.guix.gnu.org/72840

The latter is important: users should know that they can trust Guix to not change overnight without prior warning, that scripts, manifests, OS configs, etc. won’t break unexpectedly.

If you’re a Guix person, now’s the time to make your voice heard!

As a user, there are times when I’m in the mood for fiddling with the code I use, and others when I just want to get other things done.

I recently had my config “broken” by changes in a new version EXWM, the Emacs window manager.

It’s a situation where malleability gets in the way and could encourage some to turn to rigid solutions where they can hope to be an undisturbed “end user”, at the cost of decreased autonomy.

I’m a big fan of malleable and extensible software packages à la Emacs; I think they’re the way to provide practical freedom to users, when licenses provide freedom “de jure” only.

That one can gradually discover the code and fiddle with it blurs the user/developer distinction and reduces the dependency of one group on another.

#FreeSoftware

🧵

Incidentally, that’s another illustration that retrofitting namespaces in an ambient-authority monolithic kernel is tricky.

This is in contrast with the Hurd, where per-process views are an inherent part of the design.

Recommendation is now to disable network namespaces to mitigate vulnerabilities due to their interaction with unprivileged user namespaces:
https://www.openwall.com/lists/oss-security/2024/04/14/1

That’s a sad outcome.

The folks at arte.tv have put geolocation gates that prevent access over #Tor, but at least they have a sense of humor.

@clacke Speaking of which, the story of “UTC” is fun:
https://en.wikipedia.org/wiki/UTC#Etymology

Today it's been 10 years since the first #GNU #Guix commit! 🎂

Some of the 600+ people who made it possible wrote their story:

👉 https://guix.gnu.org/en/blog/2022/10-years-of-stories-behind-guix/

🎉 🥳 👏 #FreeSoftware #Guix10Years

@ekaitz_zarraga @Ninjatrappeur In general I’d recommend records: they’re disjoint types (so you get type checking and can’t put in a value for a nonexistent field) and they’re fast. There’s destructuring via ‘match’ (or ‘match-record’ in Guix).

Dear imperialist friend, this character is not any more “special” than those you’d find in Зеленський or احمد بن بلة.

The #GNU #Shepherd 0.9.0 is out!
👉 https://lists.gnu.org/archive/html/guix-devel/2022-04/msg00089.html

Improved logging, on-demand service startup à la inetd and systemd, concurrency through Fibers. Running on your #Guix System machine anytime now!

#Shepherd: systemd-style socket activation ✅

Now to try all that out on #Guix System and Guix Home services.

@AbbieNormal Yeah, it’ll provide better logging, possibly faster startup, and services started on demand—e.g., sshd actually started only when a client connects.

The #GNU #Shepherd finally getting concurrency via #Fibers!
https://lists.gnu.org/archive/html/guix-devel/2022-03/threads.html

Now with preliminary inetd-style service support, hopefully soon with systemd-style socket activation.

Concurrent programming with Fibers in #Guile is delightful!

@clacke @ekaitz_zarraga Yes, https://hpc.guix.info/blog/2018/01/pre-built-binaries-vs-performance/ talks about “fat binaries” and function multi-versioning, which are like you describe.

@clacke @ekaitz_zarraga True! The difference I think is that Gentoo makes reproducibility and provenance tracking harder because of that.

@ekaitz_zarraga I suspect the new instructions mostly come from extensions, in particular SIMD extensions since ~2000 (remember MMX? :-)).

They trickle in slowly as people update their hardware and software.

See also: https://hpc.guix.info/blog/2022/01/tuning-packages-for-a-cpu-micro-architecture/

TIL (in a talk on RISC-V) that, since 1978, the x86 ISA got three new instructions *per month*.