Re: Introducing Guix "Features"! (Was: Syntactic Diabetes)

[Liliana Marie Prikler <liliana.prikler@gmail.com>]

Am Donnerstag, dem 01.02.2024 um 05:29 -0800 schrieb Felix Lechner:
> On Thu, Nov 30 2023, Attila Lendvai wrote:
> 
> > the use of 'service' to describe two rather different abstractions:
> > a component of an OS vs. a deamon process run by shepherd.
> 
> Indeed, the use of 'service' in much of Guix appears to be a grand
> misnomer. It probably occurred because the meaning expanded over
> time.
I'm pretty sure that "etc-service-type" was not something that expanded
the meaning of "service" over time, but rather something that had been
there for most of the time already.  And if you think about it,
symlinking stuff to /etc *is* a service.

> It's like we are looking back in time at the Big Bang. Our "services"
> are the microwave echoes of Guix's initial, creative spark!
> 
> Please consider a recent, helpful reply to help-guix. [1] Carlo
> mentioned the term "service" eleven times, but none of them referred
> to what I believe most readers of this message would call a service
> in other contexts. What's a newbie on help-guix to think?
You might be experiencing an effect where repeating a word often enough
makes it appear meaningless, but not only is the reply helpful, it also
highlights the exact point you're criticizing: Typically, for stuff
like configuring clamav (which I think many would consider a service),
you'd have a dedicated clamav-service-type, which of course itself
extends more low-level services.  The mere fact that we haven't yet
gotten around to implement that doesn't mean that our terminology is
somehow off, it just means that there's something missing that
something is missing.

> Should Guix services instead be called "features"?
> 
> Those "features" are central to any operating system definition.
> Other choices like "provider" may not fully capture our collective
> uses throughout the code and the documentation. I am especially
> thinking about 'modify-features' and '%base-features'.
I think it's better to do descriptive linguistics over prescriptive
linguistics.  The use of `guix shell' prevails over `guix environment'
not because one of them uses a "more correct" term, but rather because
one of them provides a nicer interface that requires less typing (among
other benefits like reduced headaches, etc.)

As for the term features, that's a word that can mean literally
anything in between and around packages, services, etc.  Typically,
packages or services implement features, for instance evolution
implements "reading mail" and "writing mail", whereas (a) gdm(-service)
implements "logging into your account".  I personally don't think your
proposed relabeling would be worthwhile, especially if we have more
important features (like syntactic sugar) to discuss.

Cheers