smeuletz (alexsmeu.delete@this.yahoo.com) on January 7, 2020 3:07 am wrote:

> So, even if it was cool to be backwards compatible years ago (and is
> cool today if the product does it), why bother with it so much now?
> And again, why not at least deprecate stuff that became meaningless 10 years ago so that
> in 20 years, when we use quantum computers we don't stumble on sched_yield again (joke).
>
> Regarding porting programs from Windows and UNIX
> First, there is no more stringent need to port stuff from Windows. And if we really need to port some important
> program, we can use docker-like technologies which are suitable for all these backwards compatible hacks.

I would just like to point out that these two paragraphs are contradictory; Docker depends on the kernel having a stable userspace ABI to function, and then deals with the user land not having a stable ABI by bundling up all the backwards compatible hacks you need atop a stable kernel ABI.

And there's a quantative difference between kernel and userland here; each machine runs one instance of the kernel, but that single kernel can support many different userlands on the same machine, providing precisely as much (or as little) sharing between the userlands as you desire. As a result, I can build containers for Linux apps of various ages, and (e.g.) share an XWayland AF_UNIX socket from a Fedora 31 container with a Debian 2.0 container, and have Debian 2.0 applications work as-if they were running on bare metal.

This wouldn't be possible if the kernel didn't maintain backwards compatibility - I'd have to have a hypervisor that supported all the various IPC mechanisms the kernel does, and that can forward IPC between containers, bearing in mind that I'd then need to run multiple kernels each with their own view of the world in virtual machines. And, of course, if I push the instability back one level, I'd also need multiple nested hypervisors, so that I could run VMs written for one generation of hypervisor on a different generation of bare metal hypervisor.