The community must separate research from market capitalization. -Rob Pike http://herpolhode.com/rob/utah2000.pdf Linux GUIs are dinosaurs and I don't think there is much that is salvageable. As many have noted, MS are now more responsive to users than Linux distros. For me Linux GUIs seem so broken and clunky that they are almost be unusable. Devs seem determined to find new and innovative ways to break users' workflow. Instead of using the GUI as a pen where users can be herded, watched and harvested of their personal data, I think the GUI needs to be penned up watched by users, to keep it under control and out of trouble. This requires VMs and mobile has paved the way. With the GUI out of the way, we can focus on the user-space system. My proposals for that foundation are Plan 9 and Go (with C for the performance parts).Like Gobolinux, Plan 9 addressed the directory hierarchy issue, except they use bind mounts instead of symbolic links. That's an improvement
but I would rather use something like lightweight subvolume clones, as an improvement on hard links. (Would that work?) I'm not worried about the partition issue: mfrs are planning hard drives with hundreds of terabytes. I don't know how this could work within Debian. Maybe people are right, and the distro-as-packaging-service is an anachronism. I don't think moving into orbit around Red Hat will solve that. I would like to keep something like dpkg/apt, but adapt it to a compile-from-source model, maybe compile from "launch" so each program could use pre-audited source and signed "recipes." Binaries are not only removed from distribution, but from the users' hard drive so they exist only in RAM. All software is "interpreted." The division between compiled and interpreted languages will narrow and eventually disappear (PyPy seems to pre-sage this trend). It's a new concept for me, but I was surprised only by how long it took for me to notice it. This demo is one of the things that got me thinking about it: https://golang.org/