4 ms·
> There's nothing DIY about a ready made Linux distribution such as the Fedora Desktop. Until you want to change something. It is trivial to target a single us
by AnIdiotOnTheNet 9y ago
> There's nothing DIY about a ready made Linux distribution such as the Fedora Desktop.
Until you want to change something. It is trivial to target a single use case and make it easy with Linux (most of the time) or almost any other OS, but that isn't what we want from Desktop OS, we want flexibility so we can conform it to our workflow, and Linux Desktop is absolutely terrible at that because it's become a giant Rube Goldberg machine of schizophrenic and often half-thought out components written with the idea that they're in a multiuser server environment but pretending they were designed for personal desktops.
If someone built a new, joy to work with and develop for, userland on top of the Linux kernel that wasn't oriented towards the idea that the system is a multi-user network server in 1983, I would subscribe the hell out of it (though in my opinion it had better be open source regardless). As a side project, I'm actually playing with concepts in this vein, but I'm really not that good a developer and don't have a lot of time to devote to it, it's just that no one else is doing it.
- swiley 9y agoIf you want to change something then run an OS that's built with the expectation that you change things (Alpine, Gentoo etc.)
- AnIdiotOnTheNet 9y agoThey aren't any better because they still use all the same software and follow all the same broken paradigms.
- swiley 9y agoNo they don't? Alpine uses the BusyBox deamons (initd, evdev etc.) And Gentoo uses openrc. Both of these are much easier to tweak than systemd. Same for the network manager replacements and so on. Different distros (that aren't just modified Debian or redhat) can differ by quite a lot.
- majewsky 9y ago> Both of these are much easier to tweak than systemd. Disagree. The documentation for systemd is very detailed, and while it is in fact different from sysvinit (which I believe is 90% of the criticism), it is remarkably consistent with itself (compared to a feature-equivalent set of 100 separate tools). The big achievement of systemd is that it steamrolls over a ton of idiosyncrasies that distributions built over the decades. These days, I can just SSH into a random distribution (Arch, Debian, Ubuntu, Redhat, Suse) and everything works as I expect (querying the status of services, finding logs, issuing administrative commands, etc.).
- pritambaral 9y ago> it's become a giant Rube Goldberg machine of schizophrenic and often half-thought out components written with the idea that they're in a multiuser server environment but pretending they were designed for personal desktops I don't get where you're getting this idea.
- striking 9y agoWayland is an attempt to replace X11. PulseAudio is a (now decent) attempt to clean up the Linux audio space. There are attempts being made. Your comment is too vague to be falsifiable, though. Would you mind being more precise as to what you'd like to see changed?
- AnIdiotOnTheNet 9y agoIn the short version: Package management and spreading files all over the hierarchy is a really terribly inflexible way to manage applications. Applications should just be folders that contain any resources the application requires that aren't in the base system (this requires a defined and stable base system). This requires no special management software whatsoever and fits intuitively into the file management metaphor that already exists. See: Application Bundles, AppDirs, AppImage, and just about any desktop os before Windows 95. Permissions should not be oriented around protecting the system from the user. That's a concept that makes sense in multiuser servers and organization networks, not personal devices. Permissions should be oriented around protecting the user's data from misbehaving applications. Base system libraries should be stable on at least decade long timelines, barring a severe security issue that cannot be mitigated in any other way. I generally agree with Linus's opinion regarding breaking comparability. Unfortunately, a lot of the developers of the userland stuff aren't Linus. In general, the OS should not presume it knows what my workflow is. This is the problem with the difference between "easy", which requires understanding what I want, and "simple" which is making systems understandable enough that I can conform them to what I want. It is my opinion that Linux Desktop has put so much effort into being easy that it has destroyed any simplicity it once had (see common Linux Desktop evangelist argument: "normal users don't need that"). The commandline environment and toolset are ass. PowerShell implemented the same concept significantly better. Explaining my ideas for the how a GUI interface should work would take a long time and are not really relevant because they could exist as just another compositor in the current ecosystem.
- ghaff 9y ago>Applications should just be folders that contain any resources the application requires that aren't in the base system (this requires a defined and stable base system) You're basically describing OCI/docker/moby containers.
- ghaff 9y agoIt's not clear to me what the userland has to do with "the idea that the system is a multi-user network server in 1983." Pretty much any PC GUI out there today would be at least recognizable to a Windows 3 user OR a Unix workstation user--one of which was based on a multi-user network server but the other was not. There are alternative models out there. There are Chromebooks where you basically live in the browser (which many people do anyway for most of their computing) and there's the app store model we see on phones and tablets. I guess there's a potential for new paradigms that blend the flexibility of current GUIs with simplicity but I'm not sure a multiuser server foundation has much to do with it.
- AnIdiotOnTheNet 9y agoIt's mostly about the way permissions are handled. User-oriented permissions are about protecting the system from the user. This is a useful notion when people connect to a multiuser server via a terminal, or for shared network resources, but is nigh useless and generally just gets in the way for personal computing devices. The mobile OSs orient the permissions system around the applications, protecting the user's data and resources.
- ghaff 9y agoThat's fair enough although one of the reasons that works is that applications are sandboxed and limited in terms of what they are allowed to do. Certainly one could imagine a tablet/phone OS that was more oriented to desktop use cases (mouse support, etc.). But now we're back to Windows 8 and possibly Surface Pros and that just hasn't been a particularly popular model.
- AnIdiotOnTheNet 9y agoBut that's exactly what we want: applications that are only allowed to access what the user has permitted them to access. And I believe it can be done better than popping up a dialog all the time like UAC or just once at first launch like Android/iOS. You don't need to tabletize your interface to implement this concept. Bubblewrap, FireJail, and some similar software already exists that implements this in Linux (poorly in my opinion, but it works).
- JdeBP 9y agoPeople have done it. The self-contained application bundles idea that you espouse can be found as Daniel J. Bernstein's "slashpackage" mechanism, from the turn of the century. It gives side-by-side installation of multiple versions of packages and simple packaged files installation, upgrade, and removal procedures. * https://news.ycombinator.com/item?id=10357816 https://news.ycombinator.com/item?id=10357816 * http://code.dogmap.org/slashpackage/ http://code.dogmap.org/slashpackage/ The ability for users to isolate their own applications from one another is one of the concepts (removing the superuser) that underpins GNU Hurd, where one can sandbox one's own programs using translators and servers that substitute for the root ones. * http://jdebp.eu./FGA/hurd-daemons.html http://jdebp.eu./FGA/hurd-daemons.html
- AnIdiotOnTheNet 9y agoI am aware that people have tried it, ROX Filer knew about AppDirs, I think WindowMaker/GNUStep supported Application Bundles, there are currently (in keeping with Linux Desktop tradition) at least 3 implementations of the concept (FlatPak, Snap, AppImage), all of which are overengineered (but AppImage is the closest to correct by a wide margin). But there is no system I can install today that makes it the primary way of doing things. That's what I'm trying to build, a new OS (or perhaps "operating environment" would be a more precise term), not a "distribution", on top of the Linux kernel, that takes the good ideas that already exists and makes them the primary way of doing things. It's likely that even if I did finish the project people would ignore it. Maybe only I think these are good ideas, maybe I'm missing something, but I don't care. When I look at the future of the desktop operating system I see Windows 10 continuing down its path of becoming increasingly uselessly stupid, and the Linux Desktop, and those two futures make me want to become Amish. I'm building this because if I don't build something like this because if I don't I'll probably just leave computing altogether.