Home | Contact

[This page was converted from the Gemini version
at gemini://gemini.ctrl-c.club/~lars_the_bear/morelikewindows.gmi]

More like Windows every day

I started using Linux on desktop and laptop computers in the '90s, long before it became a relatively mainstream thing to do. My first Linux distribution took the form of a shoe-box-sized stack of floppy disks. Since then I've developed software for and on Linux more or less constantly. I've written bootloaders and kernel drivers. I've ported Linux to unusual platforms. Until recently I maintained a Linux distribution for Android devices. Nearly all my commercial clients use Linux.

I say this not to boast, but merely to point out that I'm not a newcomer to Linux. I've seen it grow from a hobby project to the backbone that links the world's data centres. And I'm forced to draw this conclusion:

Normal people don't run Linux on their desktop or laptop computers.

I honestly don't know whether this is something to lament, or not. Linux has a strong, even dominating, presence in commercial computing. It powers the majority of the world's mobile computing devices, even if their users don't actually notice it. Do we really need Linux on desktops as well? More particularly, do we need Linux to compete with OS/X and Windows in the domain of accessibility to non-technical users?

Let's face it -- people who are Linux enthusiasts, or who deal with Linux on servers every day, are likely to run it on their desktop systems, however fiddly it is to do so. People who aren't enthusiasts, or who don't already have an investment in Linux, won't run it on their desktop systems however easy it is.

It can't be denied that installing and operating a modern Linux like

Ubuntu or Fedora is easy. It's certainly as easy as Windows to install and use. Modern Linux systems are as good as Windows at hiding the underlying complexity. As more and more business operations go on line, the lack of proprietary software for Linux becomes less of a problem (although it's still a problem). I'm sure that, outside the IT industry, most people who use computers do most of their work with a web browser. A modern laptop might as well be a web browser in a case. In fact, that's pretty much what a Chromebook is.

But for all that, normal people don't use Linux. They'd rather pay money for a less reliable, less well-documented, less secure, and less regularly updated commercial product. Not for considered reasons, but simply because that's what everybody else they know does, and Windows came with their computer. Nearly every laptop and desktop computer comes with Windows pre-installed -- outside the developer community you'd have a job to sell one that didn't.

Modern desktop Linux distributions are very slick. They're easy to use, and they look nice. However, they have achieved this superficial glossiness at the price of a huge increase in complexity, all of it tightly integrated. An increasing amount of Linux desktop software is understood by a decreasing number of people. Although Linux is not proprietary in the way Windows is, desktop Linux is becoming more like Windows every day.

What we have now is a situation where every Linux software component depends on every other component. Some of those components are extremely complex, and some consume huge amounts of system resource. However expert you are, it's difficult to select specific software elements to use, because there's such a tangled web of interdependencies.

My Fedora 30 desktop system is running 168 processes directly after boot, not including kernel pseudo-processes. Of these, there are about thirty whose functions I don't even know. Of the others, I suspect that at least half are things I will never use, but that I can't remove without breaking something else. All this software uses memory, storage, and CPU although, to be fair, these things aren't particularly expensive at present. What I want to be running when I boot my Linux system, other than software I installed myself, is an X server, a window manager, a display manager, and an init process. And that's all.

It seems to me that there are many things that most mainstream distributions install by default, that have no real value for any reasonably technical user. In this category I would put at least the following:

And so on. Frankly, I have no explanation for the ubiquity of DBus and UDev, but I don't think we'll be able to excise those things now. I don't mind software getting installed that is easy to uninstall but, very often, the things on my list can't be uninstalled without breaking the entire desktop.

This situation has come about because we're positioning Linux as a viable desktop/laptop platform for non-specialists -- something it won't ever be. At the same time, we're making Linux less appealing for people who do need the technical edge it offers.

Rather than just complain about this situation, I've decided to review two low-complexity Linux distributions -- Alpine and Devuan. These are at opposite ends of the spectrum of lightweight Linux variants: Alpine is mostly intended for containers like Docker, while Devuan purports to be a general-purpose Linux. Are either of these real contenders for those of us who want a less Windows-like Linux desktop?

Alpine

_At the time of writing, the latest Alpine release is 3.12 -- some or all of what follows may be out of date if there is a substantially later version.

Alpine has become very popular for running in Docker containers and the like, because of its very small size and low complexity. Alpine now has a large repository of software, and it's clear that it can be run as a desktop environment. But is it a good idea?

There are two striking features of Alpine, both with far-reaching implications. The first is Alpine's rejection of a full set of command-line tools in favour of an all-in-one BusyBox installation. This has a radical effect on installed size, at the expense of a reduced level of shell support. The second is even more troublesome -- Alpine uses MUSL rather than GLIBC as its standard C library. Since almost every Linux application in existence uses the C library, this has the potential to cause serious compatibility problems. There are good reasons for using MUSL, particularly in a server installation, but it's use on a desktop system is less defensible. More on this later.

Alpine installs easily from an ISO image of about 400 Mb in size. The installation provides only a console environment with a minimal set of tools, but the repository definitions are set up correctly so it's easy enough to install new software -- provided you have a working network connection.

Wired networking is easy to set up, but wifi is a bit trickier. To be fair, the documentation is pretty clear, describing how to configure wpa_supplicant manually, and bring up an interface at the command prompt. This approach is fine as a one-off operation during installation, but I kind-of expected that a more elegant solution would be available after a full set-up.

If there is one, I couldn't find it. But more on that later.

Fundamental power-management operations, like suspending a laptop when the lid is closed, had to be set up manually, by editing ACPI configuration files. I don't mind doing this and, again, it's well-documented. Unless you spend a lot of time fiddling with ACPI, I think you'd struggle to set these things out without the documentation -- the processes are pretty arcane.

It didn't take long to get a functional console environment, with power management, networking, and audio all working, and my usual build tools installed. But, let's face it, how many people work exclusively without a graphical desktop these days?

And that, sadly, is where the problems start. The Alpine documentation on this subject is clearly noted to be out-of-date, applying to the 3.6 release. The procedure for getting X and a basic Xfce4 desktop working seemed straightforward. However, following the documentation left me with a system that was dysfunctional in some important ways. I couldn't find a display manager that worked apart from lxdm -- quite possibly the ugliest piece of software ever produced. Appearance aside, this is a fairly unusable display manager. Even lightdm did not work. I'm aware that "did not work" is a pretty inadequate problem description, but it's hard to articulate better than that -- sometimes I'd log in and just get a blank screen. Power management worked when it felt like it -- sometimes suspending from a menu would work, sometimes it would lock the screen, and sometimes it would do nothing. The Xfce4 power manage applet would crash unpredictably. I'm not sure why the choice of display manager should affect things like this, but it did.

I had to make a number of changes to system initialization that were not documented. For example, I had to arrange things so that the chronyd time synchronizer did not start until there was a viable Internet connection -- it takes ages to time out if there isn't one, and this delays start-up. I had to fiddle about with policykit so that unprivileged users could suspend and reboot the machine. If you want to support special keyboard keys -- like volume up/down and media keys -- you have to install and configure support yourself. It took an hour or two to get things basically working -- which isn't the end of the world. You do have to know what you're doing, though.

With lxdm (ugh) and Xfce4, I had a more-or-less functional desktop, except that some applications I use all the time -- like Mozilla Thunderbird -- were rather flaky. Still, I was able to do development work, and edit documents using LibreOffice. And listen to music and watch YouTube, etc.

Annoyingly, there was no way to configure a wifi connection except at the command prompt. That's OK in a static installation, but a nuisance on a laptop when travelling. The documentation hints at tools to do this, but none of them were documented well enough that I could get them to work. On a desktop system this is perhaps not such a big deal but, on a laptop, it's almost a show-stopper.

The real problem for me, though, was the lack of binary compatibility with things that I couldn't build from source. The official Dropbox client never worked, despite hours of fiddling with libraries -- although my own command-line version did. The BlueJeans and Microsoft Teams video-conferencing applications didn't work -- although they don't always work very well on a mainstream Linux, either.

The sad fact is that a lot of proprietary software for Linux makes assumptions about the Linux installation, and Alpine's design choices defeat those assumptions. I'm used to using Alpine in containers, but I didn't realize how far-reaching would be the consequences of using MUSL as the C library. I know that it's possible to get Dropbox working on Alpine, but it's extremely fiddly -- far more fiddly that I have time for. The more I used Alpine on a laptop, the more I noticed compatibility problems.

I must point out that none of this should be taken as a criticism of Alpine. Its maintainers do not position it as a desktop system, and it isn't a priority. As a base layer for a Docker container, Alpine is perfect. I can imagine using it on a Raspberry Pi, although I have my own custom installation for that.

In a container, the difference in image size between 10Mb and 100Mb could be highly significant. In a desktop system, it's irrelevant. It's not even particularly relevant on a Raspberry Pi.

If there were a simple way to run Alpine with full GLIBC compatibility, I'd be willing to persevere with it as a desktop/laptop Linux platform. The light resource usage and lack of unnecessary baggage certainly appeals to me. There's no unavoidable installation of a load of processes that serve no useful purpose. Right now, however, I think I will keep it for containers.

Devuan

At the time of writing, the latest Devuan release is codenamed 'Beowulf' -- some or all of what follows may be out of date if there is a substantially later version.

Devuan isn't billed as a particularly lightweight Linux. Rather, it's promoted as "Debian without systemd". In reality, it is pretty lightweight because the maintainers have removed everything that requires systemd. In modern, mainstream Linux distributions, that means almost everything.

I won't be joining in the systemd flame-war, but I believe that doing without systemd has the potential to create a less complex, less tightly-coupled Linux. The fact that lack of support for systemd removes from consideration most modern integrated desktop environments, like Gnome 3, also leads to greater simplicity. A standard installation of Devuan provide the Xfce4 desktop, which is more than complicated enough for me.

Devuan is a much larger install than Alpine -- the basic desktop download is about 4Gb. After installation of all the tools that I consider essential, it was using about 14 Gb of disk space. That compares favourably with a basic desktop installation of Fedora or Ubuntu but, of course, storage is comparatively cheap these days.

Devuan isn't intended to be minimal, and it still installs a lot of stuff that I deem inessential -- Pulse Audio, for example. Unaccountably, the default installation sets up the saned scanner daemon and all its dependencies. Again, if I needed to use that -- and I guess there are people who do -- I could install it myself.

Unlike Alpine, the basic desktop display actually looks good, in a millennial kind of way. An exception is that it uses the ancient xscreensaver for screen-locking, but I guess there's no robust alternative that doesn't have tendrils into Gnome or something of that kind. I had to fiddle about with a lot of the font settings to match my screen, but it was mostly clear how to do that. The standard installation provides LibreOffice, but it can be uninstalled without breaking anything else if you don't want it. I found it very difficult to get decent font settings in the LibreOffice user interface. The reason, I think, is that LibreOffice reads a GTK configuration file that isn't present except on Gnome desktops.

The wicd connection manager is included and preconfigured, so managing wifi connections is straightforward.

Because Devuan uses a mainstream C library and tools, binary compatibility with proprietary software isn't a big problem -- well, no worse a problem than it always is in Linux. I had no problems running the proprietary Dropbox client, for example (or, at least, no more problems than I always have). I can't use the Dropbox shell integration, because it only works with Gnome; but that's not a show-stopper.

When I tried Devuan on a laptop, it didn't support suspending automatically when closing the lid. The setting in the Xfce4 power manager that was supposed to do this had no effect. So back to the configuration files. That's fine but, unlike with Alpine, this wasn't documented, and I had to figure it out myself.

I found a few instances where the software in the Devuan repository was very out of date, compared to that available for Fedora or Ubuntu. On the other hand, some of this software is fairly specialized, like the Arduino development tools, and I was somewhat surprised to find it in the repository at all.

In the end, Devuan is something that I can envisage using long-term on a desktop or laptop computer. I wouldn't recommend it to somebody who isn't already very familiar with Linux, but I don't think its limitations will amount to a significant problem for a Linux developer.

So where does this leave us?

To some extent, using these low-complexity Linux distributions is like stepping back twenty years in time. Most things can be made to work, but it takes time and dedication. There's an awful lot of configuration-hacking. The Devuan installer configures an old-school screensaver, that actually wastes energy by displaying animations -- just like we did before we realized that CPUs burn fossil fuels.

Using Xfce4 again makes me appreciate how slick Fedora with Gnome 3 is. With Fedora, everything just works, straight after installation. Even the display fonts get configured to match the screen. Shutting the laptop's screen suspends it by default -- and, if you don't want that behaviour, it's easy to change. the screen dims and then blanks if you aren't doing anything. Network connection management is built right into the desktop. And so on.

But that level of integration comes at a high price, and it's not a price

that I want to pay.

Gnome, in particular, is very opinionated. The developers have decided what is good for us, and made it difficult to do things any other way. it's very Windows-like to have to install extra software just to get the window manager to respond to keystrokes the way I want. Reducing the configurability of a piece of software makes it easier to support and maintain, but garners no friends among expert users.

Moreover, mainstream desktop Linux distributions require enormous resources -- of storage, memory, and CPU. It's very nice that, when you cycle between windows, the window manager draws a little thumbnail of the windows' contents. But it's certainly not essential, and I doubt it will increase your productivity.

My Windows-using friends and colleagues do admire how nice Gnome 3 looks, but they don't go away and install Linux. Nor are they ever going to.

There's only so much Linux expertise in the world, and a lot of it is unpaid. People will work on things that bring them rewards -- enjoyment, status in the developer community, increased employability, whatever. It's much more fun to write a utility that animates desktop workspaces onto a spinning sphere, than it is to fix a power management bug in Xfce4. So long as Linux is seen as a viable platform for non-expert users, the most rewarding development efforts will be in supporting that kind of user. There isn't enough expertise left over for improving fundamental, not-highly-visible, aspects of Linux. In the long term, this can only be bad for Linux, and for the open-source movement in general.

If you're sceptical about the direction that Linux development is going, you should consider supporting projects like Alpine and Devuan (and Gentoo, and others). They might not be perfect, but at least they stand some chance of becoming perfect enough. I don't think mainstream Linux has any chance at all, since it's headed into a dead end, taking the developer community with it.

[ Last updated Tue 22 Feb 19:28:25 GMT 2022 ]