The happy path has improved a lot. When Linux is working it's reasonably usable. But once something breaks it breaks HARD and recovery is still miserable.
For reference I've been using Linux since Red Hat 5.2 circa 2000. I cut my teeth debugging problems without internet access. I ran an LTSP lab at my high school. I remember the hell that was XF86Config (I was there, Gandalf, I was there 3000 years ago).
....and like the previous commenter I run Windows on my personal machines because I want to spend my free time using them, not debugging them.
....which is why Chromebooks took over the consumer market. Oh, wait.
I’ve certainly run into some odd situations on my desktop Linux machine over the past 6 years since I started using it full time, but I think most of them were related to the nature of how I use the machine more than inherent instability. I think I’ve spent many more hours of my life unwinding piles of malware and bloat from non-technical folks’ Windows machines than debugging this one.
They don’t know what Linux is, and know nothing about tech, they just know that we had a 30 minute lesson on “here’s Firefox, this icon means you need to install updates, here’s how you print”.
Oh and this was Linux Mint back in ~2016
Things have only gotten easier since then
I have been using Linux since 2000s as well. I do remember the rpm hell, dealing with x config issues etc. It is NOT the same experience now a days. I don't have the time or inclination to mess around so I use Fedora + KDE and that basically stays out of my way. I don't rice my desktop or do any hacking around beyond basic automation and I have had zero instances of the system just breaking.
* I wanted to update a Raspberry Pi from Ubuntu LTS 22 to LTS 24. Turns out this is basically impossible. Ubuntu themselves tell you not to do it and their recommended solution is to wipe the system and try again. I ignored them and tried to do it anyway and my Pi ended up refusing to boot. Great.
* I needed to update a Raspberry Pi to change the list of WiFi networks it knew about. Except apparently there are two different networking stacks for Linux with different config files and I edited the wrong one.
* I built a new TrueNAS server. Turns out that you absolutely cannot configure the networking from the GUI. There's a section there, sure, but every time it refuses to save the information until you "test the changes" and that fails to reconnect every single time. You have to locally plug a monitor into the machine, boot it, and log in with a keyboard to get to the config there.
* Not strictly a bug, but I installed Debian in WSL and it doesn't include `man` by default. So I get a command line and no help for it. Brilliant.
* I want to install jj
* Its docs say to use cargo-binstall
* How do I get that? With cargo, so sudo apt install cargo
* `cargo binstall --strategies crate-meta-data jj-cli` -> `error: no such command: `binstall``
* `cargo install binstall` -> `error: cannot install package `cargo-binstall 1.17.7`, it requires rustc 1.79.0 or newer, while the currently active rustc version is 1.75.0`
* `sudo apt install rust` -> E: Unable to locate package rust
* `sudo apt install rustc` -> `rustc is already the newest version (1.75.0+dfsg0ubuntu1~bpo0-0ubuntu0.22.04).`
Apparently the guidance is to manage your rust versions with a tool other than apt that you install with `curl ... | sh` because no one ever learns anything about security
.....yep, just as user friendly as I remember.
The other path I can see (looking at https://docs.jj-vcs.dev/latest/install-and-setup/#windows) is that you could maybe instead use winget directly.
Though honestly IMO this is more of a failure on the jj devs to not provide something that can be installed straight using apt, I guess (looking at https://docs.jj-vcs.dev/latest/install-and-setup/#linux). For Arch for example you just install it from the official repos.
"Ubuntu themselves tell you not to do it" - you do see it right? Let us see how you forgive Windows for breaking things by ignoring Microsoft's advice and blame them anyway when it breaks.
> * I needed to update a Raspberry Pi to change the list of WiFi networks it knew about. Except apparently there are two different networking stacks for Linux with different config files and I edited the wrong one.
Why? Why not connect it to the network you want so that it just connects to that going forward?
> * I built a new TrueNAS server. Turns out that you absolutely cannot configure the networking from the GUI. There's a section there, sure, but every time it refuses to save the information until you "test the changes" and that fails to reconnect every single time. You have to locally plug a monitor into the machine, boot it, and log in with a keyboard to get to the config there.
And TrueNAS's shortcomings are somehow Linux's fault just like every Windows thirdparty software issue is Windows' fault?
> * I want to install jj * Its docs say to use cargo-binstall
No, they don't ask that as the first choice - this is what they say in https://docs.jj-vcs.dev/latest/install-and-setup/:
Installation¶
Download pre-built binaries for a release¶
There are pre-built binaries of the last released version of jj for Windows, Mac, or Linux (the "musl" version should work on all distributions).
Cargo Binstall¶
If you use cargo-binstall, ....
$ > curl -L --proto '=https' --tlsv1.2 -sSf https://raw.githubusercontent.com/cargo-bins/cargo-binstall/main/install-from-binstall-release.sh | bash
+ set -o pipefail
+ set -o pipefail
+ case "${BINSTALL_VERSION:-}" in
++ mktemp -d
+ cd /tmp/tmp.8IdPJtQBlE
+ '[' -z '' ']'
...
+ case ":$PATH:" in
+ '[' -n '' ']'
$ > cargo binstall --strategies crate-meta-data jj-cli
INFO the current QuickInstall statistics endpoint url="https://cargo-quickinstall-stats-server.fly.dev/record-install"
Binstall would like to collect install statistics for the QuickInstall project
to help inform which packages should be included in its index in the future.
If you agree, please type 'yes'. If you disagree, telemetry will not be sent.
...
INFO resolve: Resolving package: 'jj-cli'
WARN resolve: When resolving jj-cli bin fake-bisector is not found. But since it requires features test-fakes, this bin is ignored.
WARN resolve: When resolving jj-cli bin fake-diff-editor is not found. But since it requires features test-fakes, this bin is ignored.
WARN resolve: When resolving jj-cli bin fake-echo is not found. But since it requires features test-fakes, this bin is ignored.
WARN resolve: When resolving jj-cli bin fake-editor is not found. But since it requires features test-fakes, this bin is ignored.
WARN resolve: When resolving jj-cli bin fake-formatter is not found. But since it requires features test-fakes, this bin is ignored.
WARN The package jj-cli v0.39.0 (x86_64-unknown-linux-musl) has been downloaded from github.com
INFO This will install the following binaries:
INFO - jj => /home/xxxxx/.cargo/bin/jj
Do you wish to continue? [yes]/no yes
INFO Installing binaries...
INFO Done in 7.549505679s
$ > jj version
jj 0.39.0-d9689cd9b51b4139d2842fcf6c30f65f4eed8cd1
$ >
Having said all that, none of this stuff you mentioned even remotely resembled an average user's workflow who just uses his computer for listening to music and browsing the internet with some occasional document editing thrown in. Despite its warts and shortcomings, Linux does a much better job today than it used to.
Not giving a supported upgrade path between version N and N+1 of your operating system is unacceptable, user hostile, and not something a home user could deal with. "Install from scratch, wipe all your files, and set everything up again" is not OK. You can upgrade Windows from 1.0 through 11 without Microsoft saying "nah, this is impossible": https://www.youtube.com/watch?v=cwXX5FQEl88
> No, they don't ask that as the first choice - this is what they say in https://docs.jj-vcs.dev/latest/install-and-setup/:
"the binaries" are a tarball whose instructions refer back to the previous document, whose "Install > Linux" section starts "from source" and says "go obtain Rust > 1.88", so all of the previous problems still apply.
Again with the assertions without checking things. This is the path of "the binaries":
https://github.com/jj-vcs/jj/releases/tag/v0.39.0
I downloaded the file myself and extracted to see:
$ > ls -l jj-v0.39.0-x86_64-unknown-linux-musl.tar.gz
-rw-r--r--. 1 xxxx xxxx 10373711 Mar xx xx:xx jj-v0.39.0-x86_64-unknown-linux-musl.tar.gz
$ > tar xzvf jj-v0.39.0-x86_64-unknown-linux-musl.tar.gz
./
./README.md
./LICENSE
./jj
$ > ls -l jj
-rwxr-xr-x. 1 xxxx xxxx 27122184 Mar 5 02:33 jj
$ > file jj
jj: ELF 64-bit LSB pie executable, x86-64, version 1 (SYSV), static-pie linked, BuildID[sha1]=70d48428bc2100069e6813aff97e3dce8d2bb4a0, not stripped
$ > ./jj version
jj 0.39.0-d9689cd9b51b4139d2842fcf6c30f65f4eed8cd1
$ >
Not "figure out how to extract a tarball, find somewhere unspecified on your path to put things blah blah" but "to get started go read this doc whose first step is to install rust, which your package manager isn't capable of".
This is a fairly standard Linux experience, not one reserved for developer tools.
On Windows, if you're not going through an app store you get an EXE or MSI installer that you double click and it does everything else necessary. Every time.
I note you also dropped the line of argument about the OS updating, where you were chiding them, saying they did need to follow instructions in that case. Of course, the instructions in that case are indefensible - you cannot seriously suggest an OS is production-ready for the real world if the instructions are "this cannot be updated. Seriously, don't even try.".
Yes, because as per the poster, they are not a novice:
> For reference I've been using Linux since Red Hat 5.2 circa 2000. I cut my teeth debugging problems without internet access. I ran an LTSP lab at my high school. I remember the hell that was XF86Config (I was there, Gandalf, I was there 3000 years ago).
No one is expecting a novice to know how to run curl, untar and compile. This is not that situation by the very admission above.
> For somebody who just started using it, "following the install instructions at the top of the page" is a perfectly reasonable thing to be doing. It is not the user's fault if those instructions are bad and you could totally get it working more easily if only you already knew what you were doing.
Did you actually go to jj's github which the poster mentioned? This is what is literally the top of the Installation page:
Installation and setup¶
Installation¶
Download pre-built binaries for a release¶
There are pre-built binaries of the last released version of jj for Windows, Mac, or Linux (the "musl" version should work on all distributions).
> I note you also dropped the line of argument about the OS updating, where you were chiding them, saying they did need to follow instructions in that case. Of course, the instructions in that case are indefensible - you cannot seriously suggest an OS is production-ready for the real world if the instructions are "this cannot be updated. Seriously, don't even try.".
I admit that I was shallow on this point. I did research further and Raspberry Pi situation isn't great when it comes to upgrades. Most people are using separate SD cards to host the OS and doing a hard upgrade. I admit and apologise to @Arainach for not checking further on this point and ignoring it.
Edit: I guess today was the day I couldn't ignore Linux bashing from an experienced user and got somewhat carried away. My tone could and should have been softer.
I'm not the guy who wrote that, but I had the same use-case myself. (Except that I happened to choose the correct networking stack so I didn't have a problem). I wanted to set up a Raspberry Pi in my parents' house that would run Tailscale so I could use it as an exit node. (With my parents' full knowledge and permission). I wanted to pre-configure it with their WiFi password so that when I showed up for Christmas, I didn't have to spend any time configuring the device, just plug it in and go have dinner. (Then they changed ISPs, got a new router with a new WiFi password, and I had to ask them to plug it into the wired network so I could connect to it remotely and change the WiFi password again, so I had to do that work twice. But thankfully, I didn't have to walk them through the steps, just say "Hey, please plug it into the router with an Ethernet cable until you get an email from me telling you I've reconfigured the WiFi".)
As a separate point, it seems quite feasible to run Android apps in VM on Linux based phone and make the experience fairly seamless. Something like what Waydroid provides.
But why?
The premise of Waydroid seems to be to bring Android apps you want to your Linux desktop. But why would you want the phone in your pocket to run Desktop Linux so that you could then run Android apps on your Desktop Linux mobile phone instead of just running Android on your phone?
What desktop Linux features do you want on your phone that would justify this complexity?
My hope is that installation of the Android apps on Linux phone could be made seamless.
I’m sure Google would deny Google Play Services to any popular fork that didn’t follow their rules. But they would do the same to any Linux desktop or whatever that didn’t follow their rules, too, if it became popular.
There are open UI shells from KDE and GNOME, multitouch gesture support, Android emulation... it's all there.