NetBSD 11 desktop on a Raspberry Pi 4 | Hacker News Reader

NetBSD 11 desktop on a Raspberry Pi 4

neongd.com

121 pointsjaypatelani71 comments

As the saying goes. Of course it runs NetBSD.

They have a nice wiki page around running on Arm which might be helpful (and subpages for the main boards).

https://wiki.netbsd.org/ports/evbarm/

I set it up on an Arm SBC last year and I have to say it did not seem as involved a process as that blog post. I hope it hasn't become more complicated.

NetBSD is so cool. I love how broad the hardware support is, and how straightforward the whole architecture is. My only complaint is that OpenZFS isn't really a thing yet, and the built-in implementation is very outdated. Otherwise I'd have it on my Orange Pi running my NAS.

I've been trying NetBSD 11.1 on an RPi1 these last few days.

It takes a long time to install anything and I don't expect to surf the web on it, but it's nice to have this OS as an option.

I wanted to try out openconnect-saml on it, but the install failed because of lack of RAM/swap. I have managed to run xfreerdp over openconnect on it though (FreeBSD had better performance).

What's the boot time like with that setup?

2 to 3 minutes.

I'm using a Class 10 card. Maybe a better card would speed things up.

I have been using a RPi1 B+ (512MB RAM) for 2 months now as DHCP server for my local network. It is still running 11.0, sysupgrade ran fine but I still see version 11.0 after rebooting, I have to figure out what I am doing wrong.

Software installation is indeed slow, my booting times are not terrible though (running just sshd, ntpd and dhcpd probably helps).

I've been dealing with NetBSD on VMs and my desktop lately and it's really lovely. By far the fastest install and boot of any OS I've tried on my system

I really love the amd64 "live" .img they provide too. It's a disk image you just write to a disk and on its first boot it expands the image to the full size of the disk, reboots, and your system is ready to go

For anyone curious to try, do note that EFI boots significantly faster than BIOS. I forget exactly why but BIOS boots requires an unpacking step every boot that EFI boots do not

The one thing keeping it from my daily driver on my desktop was my GPU is not yet supported. I do love the OS though. PkgSrc is nifty

I've been playing with*

> pkg_add pkgin

NetBSD noob question: what’s going on with this command? Is pkg_add not a package manager? If you already have pkg_add why do you need pkgin? I am familiar with `pkg` on FreeBSD, and `pkg_add` on OpenBSD, but they are both perfectly good enough on their own, I never wanted to install another package manager.

If you're familiar with Debian tools pkg_add is like dpkg in this case it's used to install the package containing pkgin which is like apt.

Pkg_add simply installs binaries you ask it to and dependencies but can't do searches or broader package management like pkgin.

how does it find pkgin then? it’s not pkg_add ./pkgin but presumably downloads it?

or is it just some special case?

PKG_PATH is set and pkg_add automatically adds details to make it point to a package.

With PKG_PATH set (he does it in the .profile in the blog post)

pkg_add pkgin becomes pkg_add https://cdn.NetBSD.org/pub/pkgsrc/packages/NetBSD/$(uname -m)/$(uname -r | cut -d_ -f1)/All/pkgin

So you can also just use the second command and not set PKG_PATH, or atleast not for every session.

I prefer to do it that way since pkgin has its own repository config, and if I use pkg_add after installing pkgin it'll be to install packages from the pkgsrc and that might clash if PKG_PATH is set.

I know it seems contrived.

https://www.netbsd.org/docs/pkgsrc/using.html

If you are familiar with FreeBSD, then the parallel answer to the others in this thread is that on FreeBSD there's the bootstrap pkg that you get in base and the pkg that's in the ports tree itself. It's not quite the same, but the general idea holds that there's a less featureful package installer in base and a more featureful one in the package system itself.

* https://man.freebsd.org/cgi/man.cgi?query=pkg&manpath=FreeBS...

* https://freshports.org/ports-mgmt/pkg/

got it thanks.

I think the biggest problem with many of the non-Linux open source kernels these days is high idle CPU consumption — on both real hardware, and under virtualization.

Did latest OpenBSD & Haiku recently under KVM. Idle OBSD was using around 20% CPU to do nothing. Haiku, 30%. And it’s not just a virtualization thing. Thinkpad battery test yielded similar results.

That's the blocker for me too. I'd love to roll NetBSD, OpenBSD or Haiku.. but how well does it work with hardware, energy consumption and resource management? It's really just the mainstream OSes or nothing, sadly.

Despite the high idle, Haiku was really impressive. Had all of my staples (vim, tmux, git, ssh), recent versions, in their package manager, and it worked beautifully. Stable, rapid booting, responsive.

funny, most people, when they free up a Raspberry Pi, install Home Assistant or Pi-hole.vThis guy installs NetBSD 11.

He spends half the tutorial getting Wi-Fi working, then installs a modern desktop environment styled after a desktop environment from 1993.

next level nostalgia

He's got me thinking I could recreate something that looks like 4dwm and put my NetBSD Pi into a defunct O2 case. A "good" O2 from 96-97 would have been 200MHz w/ 256MB of RAM and a 1600SW display (1600x1024).

I see in the list of NetBSD packages that the parts for a fairly good Pi-Hole replica seem to be there. This writer seemed to need WiFi and a desktop, but of course neither is necessary for a Pi-Hole type of host.

https://ftp.netbsd.org/pub/pkgsrc/current/pkgsrc/index-all.h...

EMWM would have been far better and lighter than NsCDE, even if compiled from mainline releases with everything: emwm, xfiles, ximaging...

The pixel art font is so bad I can only read the code snippets

But it's not AI slop! /s

yes, the irony

Read the original gemini version then. I have not looked at it, but gemtext has no support for setting fonts at all, so it can not use a font you dislike unless you set that up in your gemini client.

Yeah, exactly. I gave up very quickly.

I read this before opening the page, and I figured it couldn't be that bad. I instantly felt my eyes straining to read the introduction.

Reader view in Firefox does the job.

It looks like a pretty standard bitmap font to me. Very legible to my eye. What’s wrong with bitmap fonts?

> Very legible to my eye.

Your retort is ableism. Some people have vision problems such as myself and I find reading the cute yet blocky font to be very uncomfortable. Thankfully, reader mode fixes this.

People should be able to express themselves how they like

bitmap fonts were the norm on computers for many years, and are still the default if you are running a computer in text mode (ie not running X11 or Wayland on linux / BSD), or using basically any low powered device with a screen. I was surprised to hear that anyone had trouble reading these, since they were just what fonts looked like on screens with certain constraints. I can understand that some people have become so used to high DPI displays and antialiased truetype fonts that they consider fonts with visible pixels ugly. If anything I would expect them to be easier to read with poor vision, since there is not much subtlty to miss out on.

English-only was also the norm for many years. First you make it work, then you make it accessible. But there's no reason to dial back on accessible formats once you have them.

the accessible format here is one in which reader mode works for those who prefer it

These fonts are low resolution and looks blurry unless I focus on it more.

Try to have this font and regular one next to each other (scroll to the the code snippets)

now start moving further away from the screen till you can barely read it.

Which one becomes unreadable first?

Accessibility is not about if you can read it... but can other people read it?

It isn't a bitmap font. It's Berkelium64, which is a re-work of a Commodore GUI font for GEOS with a fair amount added. It's proportional, rather than monospace, and it only looks like it is a bitmap.

* https://www.kreativekorp.com/software/fonts/c64/#berk

I switched the page over to unscii, and I found it to be a lot more legible.

In Firefox settings menu:

Accessibility, Fonts, Advanced Settings, Allow pages to choose their own fonts.

Presumably other browsers let you do the same. I don't usually allow pages to choose their own fonts, but I turned it on specially just for this page and it looks like a normal pixel font to me.

"Get Balena's Etcher (if you don't have it)"

You lost me there. Why are ppl still using this electron monstrosity when there are so many better alternatives? (Like dd, WIN32Imager, USBImager, DiskImager, etc.)?

I mean, Etcher is bigger than the entire NetBSD image, what the hell?

dd and Etcher do not exactly have the same target audience.

Given the topic, I'd say the target audience preference would likely be thus:

Admin/root superuser of rpi4: dd

Admin/root superuser of NetBSD, other BSDs, or linux: dd

> dd and Etcher do not exactly have the same target audience.

But the others do. They are all minimal GUI apps.

You usually don't even need dd, plain cat <iso >device is fine, or pv if you want to be fancy and want progress bars.

Less of a chance of swapping if/of and destroying something unexpected.

This is one of those things that SystemD can make more complicated. When you DD a disk image on something it rewrites the disk header, which can cause SystemD to reparse the header and do some weird stuff with the device that causes the write to fail or get corrupted. Etcher works around that problem.

That said, the more lightweight "Disk Image Creator" also gets around this problem without dragging in an entire self contained browser.

What is this "systemd" you speak of? Getting in between a man-as-root and his dev nodes? Sounds most un-unixlike.

It's IBM's way of embrace-extend-extinguish'ing freedesktop

> That said, the more lightweight "Disk Image Creator" also gets around this problem without dragging in an entire self contained browser.

Exactly my point. Also that self contained browser hides a spyware (telemetry without notice and opt out), so illegal in the EU.

> You usually don't even need dd, plain cat <iso >device is fine

Definitely not. In fact, even dd is insufficient unless you use os specific flags to flush the disk cache, otherwise your stick might be corrupted. Read about the fdatasync syscall.

Okay, so cat && sync

I like it. It’s cross platform, so you can recommend it to anyone. It prevents you from writing to system disks accidentally. It verifies the image after writing.

Who cares if it’s an electron monstrosity? It’s a portable executable, you run it once and then never again.

> It’s cross platform, so you can recommend it to anyone. It prevents you from writing to system disks accidentally. It verifies the image after writing.

USBImager likewise, cross-platform, stupid-proof, verifies write, decompresses on-the-fly, but it's multilingual too and does not collect your data like Etcher. All of that in just 300 kilobytes in a single portable executable.

https://bztsrc.gitlab.io/usbimager/

I prefer to recommend that instead to anyone.

I'm not who you're replying to, but maybe alternatives need better marketing. However I'm going to check this out and if I have a good experience I'll recommend it to people I know in the future instead. Thanks.

I’ll try that next time :)

There are a few too many Linux builds, imo, that I wouldn’t feel comfortable recommending it to a newbie Linux user. It looks like on Linux that the authors are obsessed with making the smallest possible binary instead of the smallest possible binary that runs anywhere.

Can one point me to the lines of code in the Linux kernel that contain AI slop?

They're probably commenting on Linus Torvald's decision to accept commits that may or may not have been assisted with agentic coding in part or whole, getting accepted/rejected on a merit basis instead.

(Apparently that's too much gray area for some zealots)

Are they in the room with us right now?

Sure:

  (git ls-files; git log --grep='^Assisted-by: ' --pretty='format:' --stat --name-only | sort | uniq) | sort | uniq --repeated

...will show you the names of all 6,251 tainted files. You can `git blame` them at your leisure.

"tainted". Can you point to something in any of those 6251 files that is "slop"?

Linus himself recently worked on a cache contention issue and used AI to assist.

https://www.tomshardware.com/software/linux/linus-torvalds-p...

Looks like quality code, not slop.

I find it hilarious that the author deliberately chooses to talk the novice user through the use of Vim (including how to quit it) rather than suggest something simpler such as pico or nano.

Or the newly released atto(1)

https://man.netbsd.org/atto.1

Afaik pico and nano aren't installed by default on NetBSD but vi is

I'm using vi for twenty years now, mostly because I still haven't figured out how to close it.

Ngl nano opened the other day on a new system and I panicked. Took me more than a few seconds to figure out how to save and exit.

Yeah that gradually dawned on me when I read the comment about the recently released atto text editor.

It's not vim. It's Bostic's nvi. And in fact one of the things that vim users need to be reminded about is that the habitual :wqa and :qa are vim-isms and do not work in other clones such as nvi.

Fair enough, I stand corrected. The point I was trying to make though (and clearly I failed, since I’ve been heavily downvoted) is that the poco/nano/atto ‘lineage’ of “Microsoft Edit”-like text editors are quite a bit more approachable for newcomers than the holy-war dichotomy of vi and emacs would seem to imply.

One of the things that FreeBSD has had for a long time is the 'Easy Editor', written in 1990 for HP-UX and added to FreeBSD a couple of years later. There was even its successor 'Another Easy Editor' which the world has completely forgotten about.

* https://man.dragonflybsd.org/?command=ee&section=1

* https://man.freebsd.org/cgi/man.cgi?query=ee&manpath=FreeBSD...

But NetBSD and OpenBSD went their own ways on this, to the extent that NetBSD didn't even have a non-vi-like editor in base until just a month or so ago.

Yeah I just found it genuinely amusing that the author felt a duty to digress into quitting vi for the sake of their audience but I’ve been downvoted and apparently people must have very strong opinions on this and I apologise for any offence caused.

[sobs in ed(1)/ex(1)]

Reply on news.ycombinator.com