I’ve been experimenting with what a next-generation embedded Linux build system might look like: native builds on the target architecture, modern language package managers as first-class citizens, and AI as a primary interface to the system.
Instead of cross-compiling with a large meta-layer stack, the tool builds kernel, rootfs, and applications together using one engine, with a CLI, TUI, and AI assistant talking to the same core. All you need is the tool, Docker, and Git — no global SDKs or hidden state.
It’s pre-1.0 and rough around the edges; I’m sharing it early to get feedback from people who live in Yocto/OpenEmbedded, Buildroot, Nix, etc. I’d love to hear where this breaks on your boards, what workflows feel wrong, and whether the “native builds + AI-aware build graph” direction seems promising.
I wish you luck! The pain points you identified are definitely real and solving them would be valuable.
The workflow for user space can definitely improve some of this pain but I feel like a large portion of any embedded Linux development effort still ends up in the weeds for boot related items (secure boot, proper updates, nuanced kernel patches, bootloaders, device trees, and supporting machine variants, etc). Solving those to make them easy is a hard problem for sure.
Armbian and Qemu worked well for me when I needed to compile packages for an orange pi without enough RAM to actually run cargo build. Built the image on an emulator with more RAM then the target system, dropped the customized image on the SD card and booted right into the entry script.
(For the unaware, OP above is the article author.)
As a long time user of Bitbake and Yocto, and watcher of Yoe Linux... I kind of have to respectfully disagree with the approach.
One look at the Python or Node packaging ecosystem, and you can see how difficult trying to integrate those in a 'sane' way in embedded, with repeatability and security in mind, nevertheless wrangling something like C/C++ dependencies and native packages in a qemu-cross (or binfmt emulation) environment.
I feel like Bitbake and the wider Yocto ecosystem has essentially 'solved' cross compiling. Sure there are the incredibly complex codebases like Chromium that require lots and lots of study and patience, or esoteric compilers, etc. but for most applications I feel that especially AI tooling can write up a good basic recipe for integration with Yocto.
The unfortunate thing about AI tooling is Kernighan's law (where debugging is twice as hard as writing it the first time.) Especially in Embedded Linux, where 99.99% of the product code is written by others, trying to figure out where the AI didn't quite get it right or missed something can break things unexpectedly and impossibly for the unaware developer to fix, even breaking at runtime. So the build environment has to be simple and predictable. I think Yocto/Bitbake already strikes a good balance here, with other nice things like offline builds, repeatable builds, sstate caching across machines, PRbuild servers... things useful for big teams, so consider those as well.
Although, if someone does a 'uv'-like rewrite of Bitbake into Rust with the same feature set but faster, I am all for it...
Maybe, having Yoe or some other centralized distro make specific opinionated choices and then provide an open sstate feed, working inside Bitbake, could get a lot of the speed and customizability gains back without a big rewrite effort.
I share some of the same observations that seem to have motivated this project, but my solution was to just use Alpine on the same architecture as the target (possibly via qemu) and then export a subset of it to a filesystem image. I implemented it as a perl module collection with no dependencies other than core perl modules, and then run the export from within the image itself. Among other things, it lets me use strace to find the runtime library dependencies of the things that need to be in the image.
I've used buildroot extensively, and an occasional yockto.
My impression in recent years is that these image cross build environments are just not as frequently needed as they were back in the day of their invention.
My most recent embedded linux environments were just embedded archlinux.
No need to cross build an image, just install and run the minimized linux environment right on the target.
Of course, a big part of the need for these cross-tools is that it seems most modern embedded linux developers are running windoze on their development workstations 8-/
The problems building with Linux and GNU environments exist because they were terribly designed with assumptions like
- You're building on the same native system as GNU and Linux packages, you install them globally in the same places that servers and desktops use
- Your C, C++ compiler and entire toolchain and other binary utilities with the kernel is a one single unit that you can only change one part at a time
- You use the same up to date headers with glibc, gcc and Linux kernel
- You're building software in the same universe of all the other packages, especially gcc libraries (libgcc_s, libstdc++), glibc (especially bad since ld-linux.so is part of it)
- The build system only uses standard paths
The reason Yocto is so complicated is that developing in a Linux environment actually sucks when you're not writing web-oriented or server / VM software. Yocto fixes it. It introduces a good set of abstractions that work around terrible design decisions that were made in overall Linux ecosystem. There are a lot because the OS design is fundamentally broken, especially with C-based toolchains which is 99.999% of the ecosystem. Current C toolchains including MSVC strongly ties OS with the C's internal types and bad decisions of 70s.
As always all articles whose title asks a question are answered with NO, 99% time. By taking away the cross-compiling abilities and the workarounds doesn't fix the brokenness of Linux and overall FOSS ecosystem.
If you're looking for how a better embedded environment looks like, look at Rust toolchains. For Linux take a look at musl-libc based ones (you 100% need a systemd distro to get away from nss complexities that musl introduces). Or even better take a look at relibc. There are barely any assumptions about the target filesystem and tooling in Rust toolchains, unlike C/C++/Make toolchains. There is redox OS but it is still in slow development and they stuck with Make, which I think was a bad decision. Android uses its own build tooling but cannot run away from C/C++ tooling unless Google revives Fuschia.
I'm surprised to see no mention of Guix. It solves all of these issues, and already has a good story both for cross and native builds on a variety of architectures.
Adding new targets is deceptively easy, just copy an existing template and substitute your values.
I haven't used it, but there's also Avocado OS [1]. Myself, I use Yocto for our embedded build. It's not perfect but for our application, which involves building a bunch of software that uses both Swift and Flutter for an embedded audio product running on a RPi CM4, it works just fine. Shout out to Mender for A/B updates.
Modern SBCs are just normal computers and not "embedded" in the traditional sense. You can generally just use Debian, and spend time on the actual project, instead of wrestling with the system
I've been using Nix for "embedded" systems for few years now and it works fine. I don't quite understand why "embedded" has historically meant learning a completely separate tool. I've been building for x86 and ARM servers for 5+ years now, so why should targeting an ARM board be any different?
Like the article points out, the software stacks people use for embedded devices are the same as people use non-embedded use; Rust, Go, NodeJS, and sure still C++. The only real difference with embedded devices is non-OS components like the bootloader (u-boot, EDK2) and customizing the device tree. (And of course firmware flashing). Fundamentally those are all just packages that I can describe in Nix. I don't need a separate tool just because the board is small.
IMO the embedded space, especially in the US, is already pretty Niche. Most companies either just ship the vendors BSP example (Ubuntu/Debian/Yocto) and pay very little attention to the detail or re-useablity. Once you vendor declares the BSP EOL you are stuck unless you re-implement it yourself.
Using Nix (or Guix) has the massive advantage of a large and active community that isn't fractured like the Yoctoo/Buildroot community. (By fractured I mean there may by many, many people using those tools at $DAYJOB but due to vendor BSP customization they likely share much less with the upstream community maintained sources).
> I don't quite understand why "embedded" has historically meant learning a completely separate tool. I've been building for x86 and ARM servers for 5+ years now, so why should targeting an ARM board be any different?
Because embedded usually means dealing with complete functionality provided by the soc, bare minimum for the init system to get quick boot times.
If you just want to boot the CPU and run some IoT app, device tree + kernel + uboot should be enough.
But for anything with a GUI, You'd probably need the vendor provided binary blobs for GPU accelerated UI, proper gstreamer packages with all the patches to make it work with the binary blobs... properly packaged Qt etc ... Not to mention read only rootfs and A/B partitions for upgrades to deal with power loss headaches.
I haven't used Nix for any of these but most of these things are available ready to use with the vendor provided buildroot or yocto setups. And it would've been just more work to get this all working with yet another build system/package manager
However, reading through the docs, the author clearly knows about Nix/Nixpks and has factored some of its properties into the design: https://docs.yoebuild.org/nix.html
RAM is expensive now. Reasonably priced seems to be counter to that fact.
The whole article feels like a promo-tour.
> The catch is that nobody handed us the tools to keep the place running.
Software such as homebrew exists. Why is the article ignoring this? Not only does it feel like promo, it now feels like propaganda. Homebrew is also just one example; LFS/BLFS also exists for embedded linux. Perhaps the quality is less than for desktop computer systems but it exists.
> Edge devices have started behaving like cloud systems. They run increasingly complex stacks, are frequently updated, and are managed remotely over their entire life.
Companies want to become dependent on others? Because that is what cloud is all about. Top-down control. A mafia for later blackmail rack-up-the-price strategies.
> That puts the cross-compile burden squarely on embedded developers, and maintaining recipes for thousands of packages has been a steady drain on the Yocto community
Seems untrue. If you have a compiler that works, you can compile. Thousands of packages? I am tracking 3888 right now. That covers really most software by far, including python pypi addons. If a single person can do that, a team has an even easier time. The whole article seems to narrate a story that is just made up.
> Building everything from source in Yocto means long builds, heavy memory use, and powerful workstations.
No it does not. I do so every day. You need a reasonably fast computer and RAM, which is now driven up by the existing hardware mafia that must be put in jail, but it is easily possible for solo persons too. Now imagine a team.
The whole article is really really crap. Absolute garbage. The claims it makes are for the most part simply not correct. It narrates a story as if all linux developers are incompetent. I highly doubt that. It also never mentions any existing software here that would change the rhetoric such as homebrew, LFS/BLFS and so forth. It feels as if that article is just engineered to tell a "we thus need xyz". That's not good writing. I mean just look at this:
"We can leverage these change agents and rethink the world of connected products."
What does that even mean? Did AI write this? Because humans are usually smarter when it comes to opinion pieces. "connected products" - everything is now connected?
I wrote the orignal draft, but used Claude to integrate some feedback. In hindsight, this was a mistake, and I will not do that again. Thanks for the frank feedback.
I'm glad Yocto is working for you.
I'm by no means a Yocto expert like many of you, but I've used OpenEmbedded/Yocto since the start, so I'm quite familiar with it. I started the following project, which Khem mostly maintains these days:
Oi, more AI slop. After dealing with things that have external package management and yocto, I understand the pain. But integrating that into yocto is very doable and is a far better path forward.
And "I've been experiment with" seems to just mean "I've been vibe coding"
Whilst I find yocto perfectly fine. I also see that it has its limitations. I think there is a need to make yocto more nimble rather than designing a whole new build system.
yoebuild does seem interesting at least as a proof-of-concept.
This sounds like fun, but I'm not sure how changing build systems fixes the vendor problem. They will just ship their proprietary blobs built on top of this new build system instead of the old one, and will not change other behavior. Buying from vendors who upstream or minimally open source their work might help.
I'm a little doubtful on the post's assertion that yocto is only needed for "deeply embedded regulated products." For products that have to follow the https://en.wikipedia.org/wiki/Cyber_Resilience_Act coming up, they will almost certainly need:
And I'll echo what some other folks have said here: if you don't need features like that, bundle your custom bootloaders and kernel with a binary distribution like Debian and call it a day.
My experience is that Yocto is a bit like CMake: people hate the mess they make with it and blame the system, even though they never bothered learning to use it.
When someone starts with "why cross-compiling the system if it's bearable to compile it directly on the device?", it really sounds like they are just not comfortable with cross-compilation. To me it would be similar to say "why would you need a compiled language, with this weird compilation step that nobody understands, when you can just write in an interpreted language?". Except that developers wouldn't generally proudly admit that they don't know how to use a compiler.
A valid reason I see to require a new embedded linux build system would be if the existing ones (mostly buildroot and Yocto) were unable to build a modern OS (e.g. by not being able to integrate some security features). But I don't think that's the case. This article sounds more like very verbose marketing for a system claiming that it can do similarly without having to learn as much (which I honestly doubt).
To me it's often the same pattern:
* Meson is nice, but it doesn't remove the need to use CMake, and it's not fundamentally simpler or more powerful (it's more comfortable at most).
* Jujutsu is nice, but it doesn't remove the need to learn git, and it's not fundamentally simpler or more powerful.
* Yoebuild sounds like even if it got traction like Meson and Jujutsu, it wouldn't remove the need to interact with Yocto for embedded linux engineers.
As an embedded linux engineer, if anyway I have to interact with Yocto and therefore learn Yocto, why would I learn another one that isn't fundamentally better? Usually people who go for those "simpler" (at least on the surface) tools do it because they don't want to learn Yocto. And for an embedded linux engineer, isn't it a red flag to not want to learn the most famous embedded linux build system?
It reminds me of Balena/Resin, which seems like it's targetting people who are not embedded linux engineers, don't want to become embedded linux engineers, but still want to distribute an embedded linux. To me that's generally a red flag.
Joke: The Linux community will continue to overlap and reinvent parts of itself until there is a universal Linux kernel that can be fully adapted and customized to fit onto almost any hardware..
Architecture emulation with qemu-userspace is 5-10x slower than running a native build.
The only 'proper' no-friction cross-compilation system I know is Debian/Ubuntu, because it allows to install foreign architecture libraries into your native system, and has all the wrappers for cross-compilation in dpkg-buildpackage.
This way you don't have to maintain cross-compiler or toolchain/sysroot, you just install whatever dependencies you need with the regular `apt install` of another architecture.
I'd like to have more friendly tooling around that instead of a slow compilation of architecture emulation.
33 comments
[ 2.8 ms ] story [ 58.9 ms ] threadInstead of cross-compiling with a large meta-layer stack, the tool builds kernel, rootfs, and applications together using one engine, with a CLI, TUI, and AI assistant talking to the same core. All you need is the tool, Docker, and Git — no global SDKs or hidden state.
It’s pre-1.0 and rough around the edges; I’m sharing it early to get feedback from people who live in Yocto/OpenEmbedded, Buildroot, Nix, etc. I’d love to hear where this breaks on your boards, what workflows feel wrong, and whether the “native builds + AI-aware build graph” direction seems promising.
The workflow for user space can definitely improve some of this pain but I feel like a large portion of any embedded Linux development effort still ends up in the weeds for boot related items (secure boot, proper updates, nuanced kernel patches, bootloaders, device trees, and supporting machine variants, etc). Solving those to make them easy is a hard problem for sure.
This is the complete opposite way, actually.
We need cross-compiling that is just as effortless as native compilation.
You should be able to build complex software on a powerful computer and perform costly optimization, then run it on a low-powered device.
As a long time user of Bitbake and Yocto, and watcher of Yoe Linux... I kind of have to respectfully disagree with the approach.
One look at the Python or Node packaging ecosystem, and you can see how difficult trying to integrate those in a 'sane' way in embedded, with repeatability and security in mind, nevertheless wrangling something like C/C++ dependencies and native packages in a qemu-cross (or binfmt emulation) environment.
I feel like Bitbake and the wider Yocto ecosystem has essentially 'solved' cross compiling. Sure there are the incredibly complex codebases like Chromium that require lots and lots of study and patience, or esoteric compilers, etc. but for most applications I feel that especially AI tooling can write up a good basic recipe for integration with Yocto.
The unfortunate thing about AI tooling is Kernighan's law (where debugging is twice as hard as writing it the first time.) Especially in Embedded Linux, where 99.99% of the product code is written by others, trying to figure out where the AI didn't quite get it right or missed something can break things unexpectedly and impossibly for the unaware developer to fix, even breaking at runtime. So the build environment has to be simple and predictable. I think Yocto/Bitbake already strikes a good balance here, with other nice things like offline builds, repeatable builds, sstate caching across machines, PRbuild servers... things useful for big teams, so consider those as well.
Although, if someone does a 'uv'-like rewrite of Bitbake into Rust with the same feature set but faster, I am all for it...
Maybe, having Yoe or some other centralized distro make specific opinionated choices and then provide an open sstate feed, working inside Bitbake, could get a lot of the speed and customizability gains back without a big rewrite effort.
https://metacpan.org/pod/Sys::Export
My impression in recent years is that these image cross build environments are just not as frequently needed as they were back in the day of their invention.
My most recent embedded linux environments were just embedded archlinux.
No need to cross build an image, just install and run the minimized linux environment right on the target.
Of course, a big part of the need for these cross-tools is that it seems most modern embedded linux developers are running windoze on their development workstations 8-/
- You're building on the same native system as GNU and Linux packages, you install them globally in the same places that servers and desktops use
- Your C, C++ compiler and entire toolchain and other binary utilities with the kernel is a one single unit that you can only change one part at a time
- You use the same up to date headers with glibc, gcc and Linux kernel
- You're building software in the same universe of all the other packages, especially gcc libraries (libgcc_s, libstdc++), glibc (especially bad since ld-linux.so is part of it)
- The build system only uses standard paths
The reason Yocto is so complicated is that developing in a Linux environment actually sucks when you're not writing web-oriented or server / VM software. Yocto fixes it. It introduces a good set of abstractions that work around terrible design decisions that were made in overall Linux ecosystem. There are a lot because the OS design is fundamentally broken, especially with C-based toolchains which is 99.999% of the ecosystem. Current C toolchains including MSVC strongly ties OS with the C's internal types and bad decisions of 70s.
As always all articles whose title asks a question are answered with NO, 99% time. By taking away the cross-compiling abilities and the workarounds doesn't fix the brokenness of Linux and overall FOSS ecosystem.
If you're looking for how a better embedded environment looks like, look at Rust toolchains. For Linux take a look at musl-libc based ones (you 100% need a systemd distro to get away from nss complexities that musl introduces). Or even better take a look at relibc. There are barely any assumptions about the target filesystem and tooling in Rust toolchains, unlike C/C++/Make toolchains. There is redox OS but it is still in slow development and they stuck with Make, which I think was a bad decision. Android uses its own build tooling but cannot run away from C/C++ tooling unless Google revives Fuschia.
Adding new targets is deceptively easy, just copy an existing template and substitute your values.
https://codeberg.org/guix/guix/src/branch/master/gnu/system/...
https://codeberg.org/guix/guix/src/branch/master/gnu/bootloa...
[1] https://www.peridio.com/avocado-os
Like the article points out, the software stacks people use for embedded devices are the same as people use non-embedded use; Rust, Go, NodeJS, and sure still C++. The only real difference with embedded devices is non-OS components like the bootloader (u-boot, EDK2) and customizing the device tree. (And of course firmware flashing). Fundamentally those are all just packages that I can describe in Nix. I don't need a separate tool just because the board is small.
IMO the embedded space, especially in the US, is already pretty Niche. Most companies either just ship the vendors BSP example (Ubuntu/Debian/Yocto) and pay very little attention to the detail or re-useablity. Once you vendor declares the BSP EOL you are stuck unless you re-implement it yourself.
Using Nix (or Guix) has the massive advantage of a large and active community that isn't fractured like the Yoctoo/Buildroot community. (By fractured I mean there may by many, many people using those tools at $DAYJOB but due to vendor BSP customization they likely share much less with the upstream community maintained sources).
Because embedded usually means dealing with complete functionality provided by the soc, bare minimum for the init system to get quick boot times.
If you just want to boot the CPU and run some IoT app, device tree + kernel + uboot should be enough.
But for anything with a GUI, You'd probably need the vendor provided binary blobs for GPU accelerated UI, proper gstreamer packages with all the patches to make it work with the binary blobs... properly packaged Qt etc ... Not to mention read only rootfs and A/B partitions for upgrades to deal with power loss headaches.
I haven't used Nix for any of these but most of these things are available ready to use with the vendor provided buildroot or yocto setups. And it would've been just more work to get this all working with yet another build system/package manager
However, reading through the docs, the author clearly knows about Nix/Nixpks and has factored some of its properties into the design: https://docs.yoebuild.org/nix.html
Note, this is generated by Claude as I've not used Bazel and may not be 100% accurate. PR's welcome.
RAM is expensive now. Reasonably priced seems to be counter to that fact.
The whole article feels like a promo-tour.
> The catch is that nobody handed us the tools to keep the place running.
Software such as homebrew exists. Why is the article ignoring this? Not only does it feel like promo, it now feels like propaganda. Homebrew is also just one example; LFS/BLFS also exists for embedded linux. Perhaps the quality is less than for desktop computer systems but it exists.
> Edge devices have started behaving like cloud systems. They run increasingly complex stacks, are frequently updated, and are managed remotely over their entire life.
Companies want to become dependent on others? Because that is what cloud is all about. Top-down control. A mafia for later blackmail rack-up-the-price strategies.
> That puts the cross-compile burden squarely on embedded developers, and maintaining recipes for thousands of packages has been a steady drain on the Yocto community
Seems untrue. If you have a compiler that works, you can compile. Thousands of packages? I am tracking 3888 right now. That covers really most software by far, including python pypi addons. If a single person can do that, a team has an even easier time. The whole article seems to narrate a story that is just made up.
> Building everything from source in Yocto means long builds, heavy memory use, and powerful workstations.
No it does not. I do so every day. You need a reasonably fast computer and RAM, which is now driven up by the existing hardware mafia that must be put in jail, but it is easily possible for solo persons too. Now imagine a team.
The whole article is really really crap. Absolute garbage. The claims it makes are for the most part simply not correct. It narrates a story as if all linux developers are incompetent. I highly doubt that. It also never mentions any existing software here that would change the rhetoric such as homebrew, LFS/BLFS and so forth. It feels as if that article is just engineered to tell a "we thus need xyz". That's not good writing. I mean just look at this:
"We can leverage these change agents and rethink the world of connected products."
What does that even mean? Did AI write this? Because humans are usually smarter when it comes to opinion pieces. "connected products" - everything is now connected?
I'm glad Yocto is working for you.
I'm by no means a Yocto expert like many of you, but I've used OpenEmbedded/Yocto since the start, so I'm quite familiar with it. I started the following project, which Khem mostly maintains these days:
https://yoedistro.org/
And "I've been experiment with" seems to just mean "I've been vibe coding"
Yes, much of Yoe build is implemented using Claude Code, driven by specs.
yoebuild does seem interesting at least as a proof-of-concept.
Yes, skipping builds from source can be faster, and, I don't know that you need to throw away yocto to do that https://rootcommit.com/pub/conferences/2024/elce/yocto-binar...
From the statement about "Caches builds so no piece of software is built twice." I'm guessing that the author has not enabled sstate caching https://docs.yoctoproject.org/dev/overview-manual/concepts.h...
I'm a little doubtful on the post's assertion that yocto is only needed for "deeply embedded regulated products." For products that have to follow the https://en.wikipedia.org/wiki/Cyber_Resilience_Act coming up, they will almost certainly need:
* SBOM management https://docs.yoctoproject.org/next/dev-manual/sbom.html
* CVE (or its successor) tracking https://docs.yoctoproject.org/dev/security-manual/vulnerabil...
* License management https://docs.yoctoproject.org/next/dev-manual/licenses.html
And I'll echo what some other folks have said here: if you don't need features like that, bundle your custom bootloaders and kernel with a binary distribution like Debian and call it a day.
When someone starts with "why cross-compiling the system if it's bearable to compile it directly on the device?", it really sounds like they are just not comfortable with cross-compilation. To me it would be similar to say "why would you need a compiled language, with this weird compilation step that nobody understands, when you can just write in an interpreted language?". Except that developers wouldn't generally proudly admit that they don't know how to use a compiler.
A valid reason I see to require a new embedded linux build system would be if the existing ones (mostly buildroot and Yocto) were unable to build a modern OS (e.g. by not being able to integrate some security features). But I don't think that's the case. This article sounds more like very verbose marketing for a system claiming that it can do similarly without having to learn as much (which I honestly doubt).
To me it's often the same pattern:
* Meson is nice, but it doesn't remove the need to use CMake, and it's not fundamentally simpler or more powerful (it's more comfortable at most).
* Jujutsu is nice, but it doesn't remove the need to learn git, and it's not fundamentally simpler or more powerful.
* Yoebuild sounds like even if it got traction like Meson and Jujutsu, it wouldn't remove the need to interact with Yocto for embedded linux engineers.
As an embedded linux engineer, if anyway I have to interact with Yocto and therefore learn Yocto, why would I learn another one that isn't fundamentally better? Usually people who go for those "simpler" (at least on the surface) tools do it because they don't want to learn Yocto. And for an embedded linux engineer, isn't it a red flag to not want to learn the most famous embedded linux build system?
It reminds me of Balena/Resin, which seems like it's targetting people who are not embedded linux engineers, don't want to become embedded linux engineers, but still want to distribute an embedded linux. To me that's generally a red flag.
Architecture emulation with qemu-userspace is 5-10x slower than running a native build.
The only 'proper' no-friction cross-compilation system I know is Debian/Ubuntu, because it allows to install foreign architecture libraries into your native system, and has all the wrappers for cross-compilation in dpkg-buildpackage.
This way you don't have to maintain cross-compiler or toolchain/sysroot, you just install whatever dependencies you need with the regular `apt install` of another architecture.
I'd like to have more friendly tooling around that instead of a slow compilation of architecture emulation.