Mastodawn

New blog post introducing the WIP Duranium project (immutable postmarketOS), some of its major features, and explaining why some design decisions were made.

> Either the new image works, or the system falls back to the previous one automatically. No partially-applied state. No debugging audio when you need to make a phone call and no fussing with a broken web browser when you just want to doomscroll cat photos. It also means developers can reproduce the exact state of a user's device, making it much easier to track down and fix issues.

https://postmarketos.org/blog/2026/03/17/introducing-duranium/

#linuxmobile #postmarketos #duranium

Introducing Duranium: a more reliable postmarketOS

Aiming for a 10 year life-cycle for smartphones

postmarketOS

@postmarketOS perhaps the project logo can be a durian

Ivan Molodetskikh Mar 17

@postmarketOS how does "No debugging audio when you need to make a phone call and no fussing with a broken web browser" follow from "the system boots successfully"? The system may boot successfully with a broken audio and/or a broken web browser. I'm curious and didn't find it in the blog post

Alexia ΘΔ Mar 17

@postmarketOS @YaLTeR

the idea is that faults like this can be discovered before an update is applied as it's much easier to replicate and test the images, and a partial-upgrade state (which might break something like audio) is entirely avoided as updates are effectively atomic

Ivan Molodetskikh Mar 17

@alexia @postmarketOS broken audio makes me think primarily about audio being broken on particular hardware due to an unaccounted edge case. Doesn't that mean that you can test an image on some hardware or even a few different phones, but still miss it being broken on some particular other audio chip?

Alexia ΘΔ Mar 17

@YaLTeR @postmarketOS

perhaps but the aim of the project is moreso to prevent situations in which audio was working and then broke after an update without an easy recovery path.

Ivan Molodetskikh Mar 17

@alexia @postmarketOS but this is still possible, as I just described?

@YaLTeR @alexia @postmarketOS The key point is "without an easy recovery path". If it's anything like Fedora Silverblue, which is also an atomic OS project, if something breaks you can just undo the upgrade and it unbreaks. Undoing upgrades is much harder on a traditional OS, and often intentionally impossible on mobile operating systems

@YaLTeR @alexia that's why we're also building automated testing which will include automated testing of audio and phone calls.

eventually we hope to have a comprehensive test suite that every update must pass on every device before it gets shipped

WeirdTreeThing Mar 17

@YaLTeR @postmarketOS If an update breaks something such as audio, you have the option to roll back to the previous version without the issue.

UndeadLeech Mar 17

@YaLTeR @postmarketOS Well the audio would have to work reliably in the first place. 😅

pabloyoyoista Mar 17

@YaLTeR part of it is that we have an unexpectedly high number of audio failures consequence of partially-applied state and buggy upgrades. Why it's always audio 🤷 You can still get an update with broken audio if Pipewire or ALSA broke for your device. But then it should be reproducible in all upgrades/installs, which has really not been the case so far, speeding up fixes @postmarketOS

Jonathan Frederickson Mar 17

@postmarketOS Interesting! I have a FP5 that I'm not currently daily driving and I see there's an image for it, so might need to give it a try

I'm curious if it would be possible to run Guix on it... the docs say there's nothing stopping you from running Nix on it but Guix like Nix requires a writable store dir. Going to have to dig into this some more later

PureTryOut Mar 18

@jfred @postmarketOS As long as that directory doesn't need to be on /usr, it'll work just fine.

@postmarketOS can i flush it via web flusher?

kersnazzle 4d ago

@postmarketOS @craftyguy I just got round to reading this properly and found it really interesting. I particularly like the link off to the design document which is an excellent piece of documentation. Not going to pretend I understand it all but that's on me 😀

@postmarketOS is there a guide to go from stock android fairphone 5 without unlocked bootloader to duranium? i tried to do it by combining the wiki pages of the fp5 and trying to install regular pm edge first but even while using what i assumed to be a safe way to do it using ubports, i bricked my phone in the process (apparently very easy to do on fairphone 5)

i‘d love to try again but it’s not really sustainable to send the phone to france for repair again if i brick it again, so i would benefit greatly from one install guide from start to finish

Jens Finkhäuser 2d ago

@postmarketOS Ah, Duranium implies solidity, but it's based on AI slop.

https://www.phoronix.com/news/systemd-260-rc3

No thanks.

"Vibanium"

systemd 260-rc3 Released With AI Agents Documentation Added

The first release candidate of systemd 260 arrived in late February with the new mstack feature, dropping System V service scripts support, and other changes

@jens crazy take considering it's been less than 2 months since systemd started using AI for code review. Critique all you like but at least don't disrespect us by doing it in blatant bad faith

we switched to systemd over 18 months ago! if we wanted to avoid it we would be setting our progress back half a decade at least.

at the end of the day there's no ethical consumption under capitalism, it's a constant battle between harm reduction and moral principles, the pros did (and imo still very much do) outweigh the cons. but by all means give us an alternative that meets your standards

A lot of people are understandably scared, disinfranchised, or afraid of AI-written code. I don't think this was as bad faith as it may have come across.

Engage. Discuss. Please don't slap people down for showing a very human reaction in very awful times.

Be a community leader.

@kestral @jens i am too, i hope to god the bubble pops. i have a lot of patience for people who have concerns or questions, im more than happy to talk through my thought process because this is something i care about deeply.

but at the end of the day ive learnt that it's often a waste of time to try and reason with people who are making sweeping dismissive statements of a project i have poured my soul into just because of its dependency on a project they dislike. This goes especially for systemd, we had and continue to have a lot of discussions with them about the state of the ecosystem and how we want to use their software, if we didn't align with them on a technical level we wouldn't have made the switch.

our goal is to build a mobile OS that actually respects the user without compromising on being accessible to non-technical folks or on security. If your priority is to not use systemd then i just have nothing else to say because you clearly aren't taking the utilitarian approach and we have a fundamental misalignment.

we are right now trying to seize the moment that the current geopolitical landscape has provided us, we aren't going to succeed in making it to production in a reasonable timeframe if a large part of our focus is reinventing the wheel (and having to get all our security code audited). There are just more important issues for us like figuring out how to enable secureboot on Android devices with pmOS, making the UX and telephony functionality actually reliable, collaborating with vendors, etc...

I have no doubt that once we actually get Linux Mobile in production with at least acceptable levels of security that the devices we support will be able to run whatever systemd-free OS you want, I welcome that era with open arms. But we have to pick our battles and there is no world where it makes sense for that to be our priority.

I think out original announcement blog post explained things pretty well, i should probably just link to that when people post comments like this

https://postmarketos.org/blog/2024/03/05/adding-systemd/

Adding systemd to postmarketOS

Aiming for a 10 year life-cycle for smartphones

postmarketOS

@cas thanks for taking the time to respond. Have removed the OP as I don't want to inadvertantly speak for anyone else.

In an imperfect and - frankly - dangerous tech landscape, getting people off of corporate-controlled ecosystems and onto any kind of open mobile OS is not just important, it's nessecary.

I think your words around getting the groundwork laid so that the community can use the things they want to is a good way to think about it.

Got to crawl before we can walk.

@cas I also really hear your exhaustion with the barage of communication that comes from building things in the open. I think that's especially hard in this age of social media, and I'd imagine that puts anyone on edge when getting replies.

So, thank you for engaging.

Jens Finkhäuser 2d ago

@cas @kestral The conclusion is: ditch the toxic dependency.

I have zero fucks left for fascist AI slop.

Don't like it? Fine, do your thing.

But do not for one second expect me to be reasonable about it.

Jens Finkhäuser 2d ago

@cas My blatant bad faith is with AI slop, and I absolutely disagree: the only way this can be reversed is by mercilessly shaming everyone who uses it.

Sorry it's you 🤷

But it's not *my* choice to support AI slop, so don't try to turn this "bad faith" argument around. Putting AI slop into code is bad faith.

Jens Finkhäuser 2d ago

@cas Here's a link to a fuller argument:

https://social.finkhaeuser.de/@jens/116242753615998997

Jens Finkhäuser (@[email protected])

Writing this up again so I can pin it: AI is literally a fascist project. Friends don't let friends use it. Before I go into this, there are two types of responses to this that I have taken seriously so far. One I'll call HashTagNotAllAI, which yields the obligatory "sure", but has the same smell. I'll leave it at that. The other is that an anti AI stance also throws some assistive technology under the bus, making such a stance intrinsically ableistic. The easy thing to do is to refer...

Finkhäuser Social

Erin 💽✨2d ago

@jens @cas tell me Jens, what OSes do you run on your systems? Are they free of LLM written code? I'm going to guess the answer to that question is no, unless you're running NetBSD or OpenBSD with few additional packages

Jens Finkhäuser 2d ago

@erincandescent @cas Indeed.

@jens *we* have a strict anti-AI policy for our contributors. Why don't you go and shame Ubuntu or Fedora who are actively pursuing shipping AI features in their distro rather than focusing on us purely for using the same init system as every other mainstream distro?

im on board with your anti-AI stance, it would have been nice if systemd had stayed away altogether, but what practical outcome do you want from us?

https://docs.postmarketos.org/policies-and-processes/development/ai-policy.html

AI Policy - Policies and Processes

@cas take it from someone who's been around this particular block a few times: you will _never_ reason someone out of a position they didn't reason themselves into in the first place. There is only one way to deal with lunatics with utteraly destructive and deranged takes on social media: the block button. Hit it and move on.

Jens Finkhäuser 2d ago

@cas I shame everyone when there is an opportunity.

Your answer reads like "sure I beat my wife, but that person there kicks his, too!"

Doesn't really make anything better, does it?

Neil Brown 2d ago

It might be worth stepping away and simmering down a bit! I'm sure that your view has come across clearly.

Jens Finkhäuser 2d ago

@neil I believe it when I see a plan for moving away 🤷‍♂️

Problem with projects like systemd and the Linux kernel is: they need *their dependencies* to move away to change.

So every response about "what do YOU do wrong", or "how about THAT project" boils down to: they won't give two shits about what I write here.

The only people who might are people who *care*. And, @cas , please believe me when I say: that's why I bother replying to *this* thread.

I recognize that, and wish all the best...

Jens Finkhäuser 2d ago

@neil @cas to you and postmarketOS.

But I still do not accept any apologies about AI slop. I also will not in future.

Sorry if it came across too strongly. Break an egg to bake a cake, that kind of thing.

@jens @neil if i truly thought dropping systemd would be a net positive i would advocate for it in a heartbeat. I think we assign different values to things (as a matter of bias, perspective, etc) and we probably aren't going to overcome that. Anything else i could say here would be bitter and jaded so i'll leave it there.

Jens Finkhäuser 2d ago

@cas Fascism is not a value.

Jens Finkhäuser 2d ago

@cas I am so fucking done with fascist apologia in FLOSS.

Do NOT DARE to talk about "values" here.

Ariadne Conill 🐰

@cas @jens so let me get this straight.

you are saying pmOS, a project largely maintained by queer kids (aka: the front line for fascist genocide) fucking around with phones and other SBCs, is fascist because they are using systemd?

but also Linux kernel in general is now using AI tools for actual code generation much less review like systemd.

I think you need to step back, rethink your positions and apologize. seriously.

@jens yeah you're right, guess we should give up trying to improve society somewhat if somewhere along the line someone engages with systems we don't like. Hey let's shame and scapegoat anyone who questions our black and white thinking too....

Now that i think about it im sure there's a term for that way of thinking 🙃

AaronTheIssueGuy🏳️‍🌈2d ago

@jens @cas Ngl, calling out pmos for facism cause they use systemd is kinda derailed.

Hate systemd all you want for allowing Ai code review, but shifting responsibility to every project that’s using it is nuts considering its adoption that has been long standing before the days of Ai.

Not sure you are doing anyone a favour with wild and unbiased generalizations. The way you behave in this thread is also wildly disrespectful towards people building Foss.

@cas isn't it funny how these HURR DURR SYSTEMD BAD MUST DROP trolls never shout at you to drop the Linux kernel, despite the fact that it has the exact same identical LLMs policy and uses them for the exact same purpose (code reviews)? Curious, isn't it

https://www.theregister.com/2026/03/26/greg_kroahhartman_ai_kernel/

AI bug reports went from junk to legit overnight, says Linux kernel czar

Interview: Greg Kroah-Hartman can't explain the inflection point, but it's not slowing down or going away

The Register