No, I don't have an unreasonable amount of trust in vendor-released firmware. I know that vendor firmware has mistakes and can be poorly designed just to get something out and usable to a customer base. I also know that there are many smart people in this mailing list who do not like the ESP HAL, and I'm positive that their criticisms are valid and come from a place of good understanding.
However, the ESP HAL has a significant user base, is maintained by the company responsible for the product itself, is tested in Zephyr usage alongside NuttX usage and comes with Espressif employees who maintain the software and are generally quite good at responding to community issues, even if they don't solve them + merge patches immediately. The Espressif contributors we work with do a lot for the community and they submit high quality patches. The HAL also comes with a nice license that has guarantees. I weigh this much higher than an AI-rewrite spearheaded by a few contributors, which would need significant testing and has none of the other aforementioned benefits. At least with an "old-fashioned" re-write (by hand), I can have some confidence that there are developers out there who understand the code being written and have enough motivation about it to spend actual time writing a new HAL instead of just trying to quickly spin something up with AI and then leave it unmaintained. I am not accusing you or anyone here in favour of the AI rewrite of not taking time to understand the code you are generating, but generating and reviewing code is fundamentally a different understanding than writing it yourself and having to go through the slog of effort that shows real commitment to maintenance. That is what I am worried about (in addition to the licensing issues and massive testing undertaking). If I am worried about this, think about the external perception of people who start to wonder if NuttX is a viable, well-supported RTOS for their products or if it is just a playground for AI-written and maintained code. This is a divisive issue and there are a *lot* of people who will question the reliability of NuttX more than is already the case if we start re-writing large chunks with AI. And AI or not, like others have mentioned, this is still a massive change of its own and would require weighing whether or not we can reasonably replace the existing HAL in testable chunks one at a time and then continue to maintain our own copy. There has to be more benefits of this change than the status quo for it to be useful, and we already struggle to maintain a lot of what exists in NuttX currently. Matteo On Wed, Aug 5, 2026 at 1:51 PM Felipe Moura Oliveira <[email protected]> wrote: > Hi Matteo, > > Do you really think that vendor-released firmware is guaranteed to be > functional and well-tested? > Consider Windows versus Linux. Windows is developed by a private company > with its own protocols, testing processes, and so on, yet I do not find it > to be more stable than Linux. > Look for esp p4, when it was released it was like a beta version. > > > Em qua., 5 de ago. de 2026 às 14:04, Matteo Golin <[email protected]> > escreveu: > > > > I disagree with the suggestion that those interested in exploring this > > path should just create a separate experimental project. In an > open-source > > community, every contribution and idea is valuable, even if it introduces > > disagreement or ultimately fails. > > > > Yes, in spirit. This is why the idea is being discussed here. But > > ultimately, without consensus, we can't just have everyone's experimental > > ideas merged into NuttX to succeed/fail there. That is why we are > > discussing it first. NuttX mainline is not a playground for anything, > > that's why we take pains to ask for testing logs and get proper PR > > descriptions. > > > > The ESP HAL is used in many devices. One such device is my ESP32 based > > rocket altimeter. This altimer is responsible for deploying parachutes on > > my very expensive rocket. If that goes wrong, my rocket crashes and I > lose > > a lot of money and time. I don't want to rely on an experimental, AI > > written HAL for this. I much prefer the iterated, strictly tested and > > maintained HAL by the company itself. A NuttX written HAL would need to > be > > well tested, understood and maintained for me to want to make the switch. > > These changes have real consequences, and for companies much larger than > > just lowly old me and my hobby rockets. Keeping guarantees for users is > the > > utmost concern and we always need to weigh that when considering options. > > We are unfortunately not free to just experiment with anything in > mainline > > when we'd like. > > > > Matteo > > > > On Wed, Aug 5, 2026, 12:52 p.m. Matteo Golin <[email protected]> > > wrote: > > > > > What Raiden mentioned is right: AI generated or not, this is something > we > > > have to discuss with the Espressif maintainers. They are incredibly > > active > > > and keep our ESP support up to date. As much as I want our own NuttX > HAL, > > > we have to decide if it's worth trading for the burden of having our > own, > > > non-Espressif maintainers for it. I'm incredibly against the AI > rewrite, > > > but at this time I would also be against a by-hand rewrite unless we > > > somehow get continued support maintaining it, either by Espressif or > > other > > > devs, at the level we get from Espressif right now. That kind of > > maintainer > > > work is super valuable to NuttX, which already has few maintainers. > > > > > > Also, respectfully, thinking that a new HAL rewritten to follow NuttX > > > idioms will have bit-by-bit identical output to the original ESP HAL is > > > wishful thinking. All kinds of things, including internal data > > structures, > > > will be different. Unless I am misunderstanding that proposition, that > is > > > definitely not going to work as a testing plan for a new HAL. > > > > > > Matteo > > > > > > On Wed, Aug 5, 2026, 12:44 p.m. Tomek CEDRO <[email protected]> wrote: > > > > > >> On Wed, Aug 5, 2026 at 4:27 PM Marco C. <[email protected]> > > wrote: > > >> > Let me try to explain the methodology so you can open your mind: > > >> > (..) > > >> > Open your mind. Peace. > > >> > Marco Casaroli > > >> > > >> We are open-minded Marco, very, it is highly unfair to say we are not. > > >> > > >> You need to take into a consideration that all NuttX related work is > > >> done in our free time at the expense of any other tasks sometimes our > > >> families and so called private human life. > > >> > > >> We are also people taking personal responsibility for the products > > >> development, its delivery, and maintenance. If things take too complex > > >> and too long to deliver we are out of business. If things fail on > > >> production we are out of business. If we are unable to maintain and > > >> service products we are out of business. This out of business also > > >> means measurable financial responsibility. > > >> > > >> Most of us are owners of the company, business, product. We have > > >> experiences that made us have solid stand on the ground and think the > > >> way we think. Its okay to play around and experiment, but its not okay > > >> to impose experiments on everyone around, like so called modern world > > >> does right now. We cannot build without solid foundations, when all > > >> components change all the time, including software and recently even > > >> hardware components. Its a big complex mess right now. We do not want > > >> to make it worse or even take part ourselves. This is clearly > > >> destructive. > > >> > > >> We can play around but then we need to provide alternative path to be > > >> selected, while foundations must remain rock solid, so people have > > >> choice. I understand you use free AI to create stuff, but in the end > > >> there are living humans paying bills daily that need to process that > > >> also for free in their free time. > > >> > > >> My proposition for the people with lots of free time is to create an > > >> experimental but separate AI NuttX rewrite and see how it goes. It > > >> should be a win-win scenario where we can cope with basic tasks in > > >> NuttX while the other project may explore new worlds. > > >> > > >> Peace :-) > > >> > > >> -- > > >> CeDeROM, SQ7MHZ, http://www.tomek.cedro.info > > >> > > > > > > > > -- > > *--Felipe Moura de Oliveira* > Linkedin <https://www.linkedin.com/in/felipe-oliveira-75a651a0> >
