Thread: thoughts on post N900/Maemo era
View Single Post
mr_jrt is offline mr_jrt
2012-01-23 , 23:17
Posts: 249 | Thanked: 277 times | Joined on May 2010 @ Brighton, UK
#6
Originally Posted by sulu View Post
Well, I have to disappoint you. I totally agree with what you say here.
btw: I intend to use my N900 "forever". I had my last mobile phone for more than 5 years and that old desktop computer I was talking about will become 14 years old in summer. I expect my N900 to become at least 5 years old too and I have two replacement N900's in my closet. And by the time all of them are gone my eyes will be to bad for those little screens anyway.

The difference between Hurd/BSD and N900 is that the former actually are something special. They need their own userlands because you can't run an existing flavour of Linux binaries on them. That's totally different for the N900. It has a Cortex A8 CPU so it runs normal armel respectively armhf binaries if you use a Linux kernel.
If you want to use a BSD kernel that would be a different story because you'd need some architecture like kfreebsdarmhf but that stil doesn't make it special for the N900. It would just be a regular armel port for FreeBSD kernels which runs on the N900 too.
I may have been ambiguous here, I'm just saying that having a separate kernel for each device in the Debian style for our mobile distro is no big deal. I not saying that you can't run an existing userland on top of the N900 kernel (addressing your point about the Hurd et al.), but you don't have to be running the head Linux kernel. The backports and amazing work done so far on the head kernel buys us a lot of time, but just as the users of Debian stable run old kernels quite happily, so can we, to a (far off) point. For me being able to use a mainline kernel is just something to aspire to once we can. The desktop 3D drivers have had a LOT of attention over many years and they're still not as complete as the closed source ones. We could be waiting forever for those drivers. In true project management style, you really want to reduce the critical path Eventually, we will probably hit the point where a shim around things like the 3D drivers becomes necessary as too much software relies on impractical-to-backport kernel changes. Hopefully it shouldn't be too problematic as our side of the interface is a known, so adapting things should have a minimal performance penalty. It's hard to say though. If ndisloader can run Windows drivers on Linux I suspect we can run old Linux drivers on new Linux kernels given enough motivation.

Idealism gets the hard, boring things done (the last 20%), and for that we should all be eternally grateful. Practicality gets the majority done (the first 80%).