Kernel vs drivers releases?

Kernel vs drivers releases?

[Fortier,Vincent [Montreal]]

I woke up this morning with an idea and I tought I could ask what people
think about it...  I might be completely wrong but anyway..

I've read a lot of stuff on linux 2.6 development model which makes
either people happy or unhappy of it's really fast development speed (I
love lwn.net kernel readings... keep up the good work!).

To make the stable branch more "stable" there is even now a 2.6.x.y set
of patches which seems to be making a really nice jobs on fixing
critical (or really important) stuff (again, keep up the good work!)

One thing that I've heard most was that, using this fast development
model a lot of newer hardware gets supported quickly (also a lot of new
features...)

On the other hand many people want's to get a full stabilisation of the
actual API... Maybie even get back to a fully independant stable vs dev
tree like it used to (2.6 vs 2.7) ?  This might have been partially
addressed by the new 2.6.x.y scheme although I don't get a feeling that
everybody is totally happy with it.

This morning I tought of another approach... Maybie somebody already
suggested that earlier?

Why not completely separate the kernel part from the driver part?

We could get a bit slower development cycle on the kernel side (maybie
2-4 kernel / years) and get a really fast development cycle for the
driver set (1 / month)?

This would:
- Make people feel that the kernel API is more "stable"
- Keep the 2.6.x.y scheme running the same way to make the kernel API
even more stable
- Make new hardware supported even more rapidly

End-users could eventually simply upgrade their driver set without
updating the whole kernel ... Simplifying a lot (at least I think) the
"update" process for both the end-user and the Linux distributor.

A new 3.0 kernel could be started using this scheme with it's associated
driver set?  Then a dev 3.1 for 3 to 6 months to finally get to a stable
3.2 kernel with it's associated driver set?

Am I completely wrong?

- vin

Re: Kernel vs drivers releases?

[Arjan van de Ven]

> Am I completely wrong?

I think you are yes ;)
You seem to assume you can separate drivers from the kernel. In linux,
drivers are a full member of the kernel, which really can't be
separated. Or in more technical terms: there is no stable API between
drivers and the kernel that separates them. Drivers are an integral
part, not something that can be chopped off and kept separate. Not all
operating systems do it that way. but linux does, and the reasons for
that have been discussed to death on this list many times (there is even
a file in Documentation/ about it). But given that situation.. the
separation you propose makes no sense.

Re: Kernel vs drivers releases?

[Daniel Drake]

Fortier,Vincent [Montreal] wrote:
> On the other hand many people want's to get a full stabilisation of the
> actual API... 

Not meaning to sound harsh, I think you belong in this category:

	You think you want a stable kernel interface, but you really do
	not, and you don't even know it.

See Documentation/stable_api_nonsense.txt

Daniel