Re: A proposed Roadmap

[olafBuddenhagen]

Hi,

On Wed, Sep 05, 2007 at 01:07:26AM -0500, R. Steven Rainwater wrote:

> Problems 2 through 7 are solved in my proposed road map by releasing
> an initial version of the GNU OS that uses a 100% Linux kernel. Over
> time, we would transition to a 100% GNU Hurd kernel. This allows us to
> immediately resume work on the GNU OS and we can release a working
> version of the entire GNU OS very soon, perhaps within a year. My idea
> for the kernel transition is to go through several phases that would
> allow work to focus on specific tasks, each of which would move us
> closer to a 100% GNU Hurd kernel, while maintaining a completely
> usable GNU OS at each point in the transition.

As I already said on IRC, this suggestion is totally out of the question
IMHO.

There is just no point in releasing the GNU system based on Linux, to
compete against the hundreds of existing GNU/Linux distributions. And
once it would be release with Linux, it would be virtually impossible to
switch -- nobody would dare to go from a limited but perfectly working
kernel to something rough and incomplete. There would be absolutely no
chance of moving over to Hurd unless it's almost perfect -- but that
won't ever happen, as with the GNU system already released with Linux,
there would be even less inclination for people to work on the Hurd.

The GNU system would be totally irrelevant as just yet another GNU/Linux
distribution, and the Hurd would be marginalised even more -- a perfect
loss-loss situation.

Really, there is absolutely no point in releasing the GNU system with
Linux as the kernel. It would bring no good at all.

> This leads us to Phase 2, where we do something similar to the L4Linux
> project; we create a single server Linux running on top of the
> selected GNU microkernel. Once stable enough, this goes into the GNU
> OS distro where it can be used heavily by real users. This sort of
> real world use should help improve the microkernel and identify any
> bugs.

That doesn't work. Mach also was tested for a long time with
single-server systems. The real problems showed only when people
actually tried building proper (multiserver) microkernel systems on top
of it.

-antrik-