>> … I feel very strongly
>> that we should pick one architecture and one platform and build for
>> that.
>
> I would agree with platform.  We should aim to provide a complete
> solution, and this means bundling some kind of kernel and userland.
> I disagree that we should only target one architecture, however,
> because it would be very difficult to choose between x86 and macppc.
> x86 has the larger install base, but Mac has the largest install base
> of people who are obsessive about their UI - which might be a larger
> potential target for Étoilé.

If we are able to work with a team already producing their OS
(kernal/userland) for multiple architectures, and the maturity of each
arch we want to target is fairly equivalent, then I would have no problem
with doing what you suggest, as long as the impact on our workload is
minimal. I do agree that x86 and PPC are the top two platforms to work
towards, with amd64 as the next in line.

Etoile on x86 seems like a no-brainer. Most free software OSs are
developed with x86 in mind, and if we're going to run Etoile on a free OS
(Linux, BSD or other), then x86 support is virtually a given. And, like
you said, it has the largest install base.

I question who the userbase of Etoile on PPC would be though. My first
thoughts are Free Software advocates, users with older or minimal hardware
(original iMacs and G3s), Cocoa developers and perhaps GUI aficionados. I
assume there must be a market for Linux on PPC with the number of distros
supporting that setup... but I always thought that was so that sites with
different architectures (schools, etc) could present a consistent UI to
users. Most of our development pool will be OS X users, though, so maybe
Etoile on PPC is more important than I originally thought.

One additional thing to think about if we want to support multiple
architectures is software distribution. Do we want to allow users on
different arches to be able to swap binaries? Do we want to make it easy
for users to get and install software, without having to think about what
arch they're on? If so, we need a mechanism for handling fat binaries
then. Perhaps we can get around some of this with a really good sofware
distribution/package management service, but our audience is home users
(as far as I can tell), and we all seem extremely concerned about ease of
use, so let's keep this in mind while we're determining the bundle
structure of services and components.


J.




Reply via email to