I've found it very hard to find the right place to jump into this discussion. So, I'll just put out some of my own thoughts about what I want to see out of Fedora, and then point out how I think this matches or contrasts with Matts proposal. Fedora should be an *OS*. Here are some of the qualities that I associated with that term: - It has a clearly defined boundary, with stable APIs. Some things are not going to be part of the OS, even though they are part of the Fedora universe: for example, applications. Stable is important; if you can't upgrade from version x to version x+1 and keep installed applications working, it is not an OS, imo. And it will never be attractive to anybody outside the Fedora packager community to do something with Fedora or build something on it, if there is no assurance that it continue working beyond the 6 month (or 13 month) horizon of the Fedora schedule. - The purpose of the OS is to run applications. So, it needs to be provide a way to install, update and run applications. We obviously support this now, in a way. But we need to get a lot better (see the AppInstaller proposal). The big is that many apps are simply not available on Fedora, because packaging is not something that is interesting for many people, and mostly a wasted effort from the perspective of the app developer (see the previous point). - There need to be defined extension points for how you add new stuff to it that does not fit int the 'application' category. Things like codecs, translations, fonts, runtime environments. - It should provide a defined (or designed) user experience. It can of course provide more than one, depending on the context it is used in: client, server, cloud, etc. Also worth mentioning here is the runtime vs devel split. Ideally, there will also be a defined experience for developers, an SDK if you will. Thats enough blue sky vision for now. How does this match up with Matt's proposal ? The 'Base OS / ring 0 + 1' in the proposal could possibly match my idea of an OS as having clearly defined boundaries (core + standard is pretty clear as to whats in and whats out), but as far as API is concerned, it seems a little weak - I would expect most of the system services that we are relying on to be part of the core that needs to have a stable API: policykit, pam, logind, udisks, sssd, realmd, etc. Many of these probably get pulled in via dependencies. It would be better to list them explicitly, imo. There is a tension between defining a complete enough API, and going for a minimal platform that can accomodate the needs of e.g. cloud images. Weakening the rules for bundling in applications and allowing non-rpm content may help a little bit for making applications available. But it does not really answer the question: how do I get my app to Fedora users without becoming a Fedora packager in the process. Differentiating the wider set of packages that are not part of a defined experience as 'Fedora Commons' is useful. But we more or less already have that with spins nowadays - the spins and the packages in them are what receives most of the qa and developer attention. Where the proposal fits my vision less well is where it talks about ring 2 as containing 'environments and stacks' - I don't think that gives us any handle on defining the user experience, and will arguably be worse than what we have now, where we at least know that we qa a small number of spins/images each release. I'm not sure how well the 'concentric rings' model fits with a runtime/sdk split. Matthias PS: Somewhere in this discussion, it was brought up that Fedora infrastructure is not running on Fedora. I find that a really depressing state of affairs. Changing that would be a great goal, imo. -- devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxxx https://admin.fedoraproject.org/mailman/listinfo/devel
