On Fri, Jun 05, 2009 at 02:38:24PM -0700, Brock Pytlik wrote: > Nicolas Williams wrote: > >Not necessarily. A registry, for example, would allow us to solve that > >problem. > > Could you expand on the idea of a registry a bit? My impression is that > to solve this problem, the proposal is to have a central repository at > which everyone who makes a package for distribution on OpenSolaris > registers the file locations and properties, symlink and hardlink > locations and targets, and directory permissions. To be truly safe, > would things like SMF service names and properties be needed as well? Is
Yes. > the proposal that to install any package, IPS would first check this > registry to ensure the package was registered properly? Is there an No, absolutely not, pkg(1) would never check the registry (though publishers might, if they want to). The registry adds value where pkg authors adhere to it, is advisory rather than mandatory (except for Solaris consolidations, which would have to adhere to it). See below. > example of another community where this has been done, and done well? My > impression is that this seems like a solution with a lot of overhead > which depends on buy in from the community to voluntarily register the > packages they publish. For packaging? I don't know. But registries certainly have been successful in many areas (see, for example, IANA's many registries). > As a side note, whoever is in charge of the registry would also probably > need to take on adjudicating disagreements between package publishers > about who has the right to specific files/links/etc, a job which seems > difficult to say the least. I think the right approach is to use a "first come, first served" allocation policy with "expert review" by experts in the community being required (to avoid frivolous registrations). For some sub-namespaces we might require ARC review. Advisory registrations to document the existence of names in other Linux, BSD and OpenSolaris distros, without that actually reserving the name for the same purpose in OpenSolaris, might be a good idea too. If a conflict arises where a registrant loses the race to be first but nonetheless believes their registration should win... then we'll need an appeals process (which could be based on community consensus, and/or ARC review). But such a conflict resolution process should be needed only when one of the conflicting registrations is delivered, or to be delivered by, a Solaris consolidation. Conflicts not involving any Solaris consolidations should be allowed (subject to expert or ARC review as mentioned above) and noted in the registry, including the chronological order of the registrations. I.e., the registry should be mandatory for Solaris consolidations, advisory for all others. I think an advisory registry would still add much value for third parties, since third parties should prefer to avoid conflicts. Nico -- _______________________________________________ indiana-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/indiana-discuss
