On 10/09/2007, Stephen Hahn <sch at sun.com> wrote: > Although I sympathize with the design from interaction perspective, it > seems to me that the most important aspect of the DC is in fact the > file mentioned in point 4. This lack of this file (or "image > descriptor") has already come up in our exploration for pkg(1) in > > http://blogs.sun.com/sch/entry/pkg_1_a_no_scripting > > It seems that the image descriptor should be able to describe zones > and other systems with degrees of shared components, as well as > standalone systems, install/miniroot images, and other common > deployments. > > Of the points mentioned so far: > > - I agree that the raw installed size of the image is a > constraint that belongs in the descriptor. The on-media size makes > sense, too, although I think it has to be evaluated via an actual > image build. > > - I agree that summarizing dependencies is not sufficient to determine > intent. Tracking explicit requests, either in the descriptor or > from operational history, can resolve some cases, but an image-level > policy is probably needed. Packaging has to track this information > in any case. > > - I agree with the requests for groups of packages and specific file > endpoints as goals for the constructor; I would add service FMRIs as > well. > > - Branding should be isolated to packages first, which may require > work in some consolidations (but not in others); how the > higher-level components make branding easy can be pursued in > parallel. > > - If you do intend to allow "based on installed *and configured*", > then I assume that this feature replaces the flash archive feature? > One aspect of the image descriptor I've worried about is whether > there's a marshalled form of an install, or whether the descriptor > is sufficient to recreate it on its own. (Sounds like not the > latter.)
The one fatal flaw I see is that every installation packaging or other system I have ever seen in a major product allows for scripting the installation. I am not yet convinced that you can do completely without scripting. I support the idea of a sandboxed environment that restricts what the scripts can do, but I am not convinced that you can do without one completely. However, I would like to be proved wrong, so good luck! :) -- Shawn Walker, Software and Systems Analyst binarycrusader at gmail.com - http://binarycrusader.blogspot.com/ "Beware of bugs in the above code; I have only proved it correct, not tried it. " --Donald Knuth
