On 28 April 2015 at 22:57, Sean Dague <[email protected]> wrote: > On 04/27/2015 05:53 PM, Robert Collins wrote:
> I'm still generally suspicious of the precompute / install model because > solving that ends up being... interesting some times. I also think there > is a related issue of "dependencies for optional features" which, > because they are inconsistently dealt with, exacerbate things. This > being things like drivers, db backends. > > After the giant unwind Doug, Clark, and I started writing up the > following - https://etherpad.openstack.org/p/requirements-future > > I do think we need a summit discussion, I also think pip needs some > fixes, but I think this needs a lot of heads together to get to a plan, > because many individuals thought they nailed this issue in the past, and > were wrong. Thanks for that link, I think its not questioning enough about our current setup. I had a bunch more analysis I pulled out of my blog post for legibility. FWIW I've worked in the single-global-list model before for CI, and while its definitely a tradeoff in maintenance vs surprises, its well worth it for nontrivial environments (which we are!). Dependencies for optional features fit trivially into my proposal: again I elided them to save space and hopefully aid comprehension of the main proposal. The details are: they go in two places, the global constraints and the project extras[sometag], but other than that they are managed identically. https://etherpad.openstack.org/p/liberty-requirements-management looks like an important etherpad to work through -Rob -- Robert Collins <[email protected]> Distinguished Technologist HP Converged Cloud __________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: [email protected]?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
