These are really interesting ideas, even if I don't understand all
physics equivalences :)
About the amount of man-hours needed for any different model, I think
the current model does not require little amount of man power at all,
I'd rather say keeping all things work nicely together require actually
quite substantial amount of knowledge and time fixing compatibility
issues. Decoupling could help to see and fix the issues.
However, as we start decoupling things, I see there might arise new
challenges regarding compatibility of those parts -- which is actually
where I see the discussions about rings model should give us some
answers and we should find a way how to ensure the compatibility with
reasonable amount of man-power. Anyway, I agree that in case we find it
requires too much resources, it should be a reason to change the concept.
Honza
On 01/16/2015 06:42 PM, Pete Travis wrote:
On Jan 14, 2015 6:46 AM, "Honza Horak" <hhorak(a)redhat.com
<mailto:hhorak@redhat.com>> wrote:
I'm reading the 'rings' model as somewhat analogous to the Bohr model of
the atom; concentric rings, with each element on a given ring following
the known circular path.
In practice, the theory more closely follows the modern orbital model of
the atom, where more volatile elements end up in the more outer layers
but have their own vector within that layer. It's a model that I think
more closely parallels the goal here, which I'm inferring is to allow
the less critical packages to follow a more divergent path.
Call it inexperience, but I have trouble conceiving how extended
implementation of this goal will scale. Moving from a general
"Everything should work nicely together, even if it requires compromise"
to "Some things can diverge instead of compromise, within reason" sounds
like a path where the various permutations of package and library
interactions can become overwhelming. As with the orbital model, I can
envision that quantifying the state of an "electron" - package, package
set, language, whatever - will be an exceedingly difficult task where
not all factors can be known.
Unlike the orbital model, I think we want elements to move inward as
more energy is invested, not outward. What I initially interpreted as a
model for providing developers with a graduation path now seems like a
fragmentation path, where, using the example cited above, a developer
can work with httpd-dev on Fedora Workstation and experience unexpected
problems when deploying to httpd-prod on Fedora Server and still more
when deploying to Fedora Cloud or some unflavored instance. Avoiding
those problems when fragmenting at scale is going to require a massive
amount of analysis, and I have reservations about whether the community
can provide the man-hours and motivation required.
--Pete