Rolling release model philosophy (was Re: Anaconda is totally trashing the F18 schedule (was Re: f18: how to install into a LVM partitions (or RAID)))

[Przemek Klosowski]

On 11/02/2012 07:04 PM, Adam Williamson wrote:

> Sure, like I said in another mail, we've got better at that than before.
> But as I also said in the same mail, you still have to do a version
> upgrade every twelve months. That alone is ridiculous for a 'stable'
> operating system.
>

This is an important point---it makes it difficult to deploy Fedora for 
other people. When the end-of-support comes, it usually means having to 
reinstall, because upgrade can take unbounded time, if problems pop up. 
Additionally, in my experience, a reinstall often results in a better 
configuration, free of grandfathered suboptimal settings.

I keep thinking about a scheme to roll over an EOL Fedora into a closest 
possible CENTOS. It's not trivial because I can't just look for the 
CENTOS that matches the original Fedora release, because of the 
subsequent updates. It would have to look at the as-is system and try to 
figure out the best matching CENTOS release. I am thinking about a 
sum-of-squared-differences-like distance metric: calculate sum over all 
packages of (installed_version - CENTOS_X_version)^2, for several 
CENTOS_X versions, and chose the one giving the smallest value. Of 
course some packages (glibc, kernel) would have a higher weight, but 
that could be incorporated (\sum_i((v1_i-v2_i)^2/wght_i)).