Re: [gentoo-dev] Is there a process for marking ebuilds stable?

[Brad Laue <brad@brad-x.com>]

On Mon, 2003-04-14 at 12:29, Mikael Andersson wrote:
> I don't think stable.gentoo.org is not a good solution since it's too much
> manual work included from a user for (apparently nothing) in return.

Agreed.

> I think the most efficient way to mark packages stable is statistics based.
> 
> If you compare number of installations of a  package against number of bugs
> filed and their severity i think you should get pretty decent stability 
> figures for most packages. The exception to this is packages with few users
> but the users of such packages is probably more interested in 'voting' for 
> their packages.

A tinderbox would be good to work around problems with 'unpopular'
packages. Over the course of this thread I've seen several problems
which an automated build and report system would solve.

> This is only an initial suggestion, please comment and improve :)
> 
> 1) Successful Emerges/Bugs
>   a) Count package downloads and bugs filed. If no blocker/critical bugs 
> exists after a week or two mark as stable. For important packages this rule
> could be made more stringent.

Good idea.

>   b) Count real merges/unmerges of packages and not only package downloads, 
> this should to opt-in since it would in some way need to post information 
> back to gentoo.org

The package system should probably be self-sufficient; the userbase is
too ephemeral to rely on for something like this; stable.gentoo.org is
evidence of that; a small fraction of the userbase is a) aware of it, b)
interested in using it, and c) interested in using it often enough.

Probably the only input a package should receive from a person is from
the maintainer itself.

Which leads me into another problem; currently there are no official
maintainers for a large number of the ebuilds in the tree. This prevents
the above from being doable; no one is around to represent and vouch for
the functionality of those ebuilds, just the bug reporting system.

So, is a tinderbox doable? One would have to be volunteered for each
supported architecture, or at least x86, sparc and powerpc to begin
with. One foreseeable complication would be the nearly infinite number
of combinations of USE flags, but I'm sure with discussion a way around
this could be found.

Thoughts?

Brad


--
gentoo-dev@gentoo.org mailing list