Re: problems with the concept of unstable -> testing

["Daniel Moerner" <dmoerner@gmail.com>]

On Tue, Dec 16, 2008 at 4:34 PM, Josselin Mouette <joss@debian.org> wrote:
> Le mardi 16 décembre 2008 à 23:22 +0100, Julien BLACHE a écrit :
>> Also new users have a tendency to go with testing and don't use
>> unstable much these days.
>>
>> The net effect is that there aren't enough people left using unstable
>> to uncover enough problems. Hence bugs silently make it to testing.
>
> Maybe that's because I maintain packages with a large audience, but I
> don't find that effect very important.
>
> More annoying are these effects:
>      * Bugs that trigger with a specific combination of packages. In
>        unstable they are fixed very quickly, but even when adding a
>        Conflict, one of the packages can migrate to testing long before
>        the others and keep testing in a broken state.
>      * Testing users don't check whether a bug is fixed in unstable.
>        It's not that bugs silently make it to testing, but they are
>        fixed much more quickly in unstable. That could be improved with
>        reportbug being stricter about bugs against testing packages
>        with newer versions in unstable.
>

Obviously, having more users test unstable is good.  However, I agree
that it's not necessarily a big issue.  A good deal of RC-bugs are
related to FTBFS, security advisories, package conflicts, and the
like.  These bugs can pop up independently of how much testing a
package receives in unstable, so focusing on just increasing the
number of unstable users would produce diminishing returns.

>> Being stricter wrt testing migration is hardly going to help. What
>> will help is having more people actually use unstable so bugs are
>> uncovered before they hit testing.
>
> Actually I don't think we should recommend testing at all to desktop
> users. Except during freeze times, I find unstable to be much more
> usable, and keep testing for (non-production) servers.

I think this is true as well, especially in the context of other
non-Ubuntu distributions that track Debian.  Sidux, which tracks Sid,
is able to keep almost complete compatibility with the main Debian
repositories with the exception of kernels.  In contrast,
distributions that track testing often have to have much more complete
supplementary repositories.  In part, this is due to ideological
differences, but I think it's also due to the fact that desktop users
can get more mileage from unstable.

Cheers,
Daniel

-- 
Daniel Moerner <dmoerner@gmail.com>