1 |
Sunday, 2. March 2008, Steve Dibb Ви написали: |
2 |
> Christian Faulhammer wrote: |
3 |
> > What we propose is proper testing and keywording by anyone |
4 |
> > around...not just team members. |
5 |
> |
6 |
> I agree... our main problem is manpower -- people actually working on |
7 |
> the stable bugs. I've tried to do it myself a few times, but each time |
8 |
> it just burns me out to the point where I don't want to (and won't) work |
9 |
> on anything Gentoo-related for a time. |
10 |
So, may be we should expand the number of "stability classes"? (something akin |
11 |
to my origina proposal, parst of which has been implemented, but looks like |
12 |
there is yet another usefull issue or two that did not make it yet :) (bug |
13 |
#1523, though probably not worth studying a this point as it mostly contains |
14 |
old stuff by now)). |
15 |
|
16 |
Right now with packages being only "in testing" and "stable" we basically |
17 |
cover the audiences with stances "original dev says it works - it's fine for |
18 |
me" and "I want it rigorously tested". There should be plenty of people, who |
19 |
would be happy with an intermediate level of control. May be we should add an |
20 |
intermediate category with a somewhat automated workflow? So that the |
21 |
evolution of packages in the tree would follow such pattern (plus, of course, |
22 |
there are overlays for even less stable stuff): |
23 |
|
24 |
1. as the package gets released it goes into the "testing", as it does now |
25 |
|
26 |
2. After say 2-4 wekks (take your pick) without issues and possibly going |
27 |
trhough some compile-farm (automatically scheduled at the end of this period |
28 |
if no open bugs (normal or more severe) are detected) the package is marked |
29 |
as belonging to the "tested" category. This is where many users would set |
30 |
their running stability level and, in a way, participate in testing things |
31 |
for the next level. |
32 |
|
33 |
If, any time after entering the "tested" state, package gets a bug assigned |
34 |
against it (again, normal or more severe) we start a countdown of, perhaps a |
35 |
few days, to let developers take care of the bug and if it does not get |
36 |
resolved within this time frame the package goes back to "testing". The |
37 |
decision to mask it or pull it off the tree completely remains with the |
38 |
respective devs, as it is now. |
39 |
Some packages can be marked as "critical" to make them go back to "testing" |
40 |
immediately upon getting an open bug, should such effect be desired (might be |
41 |
usefull for some security-sensitive system packages, or may be not, due to |
42 |
possible breakage this may introduce. Still we are having such downgrade |
43 |
situations already from time to time). |
44 |
|
45 |
3. "completely stable" profile, to which packages only go when explicitly |
46 |
requested and processed by stabilization team, as they are now. We should |
47 |
probably impose the requirement, that stabilization can only be requested for |
48 |
packages in the "tested" category. |
49 |
|
50 |
The good thing about this approach is that it only requires an initial |
51 |
investment of organizing and automating things but does not add any regular |
52 |
work to the devs. In fact, if the "tested" category becomes popular enough, |
53 |
it can cut the work for stable testers, may be even by something like a |
54 |
factor of 10 eventually (due to less requests for explicit stabilizaion being |
55 |
issued).. |
56 |
|
57 |
George |
58 |
-- |
59 |
gentoo-dev@l.g.o mailing list |