1 |
On Wed, 13 Aug 2008 23:33:04 +0300 |
2 |
Petteri Räty <betelgeuse@g.o> wrote: |
3 |
|
4 |
As a distribution we should strive to make as many packages available |
5 |
with as many features as possible on as many architectures (or indeed |
6 |
operating systems) as possible.[1] Not communicating important changes |
7 |
in ebuilds to arch teams, even making decisions on their behalf, we risk |
8 |
having to mend increasingly complex systems of profiles and flags. |
9 |
|
10 |
> I have been instructing people to adjust the files themselves. The |
11 |
> changes affect only the package in question and as such it falls |
12 |
> under the responsibility of the maintainer of the package. |
13 |
|
14 |
Sadly, I've been adding capitalised boilerplate pleas to the heads |
15 |
(and sometimes the tails) of hppa profile files requesting bug reports |
16 |
instead of adding silently to the ancient cruft that's there already. |
17 |
That doesn't mean that either you or I are wrong, but it does clearly |
18 |
show that we should put this all down in writing[2] when we find |
19 |
agreement. :) |
20 |
|
21 |
> > I personally think no, individual ebuild devs shouldn't touch |
22 |
> > arch-profiles. They should simply drop the (broken) keywords and |
23 |
> > file a keywordreq bug for those arches. Then the arch-teams can |
24 |
> > test and eventually decide whether to keyword the package or mask |
25 |
> > the use-flag. |
26 |
|
27 |
There should be no problem committing the ebuild with dropped keywords, |
28 |
then filing a KEYWORDREQ bug report describing the problem and the |
29 |
solution, and CC'ing the stricken arch teams. |
30 |
|
31 |
> The arch teams don't have that much staff so not adding to their work |
32 |
> load is the best way to go imho. |
33 |
|
34 |
Cleaning up *use.mask is annoying work and removing a flag from |
35 |
*use.mask could affect many users (emerge --newuse world). We should |
36 |
therefore reluctantly mask USE flags and let profile maintainers decide |
37 |
what is useable and practical to have on their OS/arch. When in doubt |
38 |
whether to mask a USE flag or drop a keyword, a package maintainer |
39 |
should commit the ebuild, drop the keyword and file a bug to have the |
40 |
arch team decide. Important note: leaving the keyword dropped is no |
41 |
option for active, security supported arches[3]. |
42 |
|
43 |
|
44 |
Kind regards, |
45 |
JeR |
46 |
|
47 |
|
48 |
[1] That's my interpretation of much of what |
49 |
http://www.gentoo.org/main/en/philosophy.xml says. |
50 |
|
51 |
[2] The end result should probably trickle down into the devmanual |
52 |
(and/or the Developer Handbook?) at some point. I feel like writing |
53 |
a section or two about arch workflows and keywording processes that |
54 |
could be included in |
55 |
http://devmanual.gentoo.org/keywording/index.html . |
56 |
|
57 |
[3] When an arch team cannot handle the workload any longer, the support |
58 |
level should be dropped to ~arch anyway. I would really hate to |
59 |
see people do what amounts to crippling packages (removing |
60 |
features), just to decrease some arch team's workload. In the mean |
61 |
time, do pile up the work regardless of staffing problems - somebody |
62 |
new may just volunteer to help an arch team out when the going gets |
63 |
tough. |