1 |
On Sun, Feb 10, 2013 at 7:51 AM, Alexander Berntsen |
2 |
<alexander@××××××.net> wrote: |
3 |
> |
4 |
> On 10/02/13 13:11, Rich Freeman wrote: |
5 |
>> - just look up your average non-core piece of FOSS software and the |
6 |
>> first thing their Ubuntu install instructions will tell you to do |
7 |
>> is to add some repository to your list. |
8 |
> And the second search result is the Ubuntu troubleshooting broken |
9 |
> installs as a result of adding other repositories. |
10 |
> |
11 |
> I accept that there may exist reasons for using overlays. "Ubuntu do |
12 |
> it!" is not one. |
13 |
|
14 |
I have mixed feelings on this. I'd never advocate doing anything |
15 |
simply because everybody else is doing it - if I wanted to use Ubuntu |
16 |
I'd be using Ubuntu. |
17 |
|
18 |
There are pros/cons to overlays right now: |
19 |
Pros include: |
20 |
1. More flexible maintenance model. The overlay maintainer can |
21 |
choose who has access to it. They don't have to worry about people |
22 |
making tree-wide commits without knowing what they're doing, because |
23 |
any damage is contained to the overlay (though obviously any package |
24 |
in an overlay could mess with anything on a user's system). |
25 |
|
26 |
2. More flexible QA model. Usually that means less QA, which has its |
27 |
own pros and cons, but it /could/ actually mean more QA, or just |
28 |
different QA. Right now we have no way of communicating to users |
29 |
(beyond masks) that packages vary in quality level, and overlays could |
30 |
be a way to accomplish this. You could also have a set of related |
31 |
overlays that provide a dev/test/stable experience. |
32 |
|
33 |
Cons include: |
34 |
1. No relationship to the tree. If somebody messes with one of your |
35 |
dependencies they will not take any care not to break your package. |
36 |
|
37 |
2. Non-mainstream experience. Because Gentoo tends to be |
38 |
overlay-averse, most users don't use them at all. |
39 |
|
40 |
3. No real organization. Beyond an entry in the layman list there |
41 |
really isn't any systematic tracking of overlays and their |
42 |
quality/etc. We don't "grade" overlays or anything like that. |
43 |
|
44 |
#1 is the biggest con I'd say. It is made worse by the fact that we |
45 |
don't have a main repository QA cycle (I'm not suggesting we have |
46 |
one). For something like Ubuntu anybody maintaining a 3rd party |
47 |
repository can monitor the release cycle and test against the new |
48 |
dependency versions before they are released and be ready on day one. |
49 |
For Gentoo you would have to pay very close attention to bugzilla, |
50 |
lists, irc, and perhaps even mail aliases (not open to the public) to |
51 |
have any idea that some change is about to happen to one of your |
52 |
dependencies if you aren't in the main tree. |
53 |
|
54 |
A fix for #1 might be some way to allow external parties to register |
55 |
interest in upcoming changes and get alerted. Then those changing |
56 |
libs could just trigger the alerts (and that system might also file |
57 |
bugs against in-tree packages to request testing). We obviously |
58 |
wouldn't consider any outside overlays blockers, but we could be nicer |
59 |
to them. Of course, that takes work and I'm skeptical that this would |
60 |
ever happen. |
61 |
|
62 |
So, those are just my thoughts on overlays. I don't think they're a |
63 |
bad thing. However, there are some things about Gentoo that make them |
64 |
less practical than on other distros. I won't argue that you get the |
65 |
best possible experience if the package is in the tree AND IT IS |
66 |
MAINTAINED. The problem is that in a volunteer-based organization the |
67 |
second half of that is hard to guarantee. |
68 |
|
69 |
Rich |