1 |
On Saturday 17 October 2009 01:29:00 Daniel Bradshaw wrote: |
2 |
> Some packages, like findutils, are pretty robust and generally just get on |
3 |
> with working. |
4 |
> Other packages, like apache and ssh, need are more fragile and need plenty |
5 |
> of configuration. |
6 |
That's almost completely user-side configuration outside the influence of |
7 |
portage. emerge findutils and emerge apache "works" the same ... |
8 |
|
9 |
|
10 |
> Packages from the second group want emerging on their own, or in small |
11 |
> groups, the better to keep an eye out for notices about things that might |
12 |
> break, to update configs, and to check that they're running happily. |
13 |
That's a very individual thing :) |
14 |
Sometimes apache is a critical service, sometimes apache is just there as a |
15 |
fallback if/when the lighttpd+php+... stack breaks. |
16 |
|
17 |
> Once the update list is reduced to packages from the first group it's |
18 |
> fairly safe to run emerge -u world and not worry about things exploding |
19 |
> too badly. |
20 |
> |
21 |
> |
22 |
> So as I say, it occurs to me that most people probably follow some |
23 |
> variation of this selective upgrade method. |
24 |
> It might be handy to have some kind of metadata in the ebuilds that can be |
25 |
> used to indicate a package that is "demanding". |
26 |
There's one issue with that - library updates and compiler updates and all |
27 |
those other variations. I had a funky crashbug in amule for a while that, as |
28 |
far as I can tell, is a mix of a confused crypto lib, wxGTK and amule itself |
29 |
not agreeing much. It disappeared with the last updates, and it would only |
30 |
trigger in this constellation. |
31 |
|
32 |
So what might work perfectly for most people could still violently explode for |
33 |
you. We already try to tag things with KEYWORDS - while that isn't the best |
34 |
indicator it gives you a general idea that ~arch is a bit less predictable |
35 |
than arch. And if it comes from an overlay you're on your own anyway. Why add |
36 |
just another level of labelling to that? |
37 |
|
38 |
From my point of view that's just adding more noise instead of fixing the |
39 |
problem (package quality and interaction bugs). But that would take lots of |
40 |
time, lots of motivated people and will never stop :) |