1 |
On Sat, Mar 27, 2010 at 11:06 PM, Doug Goldstein <cardoe@g.o> wrote: |
2 |
> On Sat, Mar 27, 2010 at 9:44 AM, Petteri Räty <betelgeuse@g.o> wrote: |
3 |
>> On 03/24/2010 08:30 PM, Peter Hjalmarsson wrote: |
4 |
>> |
5 |
>>> For qemu-kvm the problem is that there is only one implementation (i.e. |
6 |
>>> gnutls), and if I want to have ssl support I have to enable gnutls for |
7 |
>>> this package. |
8 |
>> |
9 |
>> In this case the ebuild should have only ssl use flag. |
10 |
>> |
11 |
>>> When I wrote a bug about this I got a rather short reply from maintainer |
12 |
>>> about pointing me to the policy about this. |
13 |
>> |
14 |
>> Where did he point you to? |
15 |
> |
16 |
> I didn't point him anywhere. I merely asked him for a policy on this. |
17 |
> Because senseless changes in USE flags will require my 9 VM servers |
18 |
> will need to be tweaked around for a pointless USE flag change and I |
19 |
> don't need administrative burden for the sake of administrative |
20 |
> burden. |
21 |
|
22 |
I have concerns with this which I will attempt to summarize below. |
23 |
|
24 |
1) Traditional binary distributions have releases that take users over |
25 |
these changes (dist-upgrade and similar processes.) |
26 |
2) Gentoo has no releases, so 'annoying administrative changes' are |
27 |
certainly more challenging, because it is difficult to apply them to |
28 |
new machines and not old ones. |
29 |
3) Cleanup changes such as the one proposed are good for the health of |
30 |
Gentoo. Consistency in flags actually makes it *easier* for users to |
31 |
configure their systems. Consistency in ebuild behavior makes it |
32 |
*easier* developers to maintain ebuilds. |
33 |
4) Not making changes because they may affect existing systems is |
34 |
crap. It holds the distribution back because everyone will always |
35 |
pull this card out to veto major changes. |
36 |
|
37 |
It would be interesting if we could make these changes less painful: |
38 |
|
39 |
1) For this change, attempt to detect how users are using these flags |
40 |
and migrate them to the new system. This will likely be easy for the |
41 |
80% case (/etc/portage/...) and hard for the 20% case (cross-compiles, |
42 |
and other odd things involving ROOT.) |
43 |
2) Defer changes like this to some kind of release date. Write down |
44 |
what you want to do somewhere. At the prescribed time (once a [month, |
45 |
year, quarter?]) apply all the changes to the tree and release GLEP 42 |
46 |
news items, changelogs, webpage stuff, forums posts, etc. |
47 |
3) other crap I haven't thought of. |
48 |
|
49 |
> |
50 |
>> |
51 |
>>> So I have a question: |
52 |
>>> Is there no policy about this? |
53 |
>> |
54 |
>> The policy is that USE="ssl" controls whether to enable ssl support in |
55 |
>> general. Then the specific use flags like gnutls and openssl control |
56 |
>> what implementation to use if the package supports multiple. |
57 |
> |
58 |
> Again, this policy is stated but no one can point me to anything. The |
59 |
> closest thing to a "policy" is you sending a follow up e-mail to the |
60 |
> dev list to make this a policy. |
61 |
> |
62 |
> -- |
63 |
> Doug Goldstein |
64 |
> |
65 |
> |