1 |
On Mon, 1 Jul 2013 21:56:02 +0200 |
2 |
Arfrever Frehtes Taifersar Arahesis <arfrever.fta@×××××.com> wrote: |
3 |
> > The following should be treated with extreme caution, have unobvious |
4 |
> > implications, need substantial work or are otherwise probably more |
5 |
> > dangerous than they're worth, especially if we want EAPI 6 this |
6 |
> > year: |
7 |
> > |
8 |
> > > 08. Will you vote for including support for version ranges in |
9 |
> > > EAPI 6 (assuming that a patch is available for Portage)? |
10 |
> |
11 |
> Already implemented in Paludis (not in official EAPIs). |
12 |
> Is it a mistake that Paludis supports this feature? |
13 |
|
14 |
For users, it's useful. But EAPIs have nothing to do with user-facing |
15 |
features, and Paludis does not support this feature in a Gentoo EAPI. |
16 |
There needs to be a discussion and policy on how exactly they're allowed |
17 |
to be used if they're permitted in the tree. |
18 |
|
19 |
> Already implemented in Portage (not in official EAPIs). |
20 |
|
21 |
This as an argument for a feature is wrong in at least three ways. |
22 |
|
23 |
Firstly, an implementation is sometimes necessary to see why something |
24 |
isn't as good an idea as it initially looks. |
25 |
|
26 |
Secondly, there's a big difference between having something implemented |
27 |
in a package mangler, and having experience with that feature in the |
28 |
tree. Being able to implement something is a necessary condition, not a |
29 |
sufficient one. |
30 |
|
31 |
Thirdly, an implementation of a feature for users is not the same as |
32 |
allowing it in the tree. |
33 |
|
34 |
Repository dependencies are a good example of all three of these |
35 |
points. For users, they're useful (although the limited way Portage has |
36 |
them implemented removes a lot of that). But in the tree they have |
37 |
problems that we know about (see previous discussions that you like to |
38 |
ignore when pushing for this feature) and probably some that we don't |
39 |
too, since no-one's ever tried using them in the tree. We've already |
40 |
established that repository dependencies in the tree are the wrong |
41 |
solution, so you using "there's an implementation!" in their favour is |
42 |
at best disingenuous. |
43 |
|
44 |
> > The following proposals are very bad, and implementing them would |
45 |
> > be a mistake: |
46 |
> > |
47 |
> > > 11. Will you vote for providing master_repositories(), |
48 |
> > > repository_path(), available_eclasses(), eclass_path() and |
49 |
> > > license_path() functions in EAPI 6 (assuming that a patch is |
50 |
> > > available for Portage)? |
51 |
> |
52 |
> This feature provides multiple-repository-friendly replacement for |
53 |
> single-repository-specific PORTDIR and ECLASSDIR variables. |
54 |
> Already implemented in Portage (not in official EAPIs). |
55 |
|
56 |
The point of abolishing location variables is to allow things like |
57 |
partial syncs, not to replace them with a slightly different thing that |
58 |
imposes the same restrictions. |
59 |
|
60 |
But again, if you're trying to advocate for any particular feature, you |
61 |
should really discuss them on the relevant bugs, not try to use the |
62 |
Council elections to subvert the process. |
63 |
|
64 |
-- |
65 |
Ciaran McCreesh |