1 |
Fabian Groffen wrote: |
2 |
> Hi all, |
3 |
> |
4 |
> Since I've failed to properly communicate the move of /profiles from our |
5 |
> overlay to gentoo-x86 (ok, I blame it on a surprise commit by someone |
6 |
> else :) ) I want to share my thoughts about the two here with you before |
7 |
> taking action this time ;) |
8 |
> |
9 |
> What basically annoys the hell out of me in the current system is that a |
10 |
> package as for instance mail-client/mutt is fully Prefix Aware(tm) in |
11 |
> gx86 -- including keywords -- but still is synced to the Prefix tree. |
12 |
> Let alone packages like gcc-apple, which are Prefix-only. |
13 |
> |
14 |
> It also annoys me that there are two places to commit to now, as well. |
15 |
> For example, should a fix to an eclass go to gx86, or the Prefix tree? |
16 |
> |
17 |
> As I think we'll have a transition period for a long while from now I've |
18 |
> been playing with the thought to make our lives a little bit easier by |
19 |
> allowing to eliminate duplicate packages, that is the exact same |
20 |
> version in gx86 and the Prefix tree. |
21 |
> |
22 |
> The idea is simple, from an rsync perspective. We start "overlaying" |
23 |
> gx86, overwriting all packages in there with their Prefix tree versions |
24 |
> if existing. It takes a bit of hacking in the rsync generation scripts, |
25 |
> but I think I can get the following running: |
26 |
> - update gx86 cvs checkout |
27 |
> - rsync it without CVS dirs to the candidate image |
28 |
> - update Prefix tree svn checkout |
29 |
> - remove all packages in the candidate image which exist in the Prefix |
30 |
> tree |
31 |
> - rsync the Prefix tree without .svn dirs over the candidate image |
32 |
> - distribute the candidate image to rsync1 |
33 |
> |
34 |
> Due to the full gx86 checkout, the generation frequency has to go down a |
35 |
> bit, as it takes a long time to update that tree, but that's details. |
36 |
> The idea here is that we can remove packages which are fully Prefix |
37 |
> Aware(tm), such as mutt, and only maintain it in gx86. This would also |
38 |
> make it easy to get it clear what's the todo list of us to merge in |
39 |
> gx86; if we have an empty tree, we're done :) (at that time we can |
40 |
> simply make the prefix rsync alias point to the gx86 rsync mirrors). |
41 |
> |
42 |
> Thoughts? Objections? |
43 |
> |
44 |
> |
45 |
|
46 |
Wouldn't it be more worthwhile to merge our portage changes back to |
47 |
trunk and get a prefix aware portage released officially before we worry |
48 |
about merging individual packages/eclasses? |
49 |
|
50 |
-Jeremy |