1 |
On 08/18/2011 05:15 AM, Thomas Sachau wrote: |
2 |
> Diego Elio Pettenò schrieb: |
3 |
>> Hello everybody, |
4 |
>> |
5 |
>> I have already said this before, but it looks like nobody cared. We have |
6 |
>> a problem for what concerns Gentoo-generated distfiles. |
7 |
>> |
8 |
>> This includes custom snapshots, custom packages, patches, patchsets, and |
9 |
>> so on so forth. While it was infra that (back when I joined at least) |
10 |
>> asked not to use dev.gentoo.org for hosting said fails and rather prefer |
11 |
>> to use mirror://gentoo/, they already stated that it's not a problem to |
12 |
>> do so until we have a proper system in place (system that has been |
13 |
>> considered and worked on for quite a bit already and yet is not |
14 |
>> available). |
15 |
>> |
16 |
>> Unfortunately, as long as the mirror://gentoo/ option is still |
17 |
>> maintained, we'll end up with situations like today's gnuconfig that |
18 |
>> couldn't be fetched, causing all ~arch users to see the same failure, |
19 |
>> because the distfile wasn't uploaded to the staging area. Of course the |
20 |
>> same could happen with a stable SRC_URI, but then it would fail after |
21 |
>> hitting that, rather than going through half the gentoo mirrors trying |
22 |
>> to find a file that is not there. |
23 |
>> |
24 |
>> So, anybody has reasons beside laziness, or concern for infra's disk |
25 |
>> usage (that argument is allowed to come only from infra members!), to |
26 |
>> not go this route? |
27 |
>> |
28 |
> |
29 |
> I see no improvement from this proposal. If a tarball is not uploaded, the users will always fail to |
30 |
> get it, independent from the place, where it should be going to. And even if you put the |
31 |
> dev.gentoo.org address in SRC_URI, portage will first check the mirrors before it checks SRC_URI, so |
32 |
> this would in the end result in even 1 more download try without success. |
33 |
> |
34 |
> The argument about dropped tarballs, once the ebuilds gets removed might weight a bit more, but you |
35 |
> cannot depend on other upstream keeping their tarballs around forever, so i see no requirement for |
36 |
> us preserving only specific tarballs (those created by our devs), while upstream tarballs could |
37 |
> already be gone and are not preserved, once the ebuild for it is gone. |
38 |
> |
39 |
|
40 |
What alternative are you proposing to mirror://gentoo/ if upstream |
41 |
doesn't provide a tarball, eg with large patchsets the maintainer |
42 |
constructs? Anticipating your answer might be "keep them in your dev |
43 |
space", then what would be the deprecation policy for distfiles that are |
44 |
no longer used by ebuilds? If foresee a tension between keeping all the |
45 |
distfiles vs disk space, although Patrick's point of cheap hardware is |
46 |
well taken. |
47 |
|
48 |
|
49 |
-- |
50 |
Anthony G. Basile, Ph.D. |
51 |
Gentoo Linux Developer [Hardened] |
52 |
E-Mail : blueness@g.o |
53 |
GnuPG FP : 8040 5A4D 8709 21B1 1A88 33CE 979C AF40 D045 5535 |
54 |
GnuPG ID : D0455535 |