1 |
On 8/27/13 3:01 AM, Michał Górny wrote: |
2 |
> b) assume that early src_fetch() is allowed to fail and retry before |
3 |
> build. This is much like what portage does anyway. If VCS is not |
4 |
> installed yet during parallel fetch or --fetchonly, the particular |
5 |
> fetch fails like it can fail due to 404. When the actual build process |
6 |
> starts, emerge calls src_fetch() again much like it currently retries |
7 |
> fetching missing SRC_URI items. |
8 |
> |
9 |
> This would be simpler. We could probably cache the src_fetch() result |
10 |
> (if possible) to avoid double-fetching. Or just accept the minimal |
11 |
> overhead of plain update check. |
12 |
|
13 |
I think b) is better. Having a USE flag with a special meaning sounds |
14 |
bad; adding FDEPEND could be reasonable, but then --fetchonly behavior |
15 |
may become counter-intuitive. |
16 |
|
17 |
It seems to me b) is the most reasonable option, and also has the |
18 |
advantage of being simpler as you've said. |
19 |
|
20 |
> 2. Should src_unpack() be disallowed from fetching VCS sources? |
21 |
> |
22 |
> If we introduce src_fetch() in EAPI 6, then we should probably expect |
23 |
> all the live eclasses to use that rather than src_unpack() for |
24 |
> fetching. If we agree on this, we can extend portage's network-sandbox |
25 |
> to this phase, and likely filesystem sandbox as well. |
26 |
|
27 |
I'm fine with this and think it's a good idea, but it's kind of |
28 |
secondary - things will still "work" whether or not we do that. |
29 |
|
30 |
Paweł |