| 1 |
On Thu, Jan 3, 2013 at 10:46 AM, Maxim Kammerer <mk@×××.su> wrote: |
| 2 |
> The problem is that the directory was apparently included into daily |
| 3 |
> snapshot between 00:31:07 UTC (ebuild commit) and 00:31:13 UTC |
| 4 |
> (Manifest commit). For those that don't remember, CVS does not have |
| 5 |
> atomic commits, so it's not possible to instead use something like svn |
| 6 |
> export / git archive. |
| 7 |
|
| 8 |
Note that this problem is compounded by Manifest signing: because |
| 9 |
ebuilds are committed before the Manifest is signed, there is a |
| 10 |
potentially long gap between the commit of the ebuilds and the commit |
| 11 |
of the Manifest. I ran into this yesterday, when I forgot about the |
| 12 |
password dialog sitting in some terminal I wasn't looking at for about |
| 13 |
half an hour, and at some point someone in IRC noticed that the |
| 14 |
Manifest was out of date. I'm not sure how common it is for people to |
| 15 |
interactively enter their password on each commit like me. |
| 16 |
|
| 17 |
In any case, we probably shouldn't spend a whole lot of effort on this |
| 18 |
given the somewhat-impending git migration, which neatly solves this |
| 19 |
problem. Maybe there's some low-hanging fruit in the commit ordering, |
| 20 |
though? I was thinking it might be possible to have the Manifest |
| 21 |
signed before committing the ebuilds, but it's entirely likely that |
| 22 |
CVS keywords get in the way of that... |
| 23 |
|
| 24 |
Cheers, |
| 25 |
|
| 26 |
Dirkjan |
Archives