1 |
Le jeudi 13 mars 2008 à 10:15 -0400, Caleb Tennis a écrit : |
2 |
> > +1 on that and if people who use binary pkgs don't tell us what breaks, |
3 |
> > we won't know. |
4 |
> |
5 |
> I'll kick it off, then. |
6 |
> |
7 |
> The binpkg format needs some way to store the actual versions of the dependencies as |
8 |
> they were on the machine the package was compiled on. Then, when emerging the |
9 |
> binpkg, someway to force those dependencies on the new install machine would be |
10 |
> nice. |
11 |
> |
12 |
> I'll give an example. Package A was built on machine 1, and has a dep on |
13 |
> >=openssl-0.9.7. Machine 1 has openssl-0.9.8 already installed. Binary package |
14 |
> built, no problem. |
15 |
> |
16 |
> Now, we attempt to install binary package A on machine 2, which has openssl-0.9.7. |
17 |
> It installs fine, deps met. But, whoops, there's some symbols missing when we go to |
18 |
> use package A on machine 2. After some time, we finally realize it's because we |
19 |
> need new openssl. |
20 |
> |
21 |
> I use this example because it's actually hit me before, but it extends to lots of |
22 |
> other scenarios. The obvious fix is to either use --deep, or just make sure you |
23 |
> need machine 2 up to date with machine 1, though that's difficult to do when you're |
24 |
> talking about machine 301 and machine 559. |
25 |
> |
26 |
> If there was a way to tell the bin package installer to make sure you met all of the |
27 |
> same minimum verisons of the deps as they were on the original compiling machine, |
28 |
> that would be fantastic. |
29 |
> |
30 |
> Now, I'm happy to file a bug and assign it (to the portage team?), but I view this |
31 |
> really as a wishlist item, and since admittedly very few devs use the binpkg stuff, |
32 |
> I didn't see it as something that would probably get acted upon anyway. I'm not |
33 |
> complaining about that either, just merely stating a fact. |
34 |
|
35 |
I think remi was more speaking about incorrect deps (say misplaced in |
36 |
RDEPEND) than problems concerning the package manager. |
37 |
|
38 |
In any case, openssl is the perfect example of what can go wrong because |
39 |
of upstream's behavior. The problem is that program A compiled against |
40 |
version X of openssl won't work with version Y>X. Currently we need to |
41 |
keep X's libs around and run revdep-rebuild to fix this. |
42 |
|
43 |
Most librairies don't cause this problem though so I don't really see |
44 |
this as a bug on the gentoo side even if it's annoying. |
45 |
|
46 |
Anyway, to keep machines using binary in sync without much headache, my |
47 |
current solution is to use a squashfsed portage tree with --deep. It |
48 |
works pretty well. |
49 |
-- |
50 |
Gilles Dartiguelongue <eva@g.o> |
51 |
Gentoo |
52 |
|
53 |
-- |
54 |
gentoo-dev@l.g.o mailing list |