1 |
Aron Griffis wrote: |
2 |
> George Shapovalov wrote:[Mon May 26 2003, 02:49:36PM EDT] |
3 |
> |
4 |
>> As such it makes sense to put some absolutely something basic in |
5 |
>> DEPEND for the package which does not have any apparent (as |
6 |
>> non-trivial, not bad-researched) dependencies. In that respect |
7 |
>> virtual/glibc perfectly fits the bill... |
8 |
> |
9 |
> |
10 |
> DEPEND="" works just as well. Since glibc is in the default profile, |
11 |
> there's no reason to list it as a dependency, just like we can |
12 |
> assume some basic shell utilities are present when the ebuild is |
13 |
> processed. |
14 |
> |
15 |
|
16 |
Depending on the degree you are talking here, I tend to disagree with |
17 |
the notion of not including some so call "basic" dependencies, because |
18 |
profiles can have variations that may include stripping out some core |
19 |
components. As we move into the embedded realm, the definition of our |
20 |
"core profile" will need to be reduced, so adding full dependency |
21 |
information now is not necessarily a bad thing. |
22 |
|
23 |
Instead, we need to clarify a truly minimal system set of packages (such |
24 |
as libc) for which their dependency information need not be included. |
25 |
This will probably be taken up as part of the embedded project's efforts |
26 |
to scale the system down, so there is not pressure to do this now. |
27 |
|
28 |
The tools already exists to build a dependecy list by tracking what the |
29 |
actual build process uses during execution, and I have seen a couple of |
30 |
ebuilds in the tree that seem to have used them. This may not be a bad |
31 |
standard practice to follow if those few core packages can be put on a |
32 |
'whitelist' for exclusion by those tools. At some distant point, |
33 |
automating this process with a tool would not be a bad thing: |
34 |
|
35 |
emerge --makedeps package-1.0.ebuild |
36 |
|
37 |
This would build the package, tracking the deps, and modifying the |
38 |
ebuild to update the build dependencies. The recent thread that talked |
39 |
about collecting branch prediction information by automatically running |
40 |
the program a few times suggests to me that a similar process could be |
41 |
applied to also detect RDEPENDS. The above flag should also imply the |
42 |
use of --digest run as a finalization step. |
43 |
|
44 |
Cheers, |
45 |
|
46 |
Zach Welch |
47 |
Gentoo Embedded Lead |
48 |
Superlucidity Services |
49 |
|
50 |
|
51 |
|
52 |
-- |
53 |
gentoo-dev@g.o mailing list |