1 |
I think that one of the crucial strengths of Gentoo is its ability to |
2 |
quickly incorporate and disseminate information, both new software and |
3 |
information about how well things work in a particular environment. |
4 |
|
5 |
If the notions of ARCH, OS, LIBC, KERNEL were truly orthogonal, this |
6 |
GLEP would make sense to me. In other words, if one could say with |
7 |
certainty that "ppc is ok", or "won't work with glibc", without |
8 |
needing to look at the other variables at all, then this would be |
9 |
fine, but I don't think that accurately represents reality. |
10 |
|
11 |
Leaving aside for the moment the question of who does QA and when and |
12 |
in what capacity, I think that the main purpose of KEYWORDS is to |
13 |
convey QA information. I will oversimplify at this point and roughly |
14 |
map arch to green, ~arch to yellow and -arch to red. This is the |
15 |
information users want and we want to convey: here's the tinderbox |
16 |
color of the package you're interested in as it applies to your |
17 |
machine. |
18 |
|
19 |
I think it would give much more flexibility and power to the QA |
20 |
process if this information was maintained completely separately from |
21 |
within ebuilds themselves. It would be a big matrix, currently |
22 |
5-dimensional: arch/os/libc/kernel/ebuild-version = color. |
23 |
|
24 |
This way, QA testers can convey the information about how the software |
25 |
behaves on exactly the systems they have, without necessarily having |
26 |
to meet the burden of collecting enough data about how it behaves on |
27 |
platforms they may have no access to. Others can do that at a later |
28 |
time, and we do not waste the timely information that we *do* have. |
29 |
|
30 |
I suggest that it would be neither practical nor particularly useful |
31 |
to have users directly accessing the master matrix, which would |
32 |
probably reside in a relational database. Instead, there would be |
33 |
some process that could be adjusted based on demand for various slices |
34 |
of the matrix that rips static dictionaries of it, perhaps in dbm |
35 |
format. ebuild-version => color. Portage should have no greater |
36 |
burden at emerge time than it does now, and it may in fact be somewhat |
37 |
less, I'm not sure. |
38 |
|
39 |
The matrix slice could be retrieved at emerge sync time, either as |
40 |
part of the normal rsync process or via a separate http fetch. |
41 |
|
42 |
|
43 |
-- |
44 |
gentoo-dev@g.o mailing list |