1 |
Problem: |
2 |
The current proposal can not specify which permutations are valid. |
3 |
|
4 |
Solution: |
5 |
I can't see one that doesn't either include ridiculous complexity or amount to |
6 |
being a "keyword explosion". |
7 |
|
8 |
Thoughts: |
9 |
The only way I can see it being fully supported is by widening the gap between |
10 |
testing and stable. Furthermore, testing would need to become *real* testing. |
11 |
Here are examples taken from the GLEP: |
12 |
|
13 |
ARCH: x86, amd64, cobalt, mips64, arm, hppa, ia64, ppc64, sparc |
14 |
KERNEL: linux, selinux, openbsd, freebsd, netbsd, macosx |
15 |
LIBC: glibc, openbsd, freebsd, netbsd, macosx |
16 |
USERLAND: gnu, bsd |
17 |
|
18 |
Now for ARCH, a recent (official?) policy states that no ARCH-specific patches |
19 |
should be applied to source code. If a patch needs to be applied, it should |
20 |
work on all ARCHs. To this end, I can't see why ARCH-specific keywords need |
21 |
to exist at all. This leaves: |
22 |
|
23 |
KERNEL: linux, selinux, openbsd, freebsd, netbsd, macosx |
24 |
LIBC: glibc, openbsd, freebsd, netbsd, macosx |
25 |
USERLAND: gnu, bsd |
26 |
|
27 |
Next I'll look at USERLAND. As far as I know, it's not possible to run a gnu |
28 |
USERLAND on anything other than a glibc LIBC. Conversely, a bsd USERLAND |
29 |
cannot run on a glibc LIBC. I can only imagine very rare circumstances where |
30 |
a LIBC would support multiple USERLANDs and an application would care what |
31 |
specific USERLAND it is. To this end, the USERLAND can be dropped and rolled |
32 |
into LIBC if it is ever required. This leaves: |
33 |
|
34 |
KERNEL: linux, selinux, openbsd, freebsd, netbsd, macosx |
35 |
LIBC: glibc, openbsd, freebsd, netbsd, macosx |
36 |
|
37 |
I'll just put forward the permutations of these that I am aware of and leave |
38 |
them for later discussion: |
39 |
|
40 |
linux-glibc, selinux-glibc, freebsd-glibc, netbsd-glibc, openbsd-openbsd, |
41 |
freebsd-freebsd, netbsd-netbsd, maxosx-macosx. |
42 |
|
43 |
A total of 8 permutations.. How many ARCH keywords do we have at the moment? |
44 |
|
45 |
Of course, this idea is based on dropping ~arch and arch altogether. I don't |
46 |
have a non-x86 arch (yet!) but as far as I know most software will work with |
47 |
no modification. For software that doesn't work and a quick fix can't be |
48 |
found, I would suggest a new portage/profiles/<profile>/package.mask. |
49 |
|
50 |
Well, that's enough for now. Who'll be the first to say how bad this idea is, |
51 |
I wonder... ;) |
52 |
|
53 |
Regards, |
54 |
Jason Stubbs |
55 |
|
56 |
-- |
57 |
gentoo-dev@g.o mailing list |