1 |
On Tue, May 19, 2009 at 9:38 AM, Fabian Groffen <grobian@g.o> wrote: |
2 |
> On 19-05-2009 09:30:13 -0500, Jeremy Olexa wrote: |
3 |
>> On Tue, May 19, 2009 at 5:49 AM, Fabian Groffen <grobian@g.o> wrote: |
4 |
>> > Hi all, |
5 |
>> > |
6 |
>> > While cleaning up the Darwin and Solaris profiles, I noticed that there |
7 |
>> > is some redundancy in masks (necessary) due to the structure based on |
8 |
>> > single inheritance: |
9 |
>> > |
10 |
>> > <os>/<release>/<arch> |
11 |
>> > e.g. |
12 |
>> > darwin/10.4/ppc |
13 |
>> > darwin/10.4/x86 |
14 |
>> > darwin/10.5/ppc |
15 |
>> > darwin/10.5/x86 |
16 |
>> > |
17 |
>> > Most e.g. masks in the ppc dirs are specific for the CPU, and not really |
18 |
>> > bound to the OS release. So I propose the following structure: |
19 |
>> > |
20 |
>> > <os>/<arch> |
21 |
>> > <os>/<release>/<arch> (inherits from .. and ../../<arch>) |
22 |
>> > |
23 |
>> > This way a single mask can e.g. apply to all PPC versions of Mac OS X. |
24 |
>> > Similar for Solaris' sparc, sparc64, x86 and x64. |
25 |
>> > |
26 |
>> > Does this make sense, or am I missing something obvious? Better ideas? |
27 |
>> > |
28 |
>> > (note that I (hopefully) kept full backwards compatability with this scheme) |
29 |
>> |
30 |
>> Like: |
31 |
>> darwin/ppc/10.4 |
32 |
>> darwin/x86/10.4 |
33 |
>> darwin/ppc/10.5 |
34 |
>> darwin/x86/10.5 |
35 |
> |
36 |
> No, just: |
37 |
> |
38 |
> darwin/ppc |
39 |
> darwin/x86 |
40 |
> darwin/10.4/ppc |
41 |
> darwin/10.4/x86 |
42 |
> darwin/10.5/ppc |
43 |
> darwin/10.5/x86 |
44 |
|
45 |
The less inheritance you have, the better/easier it is, IMO. So, I |
46 |
would greatly prefer what I suggested to what you are suggesting. I |
47 |
think. The end result is the same besides aesthetics. You can place a |
48 |
mask in darwin/ppc/package.mask with both methods proposed. |
49 |
|
50 |
> |
51 |
>> Is that how you are planning to do it? Because I think that makes the |
52 |
>> most sense. If you look at linux/ - they are linux/<arch>/ (they skip |
53 |
>> the <release> because it isn't needed) |
54 |
> |
55 |
> Well, that's just doing it the other way around? |
56 |
> |
57 |
>> Do also realize that this applies to *all* profiles, not just darwin |
58 |
>> and solaris. Windows look like the only ones that won't need changing |
59 |
>> because linux. |
60 |
> |
61 |
> Hmmm... not sure if other profiles are much in need for it. Alternative |
62 |
> is to create a prefix/arch/{x86,ppc,sparc,...}, or maybe even start |
63 |
> inheriting arch/X. In principle looks ok, but its linux biased |
64 |
> sometimes... |
65 |
|
66 |
ew. No. prefix/arch/... has no business for us. The reason I say that |
67 |
other profiles can benefit from my suggestion is when you look at |
68 |
hpux, it will eventually have hppa and 11.{23,31}/{ia64,hppa} just |
69 |
doesn't make sense when you need a mask for all hppa for example. The |
70 |
same with sunos. So, maybe not all, but some profiles can use this |
71 |
idea. |