1 |
On 16 September 2012 09:20, Brian Harring <ferringb@×××××.com> wrote: |
2 |
> On Sun, Sep 16, 2012 at 12:03:36AM +0200, Micha?? G??rny wrote: |
3 |
>> On Sat, 15 Sep 2012 13:33:18 -0700 |
4 |
>> Brian Harring <ferringb@×××××.com> wrote: |
5 |
>> > To demonstrate the gain of this, we basically take the existing |
6 |
>> > tree's deps, and re-render it into a unified DEPENDENCIES form. |
7 |
>> |
8 |
>> But in order to do this, we first have to decide exactly what kind |
9 |
>> of dependencies do we want to have. Then convert the tree to |
10 |
>> a separate-variable form with new dependencies. Then we can compare |
11 |
>> it with the DEPENDENCIES form and decide which one is better. |
12 |
> |
13 |
> Funny you mentioned that, I just finished tweaking pquery to generate |
14 |
> real world example unified dependencies; these *are* accurate, just to |
15 |
> be clear. |
16 |
> |
17 |
> Dumps are at |
18 |
> http://dev.gentoo.org/~ferringb/unified-dependencies-example/ . |
19 |
> |
20 |
> Herds, if you want to see what your pkgs would look like, look at |
21 |
> http://dev.gentoo.org/~ferringb/unified-dependencies-example/herds/ . |
22 |
> |
23 |
> If you'd like to see an *example effect* it has on what gets displayed |
24 |
> to the user (aka, after all major use conditionals are stripped), look |
25 |
> at |
26 |
> http://dev.gentoo.org/~ferringb/unified-dependencies-example/user-visible.txt |
27 |
> ; warning, that's a 55MB file. The syntax in use there isn't great, |
28 |
> but as said, it's an example. |
29 |
> |
30 |
> Total cache savings from doing this for a full tree conversion, for |
31 |
> our existing md5-cache format is 2.73MB (90 byes per cache entry). |
32 |
> Calculating the savings from the ebuild/eclass standpoint is dependent |
33 |
> on how the deps are built up, so I skipped that. |
34 |
> |
35 |
> The algorithim used is fairly stupid, but reasonably effectively; |
36 |
> essentially it intersects the top level of each individual type of |
37 |
> dep, breaking out common groupings. |
38 |
> |
39 |
> In other words, it won't pick up this: |
40 |
> DEPEND="x? ( dev-util/diffball dev-util/bsdiff )" |
41 |
> RDEPEND="x? ( dev-util/diffball )" |
42 |
> |
43 |
> and convert it into thus |
44 |
> DEPENDENCIES=" |
45 |
> dep:build,run? ( |
46 |
> x? ( |
47 |
> dev-util/diffball |
48 |
> dep:run? ( |
49 |
> dev-util/diffball |
50 |
> ) |
51 |
> ) |
52 |
> )" |
53 |
> |
54 |
> Additionally, the form used here makes *no assumption about default |
55 |
> context*; in any final solution we use, a default context would be |
56 |
> wise- say build,run. Again, an example of what I mean. |
57 |
> |
58 |
> If we said "in the absense of a context, the default is dep:build,run" |
59 |
> the following: |
60 |
> |
61 |
> DEPEND="dev-util/diffball dev-util/bsdiff" |
62 |
> RDEPEND="dev-util/diffball de-vutil/bsdiff x? ( sys-apps/pkgcore )" |
63 |
> PDEPEND="dev-python/snakeoil" |
64 |
> |
65 |
> would be: |
66 |
> DEPENDENCIES=" |
67 |
> dev-util/diffball |
68 |
> dev-util/bsdiff |
69 |
> dep:run? ( x? ( sys-apps/pkgcore ) ) |
70 |
> dep:post? ( dev-python/snakeoil ) |
71 |
> " |
72 |
> |
73 |
> The quicky algo I used assumes no default context, thus it writes |
74 |
> this: |
75 |
> DEPENDENCIES=" |
76 |
> dep:build,run? ( |
77 |
> dev-util/diffball |
78 |
> dev-util/bsdiff |
79 |
> ) |
80 |
> dep:run? ( x? ( sys-apps/pkgcore ) ) |
81 |
> dep:post? ( dev-python/snakeoil ) |
82 |
> " |
83 |
> |
84 |
> Etc. |
85 |
> |
86 |
> ~harring |
87 |
> |
88 |
|
89 |
Thanks. I have given it a quick overview for the qt herd. I still |
90 |
don't see what using DEPENDENCIES adds to what we do now with separate |
91 |
*DEPEND variables. I see no convincing reason to change what we do. |
92 |
|
93 |
As I've said before on IRC, we need a good costs/benefits overview. |
94 |
Right now I only see costs (migrating ebuilds and eclasses) and no |
95 |
benefits. |
96 |
|
97 |
-- |
98 |
Cheers, |
99 |
|
100 |
Ben | yngwin |
101 |
Gentoo developer |
102 |
Gentoo Qt project lead, Gentoo Wiki admin |