1 |
El Mie, 4 de Febrero de 2009, 7:17, Grant Edwards escribió: |
2 |
> On 2009-02-04, James <wireless@×××××××××××.com> wrote: |
3 |
> |
4 |
>> Grant Edwards <grante <at> visi.com> writes: |
5 |
>> |
6 |
>> |
7 |
>>> Whenever I see a write-up of Gentoo, it's described as a system |
8 |
>>> similar to BSD "ports" where you build packages from source. The main |
9 |
>>> benefit claimed for this approach is that you get better performance |
10 |
>>> because all executables are optimized for exactly the right |
11 |
>>> instruction set. is practically nil in real-world usage. |
12 |
>> |
13 |
>> Not true. You can eliminate many non-essential portions of a |
14 |
>> compiled program, via use flag and the freedom you get to select |
15 |
>> software, as opposed to other distros. Smaller executables are usually |
16 |
>> always faster. |
17 |
> |
18 |
[...] |
19 |
> But that wasn't what I was talking about, and AFAICT that's not |
20 |
> what reviewers are talking about when they talk about adjusting compiler |
21 |
> flags to optimize performance. They seem to be talking about building for |
22 |
> Athlon instead of P4 (or vice-versa). |
23 |
> Perhaps I've always completely misunderstood the articles I've |
24 |
> read, and they were indeed talking about USE flags that control options |
25 |
> passed to "configure" and not about things like gcc's -march and -O |
26 |
> options. |
27 |
|
28 |
USe flags can be used for anything. Note that ebuilds are |
29 |
ultimately bash scripts. And USE flags are just that: f-l-a-g-s. |
30 |
Flags are used in a script to control things that can be run -or |
31 |
not- depending on a condition, things like "if in amd64 do this, |
32 |
if not, if hardened do that, if yes and hardened to anything else"... |
33 |
That includes things like activating concrete portions of |
34 |
arch dependent code or a patch, things like passing a simple option |
35 |
to add or remove a dependency, and any other things that you could |
36 |
do manually on a shell. |
37 |
|
38 |
It can of course be used as well to adjust CFLAGS and other things |
39 |
depending on the architecture or whatever condition you want. And |
40 |
even more, they can be used to filter CFLAGS that the developers know |
41 |
that are harmful (and that's a big part of the portage stability, |
42 |
because in the past users used to shot themselves on the feet by |
43 |
adding a 20 lines long CFLAGS declaration into their make.conf's. |
44 |
|
45 |
Note that reviewers usually test a thing for 2 days, and then they |
46 |
think they are qualified to talk about whatever thing. Some times, |
47 |
these reviews are useless for this reason. They only scratch the |
48 |
surface, giving a bad impression or just a poor one. |
49 |
|
50 |
Note that I said "some times", though I think that "most times" |
51 |
is potentially a more correct qualifier. |
52 |
|
53 |
-- |
54 |
Jesús Guerrero |