[Gllug] [OT] Appreciation

[Nix]

On 16 Dec 2006, Mike Brodbelt uttered the following:

> Juergen Schinker wrote:
>> i pointed that out
>>> Actually, I'm extremely interested in this debate for practical
>>> reasons. Debian seems to be heavily defeating Gentoo, so far.
>>>
>> Naa your so wrong...Gentoo is heavily defeating Debian, it's
>> concept is better, but try it out...
>
> This has been a longish thread already, and AFAICS you're the only one
> defending Gentoo. That's hardly "heavy defeat" for Debian, regardless of
> the fact that I still haven't seen any credible arguments for Gentoo's
> supremacy.

I like the idea of Gentoo, but mostly because you can be *certain* that
the sources on your disk and the executing binaries match, and you always
know what the compilation environment was.

This can be useful for teasing out and tracking down bugs.

I think that, on the whole, gentoo has increased the quality of software
out there. But it's a bleeding-edge developers' distro at heart.

> 1/ Gentoo users aren't dependent on how developers built their packages
>
> Debian users tend to be happy with the way software has been packaged,

The advantage of knowing *exactly* how your software has been built is
greatest if it's something with a lot of config options. XEmacs in Debian
is a classic example of what not to do. (XEmacs in Gentoo is even worse,
however: it was entirely unmaintained for years.)

> and when they aren't, there's always "apt-get source foo; vi foo/rules;
> cd foo && dpkg-buildpackage". And you only have to compile the packages
> where you need something different, not the *entire* *system*.

Indeed.

>                                         And AFAICT, there's no such
> beast as "Gentoo stable"

Exactly. Claiming that Debian is inferior because it provides an option
(long-term stability) that Gentoo does *not* is frankly bizarre.

> 4/ Better performance
>
> For most apps, this is irrelevant - an optimised ls is hardly a win. In
> some few specific cases, you may indeed be better off

Even there, if the difference between arches is really extreme, glibc
will have a hwdep tag matching it, so the distributors will have built
appropriate versions of suitable shared libraries and put them in the
right places. (See e.g. /usr/lib/v8/libcrypto.so.* on UltraSPARC: v7
doesn't have integer multiply and divide instructions, and the speed
difference is quite substantial.)

Note that in general Intel-compatible CPU manufacturers try to
concentrate on improving the performance of binaries that are *already
out there*. New instructions do turn up, but their effect on performance
is generally marginal (CMOV is a rare exception), and often 99% of the
improvement can be had by rebuilding a very few programs or shared
libraries.

(When Intel forgot this, with the P4, everyone laughed at them. The P4
could be reasonably nippy, *if* you rebuilt everything. Nobody did. One
could say that the ill-fated IA64 was another example of this error.)

> The best case I can see for using Gentoo is for learning about the build
> tools and how the system hangs together. For that though Linux from
> Scratch would be a better choice, on a test system (or an emulated VM on
> your Debian box:-) ).

Yes :)

-- 
`He accused the FSF of being "something of a hypocrit", which
 shows that he neither understands hypocrisy nor can spell.'
   --- jimmybgood
-------------- next part --------------
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug