[gentoo-dev] Re: How a proper server profile should look like

[Duncan <1i5t5.duncan@cox.net>]

Rich Freeman posted on Tue, 22 Jan 2013 07:12:06 -0500 as excerpted:

> Should we therefore list all the flags on the system and which ones are
> enabled and disabled?
> 
> I guess we could, but it is a REALLY long list.
> 
> In practice I find that the way I tend to use USE flags is that I just
> ignore them until something unexpected happens, and then change them.

The one thing I wish the handbook had taught, way back when I started 
(and I read the handbook well enough that even before I had a system up 
and running on gentoo... 2004.0/amd64 wasn't quite ready for NPTL and I 
blocked on it, but 2004.1 worked... I was helping others who apparently 
had /not/ read it so well!  this wasn't there and AFAIK still isn't), 
was...

>>>>>

"If in doubt, leave it out.  Remember, because gentoo is build-from-
source, every package installed has a much higher cost in terms of 
continuing upgrades over time, than on a binary distro.  If you aren't 
sure you're going to use it, or will only use it maybe a couple times a 
year, strongly consider omitting it, thus avoiding the upgrade cost.  You 
can always install it later if you find you REALLY need it.

"That applies to both packages and USE flags (which often bring in extra 
packages) on your system."

<<<<<

One of the first things I realized out of the gate was that keeping both 
gnome and kde installed wasn't going to be practical for me, and I 
preferred the better configurability of kde so I quickly dropped gnome.

But over the years my system has gotten progressively leaner as I trimmed 
this and that, one thing at a time, because there really IS a continuing 
maintenance cost to every single package installed, ESPECIALLY on the 
~arch systems I run where the package churn is much higher, even MORE so 
for those (like me) that like to run stuff like kde prereleases from the 
overlays.  KDE for example has two feature releases a year and updates 
every month, basically 12 releases a year.  For those running the pre-
releases, it's 16, as for the couple months before a feature release 
they're on a two-week update cycle.  For those running its pre-releases, 
KDE *BY* *ITSELF* is thus several hundred package upgrade builds 16 times 
a year (plus -rX bumps if any).

I've trimmed my kde to ~170 packages at last upgrade (and just trimmed a 
couple more after that, deciding with dolphin as my GUI fileman and 
firefox as my default browser I no longer needed konqueror or its addons, 
so I think I'm down to 168 per kde upgrade here, now).  With my six-core 
bulldozer and PORTAGE_TMPDIR in tmpfs, that's actually reasonable.


I wouldn't expect ordinary gentooers to go to the lengths I have to 
reduce system bloat while keeping functionality I actually use, as the 
system set I've negated is there fore a reason and USE=-* is discouraged 
for a reason -- it TAKES someone with quite some experience and knowledge 
to properly navigate those sorts of things.

But if anything, that's all the MORE reason there should be a minimal 
profile available, for those who want as lean an installation as 
possible.  The more stuff turned on the worse it gets, especially for USE 
flags on system set packages and the packages they in turn drag in, 
multiple levels down.

That's actually why I eventually killed my system set, too much 
(including xorg-server and kdelibs) was being pulled into it by the USE 
flags, and for safety reasons, portage puts much stronger parallel-emerge-
jobs limitations on @system and its deps, many packages of which are 
piddly little things that kept portage running alone at <1.00 load 
average on a six-core!

So the smaller the set of profile-enabled USE flags and the smaller the 
@system set, the better, and a minimal profile that people can add what 
they need to, would ideally be the recommended profile for most users.

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman