Re: [gentoo-server] linux-headers

[Kerin Millar]

Ben Munat wrote:
> Great. This sounds like a major disaster in the making. I'm glad I
> asked... but I wonder how many people running a 2.4 kernel have just
> hosed their systems?

Well yes, I suppose it would be pretty bad if you were to inadvertently 
install 2.6 headers and later build glibc against them while still 
continuing to run a 2.4 kernel :o

> One interesting thing: The version grid for linux-headers on
> packages.gentoo.org now shows version 2.4.22-r1 as the latest available
> version. The two 2.6 versions are hard masked:
>
> http://packages.gentoo.org/search/?sstring=linux-headers&sourceid=mozilla-search
>
>
> However, I just synced again and emerge still wants to upgrade my linux
> headers to 2.6.8.1-r2... even though the website lists this as hard masked!

That's easy - the website is wrong ;) 2.6.8.1-r2 is not masked at all 
(this change was made very recently in accordance with the 2005.0 
release). Basically, 2.6 kernels are now the default on some arches in 
the stock 2005 profile.

> As far as the make.profile stuff goes, do I have to change to the 2005.0
> profile? I was never trying to upgrade to a new profile; I just thought
> it odd that portage wanted to emerge what sounded like the wrong headers
> package for my system (and it turns out I was right). I just looked and
> I'm actually still pointing at the 2004.0 profile. Is this bad?

No, not really. Just bear in mind that one day the profile you're using 
will disappear from portage whereupon it will whine at you. Thus, a 
change will be necessitated.

> As for rebuilding packages after emerging linux-headers, I don't want to
> emerge linux-headers! Really nice one-liner there though.

Sure.

> So, I'm figuring I should mask >=linux-headers-2.5 like Sri suggested.
> But I suppose that still leaves me with the question as to what to do
> about my make.profile. Is it important to change that with each new
> release? (And if so, shouldn't portage remind me if it's so important?)
> What are people running for their make.profile?

No, it's not overly important in many cases, at least for x86 arch. 
Conversely, it's not advisable to remain on a deprecated profile for too 
long. If you poke around in the profile, you'll see that its task is 
mainly to establish such things as default USE flags, packages that 
constitute the "system" set, minimum acceptable package versions thereof 
, default virtuals (i.e. what does it choose if you emerge something 
that depends on virtual/kernel and no suitable ebuild has been emerged) 
etc. All in all, this defines a baseline policy for how gentoo is built 
and "evolved" and one would generally do well to follow the bottom line 
there I suspect. I still haven't shifted from the 2004.3 profile on my 
server but will do so shortly.

If you use the profile I suggested then you shouldn't have to mask 
anything, although I'm not certain that the sources you are using are 
restricted as you would wish (see 
/usr/portage/profiles/default-linux/x86/2005.0/2.4/packages). As I 
mentioned before, you could rectify that eventuality by either 
overriding the "packages" file in a PORTDIR_OVERLAY or by adding your 
own sub-directory under 2.4 (again, in a PORTDIR_OVERLAY) and 
restricting the scope of the sources (gss-sources, was it?) there. In th 
e latter case, you'd need to make sure that make.profile pointed to the 
sub-directory that you defined. Or you could just use package.mask as 
per usual. In any case, understanding the profile mechanism and how to 
customise it is a useful asset to have in terms of Gentoo management.

To explain futher, remember that profiles are cascaded. So, for example, 
the 2.4 profile isn't defined by just the three files you see there. 
Files from parent directories are also inherited/overridden as 
necessary. As with anything else in the portage tree, any aspect can be 
augmented or overriden in a local overlay so it makes for quite a 
flexible system.

HTH,

--Kerin Francis Millar
--
gentoo-server@g.o mailing list