Slashdot Mirror
Beyond Linux From Scratch 1.0 Released
An anonymous reader writes "DistroWatch reports about the 1.0 release of Beyond Linux From Scratch (BLFS), a subproject of Linux From Scratch: 'The BLFS Development team is proud to announce the release of BLFS 1.0. With this release, you can take your LFS 4.1 base beyond a development system. It can be a desktop, a firewall, a multimedia player/editor, an Apache web server or all of the above. You install only what you need. Your Distro. Your Rules. Enjoy.'" Choose the closest mirror...
If you want to learn about how and why a linux distribution is put together, LFS is the way.
If I seem short sighted, it is because I stand on the shoulders of midgets
Linux From Scratch [linuxfromscratch.org] is not a disto, but a set of instructions for building a GNU/Linux system from little more than an existing distro. The Beyond Linux From Scratch [linuxfromscratch.org] project builds on this tradition, providing instructions for installing a number of other packages.
:-)
Having said that, I would recommend that anyone serious about furthering their knowledge of Linux, and at least 750MB of free hard disk space should give it a go. I got into Linux back in '93, and knew quite a lot. Then I stumbled across Linux From Scratch (LFS) - I didn't realise how much I didn't know until finishing my first complete build.
I now run an number of LFS-based systems at home and at work, and have never looked back.
BTW, I am typing this on my Gentoo [gentoo.org], 1.4_rc4 build.
Any fool can talk, but it takes a wise man to listen.
It's actually very different from Gentoo. There's
/bin & /sbin to be
a whole host of choices you can make when building
a package from source. Most of these are made for
you in the ports setup.
For instance, I may prefer *not* to have GNU
gettext linked into everything. Or perhaps I want
every installed program in
statically linked so I can recover if my
libraries somehow get hosed.
Don't get me wrong, Gentoo is good for eeking
performance out of your dist. But you don't
get exposed to nearly as many design decisions
as you do when building from scratch. I recommend
that anyone serious about *nix admin try it at
least once.
On your last point, I wholeheartedly agree.
When you upgrade/install a package from source
on a production system, you're basically trusting
that the creator of the package you're installing
has thouroughly tested it. This usually is not
the case. When you install a package from your
vendor, they've at least done some basic testing.
People with actual lives use Mandrake or Red Hat. People with l33t tendancies but an aversion to work use Gentoo. People who want to learn stuff use LFS.
Anyway, I don't understand how anyone has trouble installing packages from source. You just need to partition them appropriately. I wish distributions used a similar method so they could keep parallel versions of programs and libraries installed.
Take openssh, for example:
My typical build of openssh requires: zlib, openssl, and tcp_wrappers.
I use a structure I call /mfs ("My File System"). I could use /usr/local for the same thing, but a customized directory will prevent collisions in /usr/local (the usual default prefix for configure scripts).
Under /mfs I have directories "dist" (the tarballs), "src" (where I untar and build each package), and "pkg" where I install packages.
So I start with zlib: "./configure --prefix=/mfs/pkg/zlib/1.1.4 ; make ; make test ; make install".
I repeat with tcp_wrappers, openssl, and finally openssh. So now I have the most current versions:
It should go without saying that I configure openssh to use the zlib, openssl, and wrappers libraries under /mfs, rather than the default system libraries. To the anal purists out there, I don't do this with all libraries (such as glibc), as it would drive me insane -- but it could theoretically be done.
Here's where it gets elegant (or convoluted, depending on your tolerance for complexity). Under each package's directory, I use a symlink from the version I wish to use on a regular basis to "std".
So, under /mfs/pkg/openssh, I may have directories 3.6.1p1, 3.0.1p1, and 2.9.9p1. Let's say that I want to use the latest sshd, so I run "cd /mfs/pkg/openssh ; ln -s 3.6.1p1 std". I then cd back up to /mfs. I then issue "lndir pkg/openssh/std". Actually, I have a script in /mfs that automatically removes all existing links under /mfs (avoiding the "pkg" dir, for obvious reasons) then re-linking every package with a "std" link. (Note I don't use a "std" link -- and thus don't lndir -- for libraries without runable binaries). If you don't know how lndir works, check it out (it's from the XFree86 distribution, though that might not be where it originated.).
After running lndir (or my script), I now have /mfs/{bin,etc,var,lib,sbin}. I point my sshd startup script to always use /mfs/sbin/sshd, which is actually a link to /mfs/pkg/openssh/std/sbin/sshd. Since std is a symlink to the version I wish to use, I can change to a newer or older version by simply stopping sshd, changing the "std" link to point to another version, and restarting sshd.
The beauty is that still have other versions available to me. Say the scp provided in 3.6.1p1 has an irritating bug (not the case, but just imagine). If my normal PATH has /mfs/bin, I'll get scp v3.6.1p1, but I can fall back to running /mfs/pkg/openssh/3.0.1p1/bin/scp.
This technique is especially valuable in multi-user systems, where libraries and applications of different versions (think compilers, for example) have dedicated users who aren't ready to upgrade.
Sorry for the really long post, but this idea works extremely well, IMHO, and I'm surprised that no distros use a similar technique for maintaining parallel versions. If RPM used this technique, you'd never run into the case where a new app needs a newer library version, but upgrading that library isn't possible since other major applications require that specific version (such as KDE needing a specific version of libpng and libcrypt).
Method of processing duck feet