Thinking forward LFS-7.0

DJ Lucas dj at linuxfromscratch.org
Sun Mar 13 22:44:26 PDT 2011


On 03/14/2011 12:03 AM, Nathan Coulson wrote:
>
>
> On Sun, Mar 13, 2011 at 9:45 PM, DJ Lucas <dj at linuxfromscratch.org 
> <mailto:dj at linuxfromscratch.org>> wrote:
>
>     On 03/13/2011 11:39 PM, DJ Lucas wrote:
>     > Okay, so I was just
>     thinking...<replaceable><Deity></replaceable>
>     > help us! I figure we have at least 6 months, potentially a year
>     until
>     > the next major LFS release, and now seems like a pretty good time to
>     > explore some of the ideas that have been shelved for previous
>     releases,
>     > and even some new ones. Here is a quick list to see if there is any
>     > interest. I'll reply to my own post afterward to separate my own
>     > suggestions from the initial list.
>     >
>     > * Package Management - Always causes a good debate.
>     >
>     > * DESTDIR - Been mentioned several times and actually this is
>     not too
>     > disruptive (I did a POC about 3 years ago).
>
>     For me, these two go hand in hand. Package Management,
>     historically, has
>     always generated a fueled debate. There are many options here
>     including
>     conditional processing of the book's sources to allow for various
>     forms
>     of package management. I had previously suggested years ago that
>     we move
>     the default build instructions to include a very simple DESTDIR style
>     installation, with the final installation done manually from the
>     DESTDIR
>     target. This method lends itself well to almost all forms of package
>     management without the need of choosing a specific package
>     manager. This
>     method also gives us the option of pre-processing the book's
>     source code
>     for a specific package manager either by conditional logic (as I
>     understand the new docbook is capable of), or in a simpler form,
>     using a
>     Makefile target to pre-process the book using sed or other
>     standard unix
>     tools if the instructions are split into pre-install, build, install,
>     and post-install targets.
>
>
> no comment here, although I've personally preferred less instructions 
> to more where possible.  (DESTDIR is not something I use) .  I usually 
> throw away systems when it's time to upgrade though, or just install 
> newer libs over the old.
>
That is part of the problem that I see. We all have our own way of doing 
things. This is actually fine, but I can see the benefit of 
standardizing some items such as package upgrades in the way of a wiki 
entry. For instance, keeping a copy of s-ls, s-rm, s-cp, and s-install 
around for glibc updates (these would be statically linked copies of the 
s- tools) which hasn't been needed for years I'm told, and in fact I had 
proven not to long ago by doing an in-place upgrade of GLibC while Gnome 
was running. :-) I'm still weary of doing glibc upgrades like that on a 
live system, perhaps it is an unfounded fear, but I did it on a test 
system just for kicks and had absolutely no issues. I did do a reboot 
immediately afterward.
>
>     > * LSB Compliance - For LFS we are nearly there anyway.
>     >
>     > * Dynamic boot script - No more static list of links, this kind
>     of ties
>     > into LSB Bootscripts, but there are other options.
>     >
>     Again, these two go hand in hand for me. In the current
>     lfs-bootscripts
>     tarball, I've been working on and using exclusively the contrib/lsb-v3
>     boot scripts for over 3 years now. They are an extension of something
>     that Nathan Coulson and Alexander Patrakov had started on. These are
>     completely lsb-v4 compliant as well and are IMO a huge improvement
>     over
>     the current boot scripts. I've been using Dan Nicholson's initd-tools
>     package to provide the install_initd and remove_initd programs. Aside
>     from the fact that there is no longer any need to maintain a list of
>     symlinks for startup order, they add a lot of niceties, including
>     boot-logging and conditional startup for trouble-shooting.
>
>
>
> ah, I miss the bootscript days.  I'll have to take a look, and find 
> out what I have been missing out on
>

Yes, please! Another set of eyes and additional brain power is always 
welcome! You should still have commit privs so feel free to help 
yourself. The current 'stable' boot scripts are the remnants after we 
ripped out the i18n additions. Though they are stable, they still 
contain a lot of cruft such as boot_mesg which is largely unneeded. I 
wound up doing a complete rewrite of rc and a single conditional source 
of the configuration files in lsb-v3. IOW, at the cost of possibly 
faster conditional logic, they only get sourced by the script if running 
outside of rc in the lsb-v3 directory. I honestly don't remember what 
the 'stty sane' issue was caused by, but I have never been able to 
reproduce it since and saw no reason to source the files in every single 
sub-shell. BTW, Dan's initd-tools has moved. He is currently hosting 
them in his home directory on freedesktop.org.

>     > * Multi-lib - Shunned previously, but there are many projects that
>     > expect this environment.
>
>     I just kinda threw that in there. A couple of projects that I use
>     beyond
>     BLFS expect that environment now for 64bit builds, specifically AOSP,
>     Wine, and VirtualBox, all of which forcefully exclude the official LFS
>     as my daily driver now. I've been using the work of the CLFS devs
>     for my
>     own daily driver with some heavy modifications to match LFS proper as
>     much as possible.
>
>
> When I designed my own multilib system,  I had a few goals
> - 32bit was not a essential part of my system.  Only reason I have it 
> is for wine, basically...
> - I wanted /lib 64bit.   Therefore, I used /lib/32 for 32bit libs 
> (although /lib32 would probably be a better book choice).  [always 
> bugged me that /lib is normally the 32bit directory, especially since 
> it is the default installation choice]
> - If possible, I would have liked to see multilib avaliable "after" 
> building lfs, but this is probably not feasable.
> - I only install 32bit versions of glibc, zlib, ncurses, utillinux as 
> they're the only ones I actually needed for my 32bit uses.  May be 
> worth reviewing who would be using 32bit, and what applications it 
> would be supporting.  Multilib in the book would look nicer, if only 
> glibc/gcc/binutils had multilib tweaks.  (Enough to compile 32bit 
> software)
>
Yes, unfortunately, the tool chain packages are not enough to do a real 
multi-lib system. For my own purposes, I need a working 32bit tool 
chain, 32bit X Libraries and server, and 32bit libs up through java. 
Most of this is for Android development. For LFS's purposes, the tool 
chain is enough however and pretty much what CLFS covers (this includes 
PPL, Cloog-PPL, GMP, MPFR, MPC. ZLib, Binutils, GLibC, and GCC). PERL is 
a real kicker that brings in more 32bit libs than the tool chain 
requirements. I haven't tried to build a multi-lib without it yet to see 
if it is needed by my own tools, but I suspect that a lot uses it. If a 
user is dabbling in multi-lib, then they should be able to figure out 
exactly what they need by using the existing instructions. This paired 
with DESTDIR style installations provides some protection to the user as 
well. BTW Nathan, I did use your page to figure out what exactly needed 
to be changed in the GCC drivers as opposed to sorting through the CLFS 
patch. I made changes, but that write-up was quite helpful. Thanks.

>     > * EGlibC - Seems like Debian and friends are moving to EGlibc,
>     gives us
>     > a couple of niceties but nothing major, not sure what other
>     distros are
>     > doing, but I've seen a lot of mentions of it recently. The work is
>     > already done by the way, our fellow devs at CLFS already have it
>     covered
>     > for us.
>
>     Same thing, I've seen lots of mentions of it so I'm throwing out the
>     feelers. CLFS has already done this and I've used both. I have no real
>     opinion either way yet. I did not have anything more than minor issues
>     making GLibC proper conform to my expectations (/lib -> lib64/,
>     /usr/lib
>     -> /usr/lib64, /lib32, and /usr/lib32).
>
>     > * Modular *.d/ directories - I'm pretty sure this is already
>     covered in
>     > another thread, but it should be done by default where possible.
>     >
>
>     As mentioned elsewhere recently, this gives a lot of options, and in
>     fact are required by a few packages in BLFS. I see absolutely no
>     reason
>     to omit this as the default using the instructions currently in
>     BLFS as
>     a guideline for strict conformance.
>
>
> using .d folders has made scripted builds much nicer, and It seems 
> tidier to me.

Yes, the end product of modular parts simply looks much cleaner.

-- DJ Lucas


-- 
This message has been scanned for viruses and
dangerous content, and is believed to be clean.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxfromscratch.org/pipermail/lfs-dev/attachments/20110314/763a0887/attachment.html>


More information about the lfs-dev mailing list