LFS and 64-bit hosts

Ken Moffat zarniwhoop at ntlworld.com
Mon Oct 1 13:02:38 PDT 2007


On Mon, Oct 01, 2007 at 11:20:02AM -0700, Dan Nicholson wrote:
> On 9/30/07, Chris Staub <chris at beaker67.com> wrote:
> > There are an increasing number of users who report errors building LFS,
> > and then say they are attempting it on a 64-bit host system. It really
> > needs to be stated on the LFS website that LFS assumes a 32-bit system,
> > and point users with 64-bit systems either to the 64-bit LFS
> > book/livecd, or CLFS.
> 
> Very good point. Seems we're getting support requests on this every
> other day now. I've never been very fond of the ambiguousness in LFS
> about what hardware is supported. The only thing that's fully
> supported right now is x86. Full stop. If people in the know want to
> make it work on PPC or x86_64, that's great, but we shouldn't give
> that impression to people who are just following along.
> 
 My ears are burning.  The problem is that we have an increasingly
wide range of users but, for whatever reason, many of them are
reluctant to consider clfs.  I suspect that in part it's because
clfs is (usually) much closer to the bleeding edge, and more likely
to have issues on at least some of its platforms.  I've got fed up
saying "use clfs" where *for experienced and knowledgeable builders*
the LFS instructions can be adequately adapted.  Like most people,
I don't always consider who will read my remarks.  So, I agree that
until the jh branch is merged, the book should be telling people
that only x86 is supported.  But I reserve the right to attempt to
offer support on the lists.

> First, where should this be placed for LFS? I'd say in the book (can't
> decide where) it should be really clear what is and is not supported.
> Unfortunately, LFS-6.3 is in the wild already. We can change trunk to
> say x86 only until the jh branch is merged. For everything else, we
> should point to CLFS. If this info is already in the book, it needs to
> be made a lot clearer.
> 

 We have the conflict that the book still tries to cover other
architectures (e.g. the name of the dynamic linker in the note in
"adjusting the toolchain").  I would be sorry to see that, and the
note in "Toolchain Technical Notes", go - particularly when I hope
that the book will soon support at least x86_64 in some way.

 Maybe a big note at the start of "Host System Requirements" ?  But,
it needs to be somewhere that most people will read, and we know
that enough people skip that section.  I'm tempted to suggest
putting a note at the start of packages.html, but stylistically it
is the wrong place.

> Second, the LFS website should say this. Probably right up front:
> 
> http://www.linuxfromscratch.org/lfs/
> 

 Agreed.

> Third, I think the LiveCD currently says "linux64 - A 64-bit kernel
> for use with CLFS" in the options. While that seems plenty clear to
> me, I think it needs to be even clearer that building LFS does not
> work from a 64 bit kernel.
> 
 It should work, subject to certain conditions! ;-)  No, not building
from 'linux64' on the CD - we probably need a warning somewhere -
within the note(s) to say 'only x86' would be fine (i.e. something
about 'uname -m').  But building on a 64-bit kernel is
straightforward - use 'linux32' to acquire a 32-bit personality, and
make sure you have an i686 userspace.  But that is not something we
should mention in the book.

> IMO, of course.
> 
> --
> Dan

 We are still going to get people using 6.2 and 6.3 who think they
can do x86_64 without doing anything different.  Maybe we should
also put a message on the 'Read the LFS Book Online' page on the
website ?

ĸen
-- 
das eine Mal als Tragödie, das andere Mal als Farce



More information about the lfs-dev mailing list