Planning for Cross-LFS/Multi-Architecture 7.x Release

Ryan Oliver ryan at
Sat Apr 23 18:19:34 PDT 2005

Greg Schafer wrote:
 > Jim Gifford wrote:
 >>For the 7.x release we are planning to release a cross compiled
 >>multi-architecture book. The build process will allow you to build for
 >>any architecture to any other architecture. So you will be able to build
 >>a PowerPC starter system on a Pentium 4 and vice versa.
 > Every time this topic comes up I get a bunch of folks contacting me 
 > for my thoughts. I've already recorded my views and they're archived 
 > Nothing much has changed since last December, except that I'm now
 > convinced more than ever that cross compiling is always going to be a
 > kludge-fest. If you guys want to turn LFS into a hack-a-thon, go for it.

It is no more hack-a-thon than plfs was, it uses the same hacks ;-)

In some ways it is a lot cleaner.

Initial toolchain doesn't get clobbered anymore, so that removes the
main fsck up for newbies when they stuff up a specs edit/forget to
patch. There is only one spec edit nowadays, and that is in ch6.

We dont keep a modified linker around to kludge in in ch6, though we
build a temporary binutils (could be shortened to just build libiberty,
bfd and ld) in fact it is possible we can remove this build by
specifying -L/lib -L/usr/lib to the ch6 gcc/binutils builds

We don't have any reliance on the host, so no need to worry about
selinux headers (we don't look at the host headers so this doesn't
trip up glibc), nor fixincludes (it will fix our target headers, not
the host systems) and nothing linked against the host c-libraries at
any point gets placed in /tools

The above were the last remaining issues to be addressed to finish plfs,
this work is the logical conclusion.

Dont forget plfs was not much more than a bastardised host=target
cross-build, I know it, you know it.

 > But IMHO you'll lose one of LFS's best attributes, and that is gently
 > introducing newbies to an easy-to-understand build recipe.
  >> IMHO a saner approach would be to turn the cross LFS stuff into a 
 > doc. There is a distinct lack of good, easy-to-digest cross compile
 > documentation out there.

Aint that the truth... hopefully we can provide something digestible...

 > Ryan obviously has plenty of expertise in cross
 > compiling and transferring that knowledge into an LFS doc would be a 
 > thing. Folks who want adventure could play with the cross stuff. 
Folks who
 > just want to be build good ol LFS could stick to the tried and true 
main doc.

I've thought about this too...
We can (if host=target and running kernel ver = target kernel ver) use
the cross-build method for soley producing the ch5 glibc and toolchain,
from that point you have a choice.
1) continue using the cross-toolchain for the remainder of ch5 or
2) use the newly built native toolchain and continue with lfs as
     it is now
Both toolchains produce the same result.

No doubt all of this will be debated over the coming months.

 > My 2 cents.

Thanks for your concern


More information about the lfs-dev mailing list