Hey All,
Just trying to sit and think through some aspects of the upcoming cross-lfs book, and I wanted to throw some ideas your way.
Who knows exactly what this change of build method will bring exactly for LFS, but my prediction is that most LFS users that build the cross-lfs book will still be building LFS on the same machine they intend to run it on, which of course means same arch. The purpose of Cross-LFS will still be realized, in that it takes independence from the host system to a whole new level, however I think more people will be taking the chroot route and quite possibly using a livecd that's compatible with their target arch as a host.
Even so, the intention so far has been to offer the ability to begin the build on one machine/arch and finish it on a different machine/arch. Therefore, the question comes up, how do we get the cross-compiled tools over to the target machine and set up the user to finish the remaining book? To date, I think the suggested method has been to tar up the tools and kernel, and copy them over, *somehow*. I think it was proposed to have various hints written that could link to the book showing possible ways of accomplishing this.
This just feels wrong to me on several levels:
1) We don't know what state the target machine will be in - does it even have an OS? - are we really going to require that there be one? For, if we turn the cross-compiled tools into a tarball, I don't see how we can avoid that requirement.
2) If the purpose of just tarring up the temp-system and kernel was to keep the LFS book simple, it seems to me that it will have the opposite effect. There's too many variables, and it will be more than just a matter of 'dropping it into place'.
3) We're giving them an option in the book but not really seeing it through. We say, in effect, "you need to get to this point to finish the book, but to do that, you have to get there yourself - see you there!"
To have their system usable and able to build the remaining LFS - they'll need to not just switch over to that machine, but actually boot it with a Linux kernel, which also implies a bootloader and perhaps some simplistic bootscripts. If we're asking that much of them at that point already, why not just finish the job? With a couple more simple packages and a few instructions on how to bring it all together, we could give to them a complete solution in the form of a minimal bootable cd. By far, the majority of machines out there can boot from a cd, and the methods to produce one are already documented - we just have to bring them in.
Anyway, I realize there are likely to be different opinions on this one, but I wanted to put this forward for comments. Personally, it would make me feel much more comfortable with where we're leading the reader.
-- Jeremy Huntwork -- http://linuxfromscratch.org/mailman/listinfo/lfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
