lanas wrote: > Obviously I haven't read the CLFS book from A to Z. Do you mean by > that, that it'll go natively because of the 'chroot or boot' choice, > even though there was an extra step (compared to plain LFS) at building > and using a cross-compile toolchain in the first place ?
Yes. The final system is built natively. Unless you use the sysroot method (see below). > But it seems this has a limit, at the end of chapter 5 where the > question "chroot or boot" pops up. Let's say I do indeed use CLFS to > cross-compile a x86_64 system and that I use a 32-bit host to do so. > Doesn't that choice makes it clear that I'll have to boot the target to > continue building a full system ? Yes. > Then, does this imply that a full system on a ARM processor such as the > Linksys NSLU2 can only be fully built natively on that very slow > processor ? Yes. > Isn't it possible to build a full system by only using > cross-compilation ? Yes, via the sysroot method. CLFS has not yet released a stable version of this method, but I believe they've got it mostly working. Sysroot has a great advantage in that you can, as you say, use a super-fast system to fully compile a system for your super-slow machine, like build the entire system for your ARM on a modern x86 or x86_64. The main drawback, if you're a purist, is that none of the configure scripts are run natively. This means that instead of getting true results of a test, configure has to guess at what it thinks your hardware can do. Sometimes it guesses wrong. Generally speaking, with today's methods, you'll always get a truer result when using 'cmmi' natively than when cross-compiling. > Why would the actual system be important if nothing linked to a running > kernel is being built ? I understand that inside BLFS, building a X > server is recommended to be done on a live system (or at least it used > to be recommended a few years back). But isnt't that an exception ? > Nothing else warrants using an actual live system to produce a full > built, isn't it ? Or am I missing something ? See above. It has to do with the assumptions configure makes about the final system. When running natively, or at least, on the same general architecture as the final system, configure usually arrives at reliable results. > I'd believe that in this case, a Fedora Core 6 x86_64 system has all > the tools required. Verifying is much better than believing, though :-) Yep, so long as you grab the dev tools. > Thanks for the URL and for mentionning DIY-Linux. I wasn't aware of > either. No problem. DIY is a good resource, created by a former LFSer. -- JH -- http://linuxfromscratch.org/mailman/listinfo/lfs-support FAQ: http://www.linuxfromscratch.org/lfs/faq.html Unsubscribe: See the above information page
