I'm trying to put together a script to natively build ppc64 the way LFS builds, that is replicate LFS chapter 5 to build in /tools, then follow CLFS chapters 9, 10 (no, I haven't decided CLFS is wrong, I've hit a major problem with gcc-4.2.0 and at the moment I can't run the cross-compiled binaries I built on a fast machine on my ppc64). Unfortunately, my low-end G5 is dead slow (about 195 minutes to build gcc-4.2.0 without bootstrapping or testing) and as expected a lot goes wrong when I write new scripts.
Googling for some of the errors I've seen found the occasional response indicating that some people have built natively, e.g. on x86_64. Has anybody got any insights to share, please (apart from "prepare and test this before you think you'll need it") ? The main things I've got so far: 1. If CLFS chapters 6,8 build a package, do so in the same place because it might be needed (e.g. zlib, file aren't in LFS /tools). 2. Drop test tools and perl (because the final build can follow the CLFS book for those, and testing this early is not particularly useful). 3. If CLFS chapter 6 builds a library in both sizes, do the same. The default on my CLFS host is -m64, so I don't need to specify BUILD64 in the other cases. 4. When building in ld at the end of binutils pass 1 (for ld-new) the LIB_PATH should be /tools/lib64 on multilib. 5. The sed for adjusting the toolchain is different - look at the output of gcc -dumpspecs to see where the loader(s) are referenced, in particular the LFS book's '^' is NOT wanted on the other architectures I've got, the places to change are preceded by '-dynamic-linker '. TIA ĸen -- das eine Mal als Tragödie, das andere Mal als Farce _______________________________________________ Clfs-support mailing list [email protected] http://lists.cross-lfs.org/cgi-bin/mailman/listinfo/clfs-support
