Joe Ciccone wrote:
As i said in a previous email to someone else about this earlier in the week. Only /tools is cross-compiled. The final system relies on config.guess. config.guess relies on uname.
I'll carry that other thread over to here since this subject line describes it better :) Despite your explanation, I'm still not clear how it all works. I can see that 'tricking' config.guess is important, e.g. for $exec_prefix paths like lib/gcc/i486-pc-linux-gnu/x.y.z/specs, which would not be correct just from passing a -march in CFLAGS. But is tricking config.guess sufficient to also take care that you don't generate any i686-specific opcodes?
There is a uname hack to force config.guess report i486-pc-linux-gnu by making uname report i486. http://ftp.jg555.com/lfs/uname_ix86.c or http://cross-lfs.org/~jciccone/uname_hack.tar.bz2 . The only difference between the 2 is the tarball has a Makefile in it.
I guess rather than hacking the uname one could alternatively hack the config.guess in (each subdir of) each package, right? Replace them with #!/bin/sh echo i486-pc-linux-gnu _______________________________________________ Clfs-support mailing list [email protected] http://lists.cross-lfs.org/cgi-bin/mailman/listinfo/clfs-support
