I played around some and did notice that common.eblit filters -march and then substitutes it with -march=i686 at least for x86. I also saw bugreport #185404 and as far as I can tell from it for x86 there were no breakage except if -march was missing. The problem with this is that "man gcc" says that as -march implies -mtune and does not need to be set. I think (and have on the forums seen) a lot of people only is using march becouse of this. For me this manifested itself at least as a failing iconvdata/tst-table in glibc testsuit. It also results in an unoptimized version of glibc. We could if -mtune not set convert -march to -mtune before filtering -march. Or we could do the following change to common.eblit: --- /var/portage/sys-libs/glibc/files/eblits/common.eblit 2008-06-22 16:07:13.000000000 +0200 +++ common.eblit 2008-12-10 18:28:57.940730341 +0100 @@ -74,8 +74,7 @@ local t=${CTARGET_OPT:-${CTARGET}} t=${t%%-*} if ! echo "" | $(tc-getCC) ${CFLAGS} -E -dD - | grep -qs __${t}__ ; then - filter-flags '-march=*' - export CFLAGS="-march=${t} ${CFLAGS}" + [[ "${CFLAGS}" =~ 'march' ]] || filter-flags '-march=*' && export CFLAGS="-march=${t} ${CFLAGS}" fi ;; amd64)
your analysis is incorrect ... there is still breakage possible *** This bug has been marked as a duplicate of bug 199334 ***
