Hi Kurt, On Tue, Jul 26, 2016 at 06:20:08PM +0200, Kurt Roeckx wrote: > On Tue, Jul 26, 2016 at 06:25:16AM +0200, Helmut Grohne wrote: > > Hi Kurt, > > > > On Mon, Jul 25, 2016 at 08:36:41PM +0200, Kurt Roeckx wrote: > > > This all looks fine. I'm just wondering why gcc-multilib is a > > > problem in your case since it was already any-amd64. I'm also > > > wondering if any other arches need it, like x32 and ppc64el. > > > > Well, kfreebsd-amd64 matches any-amd64 and I'm cross building for that. > > > > As far as I understand, Debian's ppc64el doesn't do -m32. > > > > x32 does understand -m32. > > > > Given that the most recent libtool build succeeded on both amd64 and x32 > > and that both builds skipped 22 tests, I guess that -m32 is not needed > > on x32 (and maybe no longer needed on amd64 either). > > Are you confusing libtool and elfutils?
Yes, I am. So let's look at elfutils again. For x32, the failing test is run-backtrace-native-biarch.sh, so this looks a lot like a multilib issue. On the other hand, any-amd64 matches[1] x32, so gcc-multilib is being installed and we can see that from the build log. So without further details, it's hard to say why it fails in the presence of a -m32 capable compiler. Also verbose build logs would be awesome. For ppc64el, we see: | configure: WARNING: not running biarch tests, gcc -m32 does not work Indeed, a gcc-multilib package might help here. gcc doesn't build it for ppc64el and I'd very much like to not add any further multilib packages. Instead, we should move to fully triplet prefixed toolchains. Rather than running "gcc -m32", run "powerpcle-linux-gnu-gcc". Besides saving a lot of trouble, the latter is not ambiguous. Helmut [1] dpkg-architecture -ax32 -iany-amd64
