Hello all, I recently downloaded the arm64 experimental stage3 and noticed the following issues: a) According to https://packages.gentoo.org/packages/sys-devel/binutils, binutils < 2.28 are considered unsupported. I tried binutils 2.28.1 with no issues. With binutils 2.29 and 2.29.1 glibc fails to compile. There are patches available for glibc-2.24 such as https://patchwork.ozlabs.org/patch/797471/, https://sourceware.org/bugzilla/show_bug.cgi?id=21725 that can also be used on 2.23-r4, the first one fixes compilation but the second one didn't fix the ld.so crash, there is another fix I found online for 2.24+ but I didn't test it and went for binutils 2.28.1. Could we please unmask it ? b) For the systemd profile the following packages needed to go to package.accept_keywords sys-libs/libseccomp-2.3.2 sys-apps/systemd-233-r4 sys-apps/gentoo-systemd-integration-7 I also put there linux-headers-4.9 (to match the running kernel -not from gentoo sources, it's a custom one-). sys-libs/libcap-ng-0.7.8 (I enabled the caps use flag globally) sys-apps/net-tools-1.60_p20161110235919 (the unmasked version needs this patch http://www.openfastpath.org/pipermail/openfastpath/2016-October/001322.html to work with linux-headers >= 4.8) Note: Gcc 5.4.0 didn't have issues and I also tried 6.4.0 with graphite + lto + pie (recompiled everything with some exceptions for lto, such as python that for some reason results x87 assembly code). c) Portage complains that there is no arch-specific configuration and that I should contact the arch maintainer. d) The 13.0 profile is multilib (inherits features/multilib) for some reason and since there is no multilib support on arm64 it doesn't make much sense to have multilib glibc (it results downloading a pre-built toolchain etc).
I think we should split some of this up to separate bugs, maybe arrange a tracker if we didn't have one yet, etc. The profile note about features/multilib sounds like most important to figure out to me
As an additional note, the experimental stage3 used to contain leftovers in /tmp as well, which people might not even notice, taking actual space in there, if a tmpfs gets mounted on top of it. So something else to verify once we have newer stages going.
I think everything else is fixed by now, except the features/multilib concern. Albeit I don't see any multilib glibc here, or I don't know how to look properly. But still - if we have decided to not do multilib...
commit 46a09ed6187912edd0aefff1962a8bd08e613782 Author: Anthony G. Basile <blueness@gentoo.org> Date: Fri Apr 19 23:33:39 2019 -0400 profiles: remove multilib from arm64 profiles