On Wed, Oct 19, 2022 at 09:41:20PM +0200, Janneke Nieuwenhuizen wrote: > Ludovic Courtès writes: > > Hey! > > > Janneke Nieuwenhuizen skribis: > > > > This is exciting news! Looking forward to having full-source > > bootstrapped AArch64… and it looks like there’s already activity on a > > ‘wip-’ branch. > > Yes, it is! The ARM bootstrap story is still a bit flakey, as > stage0-posix does not support ARM. So, ARM would need > %bootstrap-mescc-tools and %bootstrap-mes binary seeds...meh. > > However, aarch64-linux now bootstraps from 526 bytes all the way until > gcc-core-mesboot 2,95.3. Very nice! > > We're still stuck at building a full gcc+glibc combo; > glibc-mesboot-2.2.5 builds, but possibly not correctly; as the full > gcc-mesboot0 (2.95.3) build fails at configure time: gcc-core-mesboot0 + > glibc-mesboot0 > > ?: 0 [execle "./gencheck" # "./gencheck"] > ERROR: In procedure execle: Exec format error > > Not sure what to do here. We could somehow try to debug/bisect this. > We could try to use a newer glibc; glibc-2.2.5 happened during the > OABI/EABI switch and is heavily patched. Or, we could try to remove > glibc-2.2.5/gcc-2.95.3 altogether and aim for a direct tcc => gcc-4.6.4. > We need to go that way anyway for RISCV. I noticed that when offloading from my x86_64 box to my aarch64 machine target gets (%current-system) from the originating system, not from the system doing the building. I'm testing moving the '(let ((triplet ...' code down from the top of the definition to inside the arguments. I didn't see a difference with 'guix build -d' for x86_64 from x86_64, but the derivation did change with the planned offload to an aarch64 machine. I looked into gcc-2.95 more than I wanted to last week with GHC-4.08, would it help if we specified we wanted armv4 or armv4t or strongswan instead of "generic"? -- Efraim Flashner אפרים פלשנר GPG key = A28B F40C 3E55 1372 662D 14F7 41AA E7DC CA3D 8351 Confidentiality cannot be guaranteed on emails sent or received unencrypted