Fedora Account System
Red Hat Associate
Red Hat Customer
binutils-2.47.50-4.fc46 seems to have regressed on aarch64 when compiling rustc. The first stage goes okay, but when we build stage 2 with LLVM profiling annotations, the compiler hits SIGILL even for a simple version check. error: process didn't exit successfully: `/builddir/build/BUILD/rust-1.98.0-build/rustc-1.98.0-src/build/host/stage2/bin/rustc -vV` (signal: 4, SIGILL: illegal instruction) https://koji.fedoraproject.org/koji/taskinfo?taskID=149429290 Reproducible: Always Steps to Reproduce: 1. (on aarch64) rpmbuild rust.srpm Additional Information: I tried building in a side-tag with previous binutils, and they passed: 2.47-1.fc45: https://koji.fedoraproject.org/koji/taskinfo?taskID=149428939 2.47.50-2.fc46: https://koji.fedoraproject.org/koji/taskinfo?taskID=149428542 I took a look upstream and found commit ffd2a725: https://git.sr.ht/~sourceware/binutils-gdb/commit/ffd2a7258c46d74d960502b3b3f77aaa44b8118b > AARCH64 continues an ARM tradition of updating stubs late. See > coff_arm_link_output_has_begun. Commit e189bfd9b492 broke AARCH64, This ffd2a725 fix came after the snapshot used by 2.47.50-4, and e189bfd9 came between our 2.47.50-2 and 2.47.50-4. I can't be sure that's *my* problem, but it would be a rather large coincidence otherwise... :)
FYI, I asked releng to untag that build since we saw more regressions in other rust-* packages. https://forge.fedoraproject.org/releng/tickets/issues/13500
This should be fixed in binutils-2.47.50-6.fc46 (which is currently building and should be in the Bodhi system soon).
Right - the new build has reached Stable and should be available for testing.
It worked here, thanks! https://koji.fedoraproject.org/koji/taskinfo?taskID=149793765