The: https://pkg-status.freebsd.org/ampere2/data/main-arm64-default/p7c7045637357_s8fb17896120/logs/errors/ruby-3.3.9,1.log shows: . . . --- ../../.ext/aarch64-freebsd16/openssl.so --- linking shared-object openssl.so --- ruby33 --- --- ruby33 --- linking ruby33 --- main --- --- rdoc --- Generating RDoc documentation ld-elf.so.1: /usr/local/lib/libunwind.so.8: Undefined symbol "__aarch64_ldadd4_acq_rel" *** [rdoc] Error code 1 make: stopped making "all" in /wrkdirs/usr/ports/lang/ruby33/work/ruby-3.3.9 make: 1 error make: stopped making "all" in /wrkdirs/usr/ports/lang/ruby33/work/ruby-3.3.9 ===> Compilation failed unexpectedly. That has lead to 4760 skipped port-package builds in: https://pkg-status.freebsd.org/ampere2/build.html?mastername=main-arm64-default&build=p7c7045637357_s8fb17896120 (At the time I'm writing this, the time the overall build is still in progress.) ports tree at 7c7045637357 is from: 30 Oct 2025 00:59:23 UTC src tree at 8fb17896120 is from: 30 Oct 2025 01:04:16 UTC
For reference from the log file: ===> ruby-3.3.9,1 depends on shared library: libunwind.so - not found ===> Installing existing package /packages/All/libunwind-20240221_2.pkg [main-arm64-default-job-04] Installing libunwind-20240221_2... [main-arm64-default-job-04] Extracting libunwind-20240221_2: .......... done ===> ruby-3.3.9,1 depends on shared library: libunwind.so - found (/usr/local/lib/libunwind.so)
But it turns out that ruby was updated to 3.3.10 on 2025-Nov-01: author Yasuhiro Kimura <yasu@FreeBSD.org> 2025-10-23 13:45:52 +0000 committer Yasuhiro Kimura <yasu@FreeBSD.org> 2025-11-01 03:17:43 +0000 commit e246c13eac6f6e8bf6127716cf2fc07ef5fc7071 (patch) tree b59bfd0bb007edbc5a665808ecdc4ceb46013396 /lang/ruby33 parent 6241b965356fb18caef79dc64bbdb6f85563b88a (diff) lang/ruby33: Update to 3.3.10 So the status may have changed since mastername=main-arm64-default for build=p7c7045637357_s8fb17896120 was started. But it will be likely be some time before another main-arm64-default official build starts.
The following looks to explain what lead to the failure: Wed, 29 Oct 2025 . . . git: 51e8e8b0f369 - main - Merge commit e24f90190c77 from llvm git (by Brad Smith): Dimitry Andric . . . Thu, 30 Oct 2025 . . . git: 8fb17896120d - main - intr/powerpc: create openpic_class kobj Justin Hibbits where the jail build was based on: build=p7c7045637357_s8fb17896120 and 51e8e8b0f369 was: Merge commit e24f90190c77 from llvm git (by Brad Smith): [Driver] Enable outline atomics for FreeBSD/aarch64 (#156089) The compiler_rt helper functions have been built since 12.4, 13.1, 14 and anything newer. before it was reverted. That, in turn, indicates that 8fb17896120d was able to build based on the prior system's lack outline atomics being in use when the new one with outline atomics was being built. Other, later CI activity indicated that an outline atomics world could not rebuild itself, getting linker errors for __aarch64_* symbols in various places. Such is what lead to reverting 51e8e8b0f369 . Thus buildworld success vs. failure was dependent on the live-world's status for how atmics were being handled.
As of comment #3 the issue looks to have been explained and does not look like it will repeat if things stay as they are (the world not using outline atomics).
(In reply to Mark Millard from comment #3) Actually, I got to thinking about prior experiments and my build attempt with outline atomics on a main 16 with with inline atomics running failed in the same manor as the CI builds. But I do not know what builds of main might have been built from a running 14.* (inline atomics) then or now.