Seen for months, specifically tested on: FreeBSD freebsd 15.0-CURRENT FreeBSD 15.0-CURRENT #0 main-n270474-d2f1f71ec8c6: Thu May 30 03:12:57 UTC 2024 root@releng3.nyi.freebsd.org:/usr/obj/usr/src/amd64.amd64/sys/GENERIC amd6 The WITHOUT_CLANG build option enable these build options: WITHOUT_CLANG_FULL WITHOUT_LLVM_COV WITHOUT_LLVM_TARGET_AARCH64 WITHOUT_LLVM_TARGET_ALL WITHOUT_LLVM_TARGET_ARM WITHOUT_LLVM_TARGET_POWERPC WITHOUT_LLVM_TARGET_RISCV On 16 thread and 48 thread builds on different systems, two cleanups (bw/iw and w/iw in build_option_survey terms)fail on their removals without an error message: rm -f /usr/obj//usr/src/tools/tools/build_option_survey/_.mnt/usr/share/man/man3/cap_rights_contains.3 /usr/obj//usr/src/tools/tools/build_option_survey/_.mnt/usr/share/man/man3/cap_rights_contains.3.gz; install -l h -o root -g wheel -m 444 /usr/obj//usr/src/tools/tools/build_option_survey/_.mnt/usr/share/man/man3/cap_rights_init.3.gz /usr/obj//usr/src/tools/tools/build_option_survey/_.mnt/usr/share/man/man3/cap_rights_contains.3.gz make[4]: stopped in /usr/src/lib In previous cases rm failures were caused by a different thread cleaning up a parent directory. I will post the results of a single-threaded build as soon as it completes. My theory is that it will not exhibit the failure.
8h15m57.78s later on an Epyc 7402P at 2.8Ghz... chmod 444 freebsd.submit.cf -------------------------------------------------------------- >>> World build completed on Fri Jun 7 20:20:14 UTC 2024 >>> World built in 29757 seconds, ncpu: 48 -------------------------------------------------------------- The build completed. I sense a race condition.
Is there an error message earlier in the log?
This might have been attributed to building 15-CURRENT on 14.0R and appears to be working with combinations of 15 on 15, and 15 on 14.1R. I will close this for now and keep testing.