Scenario: - System running FreeBSD 13.0-CURRENT #0 r355526M: Sun Dec 8 16:30:56 CET 2019 - /usr/src updated to r356034 (head) - Executing 'make -j8 buildworld buildkernel KERNCONF=...' Result: - In most cases, the make fails with >>> stage 3.1: building everything ... make[4]: make[4]: don't know how to make opt_global.h. Stop Notes: - Omitting -j8 gets the build to succeed It seems that the parallel makes repeatedly recreate opt_global.h, thereby thrashing each other's copies.
*** Bug 249957 has been marked as a duplicate of this bug. ***
The saga goes on... first, I realized that I reported this twice, so I closed the second bug #249957 as a duplicate. One crucial point seems to be that both the source and object directories are on NFS. When switching the object directory to a local one, the issue does not occur. Still with an NFS-mounted object dir the issue also does not seem to occur when rebuilding an already built kernel (even if the cleaning step is run, as it normally is).
I also tried adding a sleep of a few seconds after the config which generates opt_global.h, to no avail. It should be noted that all the machines are running ntp, with a few milliseconds of time difference between them. Also, if I rerun the make buildkernel with -DNO_CLEAN, it always succeeds. -- Martin
I still have this on stable/14. -- Martin