|Summary:||devel/ninja: default config of devel/llvm80 fails to build on bare metal ryzen 5 system|
|Product:||Ports & Packages||Reporter:||aryeh.friedman|
|Component:||Individual Port(s)||Assignee:||freebsd-ports-bugs (Nobody) <ports-bugs>|
|Status:||Closed Unable to Reproduce|
|Severity:||Affects Only Me||CC:||aryeh.friedman, brooks, mikael|
Description aryeh.friedman 2019-07-29 08:21:53 UTC
Using default settings for make (and/or portmaster) cmake goes into an infinite loop on task 1 (running the port configuration script). Sorry don't have a log showing where it went off the rails I just have a lot of iterations of step 0/1 This is a brand new ryzen 5 2600 machine with the kernel reporting 12 cores (6 real ones with 2 threads each). To repeat: [with no ports installed] svn co svn://svn.freebsd.org/ports/head /usr/ports cd /usr/ports/ports-mgmt/portmaster make install cd /usr/ports rehash portmaster -d x11/nvdia-driver ... lots of output ... devel/llvm infinite loop FreeBSD neomarx 12.0-RELEASE-p8 FreeBSD 12.0-RELEASE-p8 r350399 GENERIC amd64
Comment 1 aryeh.friedman 2019-07-29 08:23:04 UTC
(In reply to aryeh.friedman from comment #0) oops devel/llvm80 not devel/llvm fails
Comment 2 Mikael Urankar 2019-07-29 08:57:23 UTC
I had the same problem once: the clock on my system wasn't correct.
Comment 3 aryeh.friedman 2019-07-29 09:00:22 UTC
This might be ninja related since the build for graphics/jpeg-turbo went into a similar loop (but gave up after 100 tries)
Comment 4 aryeh.friedman 2019-07-29 09:02:08 UTC
(In reply to mikael.urankar from comment #2) Not in my case it is correct (set it with ntpdate before I started the port installs): root@neomarx:/usr/ports/distfiles # date Mon Jul 29 05:01:15 EDT 2019 root@neomarx:/usr/ports/distfiles # ntpdate pool.ntp.org 29 Jul 05:01:52 ntpdate: adjust time server 184.108.40.206 offset -0.297128 sec
Comment 5 aryeh.friedman 2019-07-29 09:19:55 UTC
(In reply to aryeh.friedman from comment #3) It is for sure related to ninja since I have now had 5 or 6 ports in the xfce4 chain die on the same type of loop with the only commonality between them is they use ninja as their build system... is a default of j 12 on make on a fast processor too fast for ninja?
Comment 6 Brooks Davis 2019-09-06 11:12:49 UTC
AFACT this isn't an LLVM issue, but rather a ninja issue.
Comment 7 Brooks Davis 2020-01-03 22:19:33 UTC
There may be a bug here, but I've seen zero other reports and I regularly build llvm with -j32. Closing this since there isn't anything we can work with.