Learn more about these different git repos.
Other Git URLs
I have not been able to successfully build gcc in Fedora 23+ since June, 19th, all the builds fail because on 32-bit arm we trigger the 24 hour timeout. Back in June, the Fedora 22 as well as Fedora 23 builds took both around 18 hours on 32-bit arm, which is terrible (compared to 3-8 hours for other architectures), but still it was possible for the builds to succeed. The Fedora 22 gcc builds (same source) always succeed even now, still around 18 hours build time. The only major difference between F22 and F23 gcc is the default libstdc++ ABI (gcc4-compatible vs. the new one), but that really shouldn't result in many hours of difference and in the past it didn't result in that. So, the question is, are F22 and F23/F24 gcc builds (eclipse build target) going to approximately same hw, or significantly different? Are the kernels in between those approximately the same?
I've already tried to %undefine _hardened_build (which is desirable for gcc in any case, slowing down the compiler for the address randomization is undesirable), which didn't help, similarly am doing: %if 0%{?fedora} >= 23 %ifnarch %{arm} RUNTESTFLAGS="--target_board=unix/'{,-fstack-protector-strong}'" \ %endif %else RUNTESTFLAGS="--target_board=unix/'{,-fstack-protector-strong}'" \ %endif which should on arm F23+ significantly reduce the amount of gcc testing by not testing -fstack-protector-strong, only the default options, but it timed out anyway. So I'm afraid I'm out of ideas what it might be; note, the compiler is the same, and really should dominate the build time.
The builders in the eclipse channel are the same for any branches... so the f22 and f23 builds should use the same builders. ;(
So, it has to be something in userspace...
We could perhaps look at increasing the timeout until we can figure out what it is?
If you could bump it say to 2 days, at least for gcc, it would be nice. I could try to also add a date command to the beginning of %check, so we could (assuming all the eclipse builders are the same hw) compare f22 and f23 bootstrap time, if things slow down already there, or if it is the same speed, but only something got much slower during testing. Unfortunately, the testing is a single command, though the log files (available only if the build succeeds) contain time stamps (start/stop time). Though of course, as make check is done with -j<num_of_cpus>, it can't be exact.
Dennis bumped it to 2 days and I just pushed out that change...
So, can you try some builds and hopefully they complete and we can see from logs what might be causing it. :)
I kicked off a build on Sunday for f24 head and it's still running. It appears to be stuck though so might need some investigation of the build
http://koji.fedoraproject.org/koji/taskinfo?taskID=11325163
30214 kojibui+ 20 0 5440 1472 1368 R 99.7 0.0 1038:55 ./macrosub
[root@arm04-builder06 ~]# strace -p 30214 Process 30214 attached ^CProcess 30214 detached
seems that process is hanging there is no output when attaching to it via strace
I am guessing this got sorted out since we have had a number of gcc builds since then?
Please re-open if there's something else for us to do here.
Metadata Update from @jakub: - Issue set to the milestone: Fedora 23 Alpha
Login to comment on this ticket.