[llvm-dev] [Release-testers] [3.8 Release] RC3 has been tagged

Daniel Sanders via llvm-dev llvm-dev at lists.llvm.org
Wed Mar 2 16:52:14 PST 2016


> > > > clang+llvm-3.8.0-rc3-mipsel-linux-gnu.tar.xz
> > > >         My machine died during the check-all stage and disappeared from the
> > > network. I've fixed it and am currently running the last few bits of test-
> > > release.sh manually.
> > >
> > > Thanks, let me know how this turns out.
> > >
> > > So far, it sounds like there are no new issues in rc3, so I'm hoping
> > > we can make that the final version for 3.8.0 real soon.
> > >
> > > Cheers,
> > > Hans
> > >
> >
> > 'check-all' is still killing the machine but I think I know what the problem is. While investigating some of the rc2
> > problems I noticed that g++-multilib wasn't installed and this was disabling the mips64el sanitizers. Installing this
> > didn't cause problems at the time but I've noticed that some of the mips64el sanitizer tests are very memory hungry.
> > I noticed several occasions where parallel asan tests were consuming ~200% of RAM and it would have been
> > swapping heavily (on a USB2 disk) to cope. If later tests require even more RAM then it would explain why the
> > system stops responding. I'm going to uninstall g++-multilib and rebuild rc3. Hopefully that will make check-all work
> > again.
> 
> It seems there's was still just enough multilib support on the machine to trigger mips64el sanitizers again but having worked
> around that the parts of lit that aren't mips64el sanitizers are working so far. llvm-lit is still going but big-endian mips and cross
> compilation was fine so I'm expecting little endian to be ok too. It's showing an ETA of about 90 mins for the remaining tests
> from clang and llvm at the moment. I'm going to run the MIPS32 sanitizer lit tests after that.

This is proving to be a long 90 mins :-). It's currently at 93% and estimating 15 mins but that estimate has been slowly creeping upwards for the last hour or so. I'll check on it first thing in the morning.


More information about the llvm-dev mailing list