Apparently the problem was a warning at configure time that is
interpreted as a failure by the buildbot.
The mentioned svn revision is a change on docs/CMake.html which,
obviously, does not affect the build at all. On the past I received
multiple similar bogus messages.
While the buildbots spit such nonsense it is reasonable to expect people
ignoring buildbot's nag messages.
command timed out: 1200 seconds without output, killing pid 22364
process killed by signal 9
program finished with exit code -1
Maybe the buildbot should be switched to do a release build (make ENABLE_OPTIMIZED=1),
instead of a debug build?
The release build has less symbols, and is much faster to link.
The buildbot is probably using a binutils version that has the huge link time bug in 2.17?
There are some false positives, yes, but on the whole, sending out such messages helps us find and fix breakage faster. That's good for LLVM as a project and as a community.
I'm sure the BuildBot owners would be happy to have help improving the heuristics on these messages, e.g., to ignore changes to documentation.
Apparently the problem was a warning at configure time that is
interpreted as a failure by the buildbot.
The mentioned svn revision is a change on docs/CMake.html which,
obviously, does not affect the build at all. On the past I received
multiple similar bogus messages.
While the buildbots spit such nonsense it is reasonable to expect people
ignoring buildbot's nag messages.
I've asked Daniel Dunbar to increase the timeout on this machine. Every build which requires libLTO to be rebuilt will fail, but because it's doing an incremental build the later builds may succeed so long as libLTO doesn't change. It's flapping pretty nastily because of that.
So Daniel, please take this as a reminder to *really* increase the timeout instead of just filing it away into your todo list