On Sat Jun 06, 2020 at 08:25:48PM +0100, Stuart Henderson wrote: > > http://build-failures.rhaalovely.net/sparc64/2020-06-04/devel/cpp-hocon.log > > http://build-failures.rhaalovely.net/sparc64/2020-06-04/devel/glog.log > > http://build-failures.rhaalovely.net/sparc64/2020-06-04/multimedia/libmatroska.log > > "ninja: error: manifest 'build.ninja' still dirty after 100 tries" > > I am glad I'm not the only one running into this. Your clocks are > probably not close enough on the build machines. ntpd isn't helping me > with this, clocks are close enough that it tries to skew them, even > with "trusted". I am currently running rdate -n on all my build > machines immediately before starting a build but it's too early to > know if that's really going to help. > > I had been thinking that maybe the machines swapping is resulting > in clocks getting out of whack. Not sure if that's the case but > I can't think of another explanation (and the i386 boxes are > heavily bogged down at times to the extent that BREAK isn't processed).
It's probably a known bug: https://github.com/ninja-build/ninja/issues/1704
