Hi, On Thu, Aug 18, 2011 at 11:24:45AM +0300, Pekka Pessi wrote: > The header in question <sofia-sip/sip_extra.h> is autogenerated by an > awk script. It would help much if I could see its contents...
This is a bug in the makefiles, not the header or the awk script as such. It only shows up under heavily parallel builds. Lucas usually builds these with `make -j10`, and even then a test rebuild showed we don't always lose the race, it depends on the order the threads complete. I've never seen it happen with -j4 on a quad core box (and I think I've run some at -j8 on a hyperthreaded quad too and didn't see it fail on that either). At the instant of the failure, the header would have been partially complete (or perhaps partially overwritten by another thread that also tried to make it) - by the time the build actually ceases though it will probably look just fine. What we're looking for is a hole in the makefile prerequisites that either doesn't serialise creation of this file correctly, or that leads to it being created by two separate goals, where the second one overwrites the output of the first (possibly while the targets of the first are actually trying to use it). I had a quick look through the makefiles myself at the time (and pinged you on #sofia-sip to see if it was already known), but no obvious problem jumped out at me, and I haven't had enough spare time to fully analyse the dependency chain there. I have seen problems like this with parallel builds and generated headers before though, so I am fairly confidant this is the kind of issue we are looking for with this one. I was less confidant that I could fix it 'by eye' though without a better understanding of all the build dependencies and/or the ability to actually reproduce it myself. It will be good if we can find a fix for it - but I'll also downgrade the severity of this bug if we don't find one by the time it becomes a release blocker. None of the normal Debian buildds have been hit by it to date, and "don't build with -j10" doesn't seem like a totally unreasonable workaround until we do. Cheers, Ron -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
