On Sat, 2003-12-13 at 20:24, Jeff Bailey wrote: > On Sat, Dec 13, 2003 at 03:30:05PM -0800, Karl M. Hegbloom wrote: > > > I notice that in debian/sysdeps/linux.mk you set NJOBS according to > > the number of CPU's on the compilation system. I think that it might > > be good to support compilation with "distcc". Have you ever thought > > of doing that? Is there any reason not to? > > It sounds interesting (particularily for slow-arch support), but I'd > prefer not to see this type of change until after Sarge releases. I > don't think there's enough benefit to risk this type of change.
FWIW, I think that the risk is probably minimal, provided you ensure that all machines in the cluster are running the same compiler version. The changes to your code will be minimal as well. I think that all you need to change is the way that NJOBS is set, and what the compiler is called. I suppose that builders would run into trouble if they don't keep the compiler versions in sync on all machines. ccache would probably also be helpful, especially if you rebuild it quite a lot. Both ccache and distcc may be used together. Hmmm... for slow arch support, I suppose that a cluster with a cross compiler would be most awesome. I really love distcc. It can build a large software system in record time, provided the build setup is properly designed for "make -j$NJOBS" compilation. I tested with the Linux kernel, and easily halved the build time, even with my compilation cluster at the other end of a relatively (compared to same room ethernet) slow dorm network connection. It's fun to watch with the distccmon program, too. I bet that a compilation cluster setup where the machines are an Internet away would work fine. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
