On Wed, 27 Feb 2013 14:59:20 +0100, Corinna Vinschen wrote: > On Feb 27 21:29, JonY wrote: > > I'm worried that I might break gcc installs if I overlooked something > > obvious. > > > > The upload will be overwriting the .hint files, java and libffi are > > empty packages (I could not get java to build yet). > > I'm not concerend about java (dum di dum), but why is libffi missing? > > > I can't figure out > > how to pack the debuginfo package, it is always empty. > > Yaakov might be able to help here.
Working on it. > > Will you be able to rollback my uploads in case something does go wrong? > > In theory, yes. If you leave the dependencies in place for the "curr" > release, then the test release shouldn't interfere and testers will > have to care for the stuff themselves. Let's assume for a start, that > downmloaders of a gcc test package know what they are doing. > > Another way to distinguish the new gcc from the current on would be > perhaps to create a "gcc472" package set, distinct from the other gcc > packages. It could install itself into /usr/local, just for the test > period. > Yeah, I know, I know, no official package should install into > /usr/local. Maybe /opt would be fine for once, too. The only way to really test GCC is to throw a lot of software at it and see what breaks, and short of someone doing a mass rebuild, I'm not sure that will happen unless it goes stable quickly. I volunteered to put 4.5 through its paces, but it remained in testing for so long that Ports almost ended up as a completely forked distro and it took me months to clean up the mess afterwards. Yaakov
