On Tue, Jan 21, 2020 at 10:10:52AM -0500, Kurt Mosiejczuk wrote: > On Tue, Jan 21, 2020 at 04:02:34PM +0100, Marc Espie wrote: > > > > This happens to me fairly frequently, whether it is on that sparc64 box or > > > my amd64 laptop. Yes, I use FETCH_PACKAGES. No, my /etc/installurl is not > > > wrong, because if I then run "pkg_add -a py-numpy" it will resolve > > > everything just fine. So this is something specific to how the ports > > > infrastructure handles the situation. > > > If things are not perfectly in synch, FETCH_PACKAGES won't help. > > It will look up *exactly* the FULLPKGNAME that's currently in your ports > > tree, so if your ports tree is behind or if the packages haven't made it to > > the mirror, you're out of luck. > > It used to be that in such cases, it would fail over to building the new > version. Here it just gives up completely. Did that change at some point?
Oh, that's a new one. I'll check whether I can reproduce/fix. Some things changed. Specifically the old test would do BUILD_DEPENDS along with RUN_DEPENDS in a strange way, but I never intended to make it fail.