On Thu, Feb 22, 2018 at 04:06:06PM -0500, Theodore Ts'o wrote: > > Thanks for the patch and the explatnation. Hmmm... so that memans > that if a package is every Arch:all, it's impossible to ever > transition to Arch:any without running afoul of the potential for the > package to be installed for multiple architectures. Grump.
The inverse, surely. If it's :any (and m-a:same), changing it to :all won't work *if it has deps*. If there weren't deps involved here, it would be fine, since replacing two installations of old-package with one installation of new-package would be exactly what you wanted. But, due to there being deps, you end up with a situation where: 1) We upgrade any->all on the primary arch (say, amd64). 2) libold:amd64 brings in libnew:amd64, but nothing pulls in libnew:i386. 3) libold:i386 and libold:amd64 are now different versions. 4) The situation in (3) isn't allowed, so we remove i386. 5) Anything that depended on libold:i386 now needs to be removed. Anyhow, with the exception of transitionals, it seems generally weird that one would want to go from any+same to all, as those tend to cater to very different types of packages, so meh. ... Adam