On Wed, Mar 27, 2019 at 06:19:14PM +0100, Thorsten Glaser wrote: > Yes, but the output, at runtime, will match the native architecture, > so it doesn’t matter if the i386, amd64, x32 packages are installed > for them, or even e.g. the armel package, with qemu-user hooked up > via binfmt-misc, the output will always be, for example, x86_64. So > this is not a point against M-A:foreign.
I'm sorry, but this is wrong again. The output of the uname-dependent script will vary with the native architecture. The promise of M-A:foreign however is that this is not the case. Masquerading the native uname is exactly the problem here. Indeed, we can say that a an Architecture: all package that does not depend on any architecture-dependent packages can safely be marked Multi-Arch foreign. There is a little twist unfortunately: Every package implicitly depends on all essential packages. This means that every package (regardless of its Architecture field) depends on some architecture-dependent package and no package can be safely marked Multi-Arch: foreign. The cases where this becomes relevant are rare, but they do exist. The way to think about it is similar to security issues: Try to be the adversary. Use only "supported interfaces" of the package. If you can tell different architectures apart by interacting with those interfaces, it shouldn't be Multi-Arch: foreign. Helmut
