Revisiting the multi-arch premises for a more capable multi-arch system

Within the Debian project, agreeing to and implementing a plan for parallel installability of multiple architecture versions of a package ("MultiArch") was a complicated multi-year process. With the Wheezy release, we are finally starting to see the fruits of this labor, prototyped and designed by Steve and a small, tireless team of dedicated volunteers.

As we start taking the MultiArch infrastructure for granted and using it in novel ways (e.g. for interpreter packages and cross compiling), a number of unresolved issues from the original spec and the in-progress MultiArchCross spec are rearing their heads again. This will also impact further work, as we prepare the ports infrastructure and eventually the archive for hosting partial architectures (e.g. x32).

In this talk I want to re-visit the theoretical foundation for MultiArch, and propose a slightly tweaked definition, which should have little impact on multi-arch dependencies already in the archive, but may help us reason better about the multiarch-aware dependency graph. Hopefully we will come to an agreement in the following discussion which can help us solve the interpreter and cross-compiling issues.