What do you think about that idea of changing how dpkg uses architectures?
The architectures like i386, mips, alpha would be the top level architectures. They would have sub-classes like linux-i386 and hurd-i386. A higher level architecture would be compatible with systems of its sub-classes. In addition there could be architectures like linux-all which means that the package is common for all Linux systems. The specifier all would match any system whatever. So a system running on linux-i386 could install packages of the architectures linux-i386, i386, linux-all and all and a system running on hurd-i386 could install packages of the architectures hurd-i386, i386, hurd-all and all. I think this system is mostly compatible with the current situation. The only problems I have found is that a Hurd system would think current Linux packages are compatible with it and the current Linux/i386 systems would think linux-i386 packages are incompatible with it. I don't know much about the real implementation of the architecture system so there could be inconsistencies here. How difficult it is to implement a system like this? Are there better solutions to classify packages of different systems? What are the package pools I have read about on this mailing list?
