Losses at Mandriva

Not only that, but libraries are actually the easiest problem to solve. Yes, obviously you can always just install enough versions so everything can start properly with its preferred version.

But, on a modern system the most interesting and difficult interaction is at runtime, as various programs talk to each other, through dbus or pipes or sockets or whatever. Trying to keep all that messaging 100% backward compatible is a nightmare and there's no simple solution to this problem: newer applications expect the services they interact with to have newer features: that's why we upgrade, after all. That's why you generally need to upgrade the entire desktop environment, because those applications and applets are strongly linked.

And, although maybe it's appropriate for LWN, you are completely ignoring the large number of non-Linux platforms that most upstream developers support. Requiring upstream developers to be fluent in not just RPM vs. DEB, Red Hat vs. OpenSUSE vs. Fedora vs. Gentoo ebuilds vs. Ubuntu vs. Slackware vs. PackageKit, but to ALSO understand FreeBSD ports and the differences building packages for the *BSDs, Solaris, HP-UX, AIX, etc. is completely infeasible. And upstream developers are simply not willing to compromise the software in Linux-specific ways, the way package maintainers often need or want to. That's no one's fault: it's the result of perfectly legitimate differences in focus and priorities.

As an end-user I'm sure it seems like a lot of wasted effort. But believe me, if you were involved in the guts of these efforts, whether on the package maintainer side or the upstream developer side, you would (modulo the inevitable conflicts that always arise) agree that there's no really practical alternative and there are even advantages to doing things this way.

ObDisclosure: I'm an upstream developer for a common software package.