This one is long and boring. You'll want to skip it entirely.

A while ago, Ulrich Drepper made getaddrinfo() in GNU libc behave according to an IETF RFC. Sounds sane, right? Well, not entirely. The way RFC 3484 rules work for IPv4 is less than ideal. On the other hand, returning network address query results unsorted is also less than ideal. Imagine if getent passwd returned lines in a randomized order. It would be perfectly functional, but piss me off to no end until I conditioned myself to always sort them the way I wanted them. I digress.

So, we shipped a stable release that behaved this “new” way. Nobody objected. I don't know if Ubuntu did or not. It would amuse me if they did, but I don't much care.

Then all of a sudden, along comes bug #438179. The gist is that Kurt thinks that trying to make intelligent decisions based on network topology is a bad idea because “it defeats the point of having multiple A-records in the first place.” We'll come back to that in two minutes. James filed the analogous bug in Ubuntu. I keep mentioning Ubuntu for a reason. I'm not going to tell you what that reason is.

The GNU libc maintainers were reluctant to change the default (which is how most other modern operating systems behave; MacOS X being a notable exception, though I heard that they were working on it), so Kurt escalated to the Technical Committee. Now nothing in our lovely and flawed Constitution says that the tech-ctte should be a technically-capable bastion of wisdom and patience; I just think it does because I don't waste my time reading Foundation Documents that don't accurately describe our governance structure.

Now for a brief interlude while I vamp incoherently for some of my less technical readers who are not implementors of a DNS resolver libraries, past IETF participants, DNS administrators, someones who've followed some of the IPv6 transition work, or have invented the Internet lately: Round-robin DNS is a giant hack, a poor man's loadbalancer, as it were, something that only works due to a de facto standard that no standards bodies seem to have thought worth preserving. gethostbyname() has been marked obsolescent in POSIX and deprecate in GNU libc for what seems like forever, and the GNU libc implementation of it suffers from bugs that probably no one should ever fix, and as such, no one should ever use it. getaddrinfo() is not a drop-in replacement for gethostbyname(), and expecting it to behave the same is daft. If it were meant to behave the same, I would expect that it using it would be simply a matter of changing calls to gethostbyname() to getaddrinfo(), rather than the amount of code rewriting you need to do now to change the same effect. Furthermore, nothing guarantees the behavior of gethostbyname() except for years of observed consistent behavior and the expectation that should anyone implement a C or resolver library that sorted gethostbyname() results, there would be much loud screaming that it was the library that's broken, and not the applications or programmers for being naïve.

I hope that was less than helpful. Let's move on with the plot.

While the Technical Committee “deliberated”, and I use the quotation marks because several -ctte members don't seem to have participated at all, Ian Jackson jumped up and down like some kind of Rumpelstiltskin, screaming that he is Der Komissar and that voting is in order, and a few other things I won't mention. On the 20th of September, Ian Jackson called for a vote. On the 20th of September, Ian Jackson uploaded an Ubuntu version of glibc with the default changed. On the 20th of September, my mother had trouble sleeping, though I suspect she has no idea it was because of this debacle.

This story has no ending. Let the rain come down.