monline

Well, this is a classic one. The maintainer of the netbase package proposed a fundamental change in how network sockets should be managed regarding IPv4 vs. IPv6 (ie. IPv4-mapped IPv6 addresses). He suggested that the default of the net.ipv6.bindv6only kernel variable should be set to 1 (whereas the current default is 0). After "some" discussion (he and two other guys had something to say about the topic in a one and a half month period) he made the change on 6th December 2009.

I guess he thought that since there was almost no interest in his proposal, it was OK to just "flip the switch". His theory was that the change was needed and only a few old apps will break ... which should be fixed to work together with the change. Well ... I guess it was never easier to break half the application servers out there than this one. The result of the change was a million and one bugreports from users of a thousand and one applications. Obviously there were a bit more "old"/"bad" applications out there than initially expected.

But that was not the real problem. What made it really bad was that Marco (the package maintainer) could not be reasoned with and nobody else in the Debian team (who could have done something) seemed to care. It took half a year to admit that the change was too early and not very well executed (flipping a switch that could potentially -and actually did IRL- break a large percent of network applications and not "shouting" at the user/admin while the switch is being flipped ... well ... imho it's a very unwise thing to do). The change was intoduced in netbase v4.38 (6th Dec 2009) and was completely reverted in v4.42 (25th Jun 2010).

I guess this is still better than some of the bugs in eg. Ubuntu or Firefox that have been there for years and still counting ...

Here're a few other bugreports related to this change:

• sun-java6-jre: net.ipv6.bindv6only=1 breaks java networking
• net.ipv6.bindv6only=1 breaks java networking

These bugs have tons of references to other (merged) bugs. Summing all the hours this bug caused for people (admins of Debian Squeeze systems) all over the world, I guess we'd get years of wasted work/debugging/frustration just because one person decided it was time for a change. This kindof reminds me of the new movie In Time. Who pays for all that lost time and who gains anything from it?