Paul A Vixie writes:
On the other note - MHO is it was a Bad Idea to put these checks into the resolver library. No objections towards having it as an option in the code for primary zones in name servers, though.
We debated this for a long while. It turns out that a lot of applications don't check the results of the names they get back from gethostbyaddr(), and that the architecturally pure solution would have meant letting these older applications, and their users, burn. I wasn't willing to do that. There was a CERT advisory about sendmail's queue files and PTR RRs that included \n characters, but there was no CERT advisory about the various applications who expand DNS names on shell command lines. Yes, the authors of those scripts should be shot. (Especially whoever wrote Vixie Cron.) But something would have to be done about Java as well. And who knew what else? It was not worth skewering the universe to maintain the purity of the design. So, we put the checking in at several levels.
Great. The joke is - all the old code I know worked well with the status quo ante. (It might have been a conscious decision not to know any code that wouldn't work, but it's a slightly different matter). So do I understand rightly, that there are some criteria for incorporating some bugs into the Internet (semi-)standards? Like, Vixie Cron scores here, avg's db - doesn't? Sendmail scores, Apple's resolver library (the earlier versions, at least) not? (these are rhetorical quesions, anyway) As of Java, the resolver part of its API is done with all the Sun's cluefulness (if any) about DNS, so I wonder why you started taking its (Sun's) antics into consideration now, and not when they were trying to cross YP and DNS.
Check out draft-andrews-dns-hostnames-03.txt for more info on this subject.
Now let's get off the nanog list. Dima