At Wed, 12 Aug 2015 07:23:59 -0400, Andrew Sullivan <a...@anvilwalrusden.com> wrote:
> > So we are in agreement that glibc's stub resolver is acting really dumb > > here? > > I think that's overstating it. It appears that glibc implemented the > protocol according to a widely-held but (at least mostly) undocumented > feature of the protocol. I think my reading of the documents is more > in line with your interpretation, but as you can see in the thread > Mark thought "add" meant something obvious. Given the wide deployment > of glibc, it's rather hard to call it "wrong" -- it's got a running > code argument, after all. I think this is probably a gap in the > specification. It's hardly the first one in the DNS. FWIW the stub resolver library in BSD variants derived from a very old version of BIND (ver 4?) has been behaving that way for more than (in my understanding) several decades: https://github.com/freebsd/freebsd/blob/master/lib/libc/net/gethostbydns.c it goes through the answer section in gethostanswer() as a one-pass operation, replacing the search name with CNAME target as it sees CNAMEs. I suspect it was implemented way before the first stub resolver of glibc, and I wouldn't even be surprised if the glibc implementation referred to the BSD behavior. -- JINMEI, Tatuya _______________________________________________ DNSOP mailing list DNSOP@ietf.org https://www.ietf.org/mailman/listinfo/dnsop
