I did that RR type implementation several months before I got the response
from the author on the proposed erratum.  Under the supposition that the
author would agree with my obvious correctness, I put in a requirement
that all characters in labels be alphanumeric.

As it turned out my correctness wasn't obvious to the author after all,
and perhaps I should have revisted the coding decision then.

The problem is, if BIND receives a zone transfer containing a record like
this, it's quite likely to dump a copy of the zone to disk, and if it's
using masterformat "text", then it won't be able to reload the zone
afterward.  This admittedly isn't a huge problem -- it wastes time on
unnecessary retransfers, and wastes disk space on copies of the "bad"
zone file left around for inpsection, but it doesn't prevent the server
from working -- but it is a problem.  It really should've been caught
before the CAA code point was issued.

Anyway, removing the following lines from lib/dns/rdata/generic/caa_257.c
should make the FORMERR go away:

        /* Check the Tag's value */
        for (i = 0; i < len; i++)
                if (!alphanumeric[sr.base[i]])
                        RETERR(DNS_R_FORMERR);