Providing a transition strategy that enables us to use TXT as
a short/medium
term strategy makes sense to me; relying on it for all time seems
like a collapse into sub-typing where the TXT RR structure
doesn't permit
or encourage it. To put it bluntly, in my opinion it is very
bad engineering.
I really do not see that there is a problem.
If DNS was being designed today we would give it a text keyword based
RR key. That is the technique that has been successful on the Web.
From a formal point of view I would be equally happy to write a VDM
or Z specification for DNS using traditional RRs or SRV style
prefixing.
You might have a point on pure aesthetics, but I do not think that
the 'bad engineering' claim is valid. Prefix is how I want the DNS to
look, Prefix is where I want to take the DNS.
Sure it might look a little ugly. Seen the I18N spec recently?
Just think of it this way, the RR itself only defines the SYNTAX
of the record. The semantics are defined by the RR type and the
Prefix. So far most RRs only have default semantics defined for the
unprefixed label, SRV being the exception.
I18N names are definitively a hack. But they are not a terrible
architecture, they are merely a slightly whacky unicode encoding.