John DeSoi <jd@icx.net> writes:
> No hard feelings about it, but I'm really surprised existing behavior
> will be broken when the technical reasons for changing it were so weak.
We've only had domains for one release cycle, so it seems to me that
there's not a lot of track record to justify a "this is how we've always
done it" position for domain-related behaviors. Especially when the
odds seem good that few people are really using domains. (Barry pointed
out that JDBC has been broken for the entire release cycle because it
couldn't handle I/O of domain types with the 7.3 RowDescription behavior
... but no one's complained to him yet ...)
I thought the rationale that client libraries would know about base
types but not domain types was pretty good, actually. From the
backend's perspective, reduction of domains to base types in
RowDescription takes extra code and extra cycles, so I'd have resisted
the argument if I had anything to stand on --- but I didn't. I'd say
your suggestion that the client side wants to distinguish domains from
base types is really the weaker end of the argument ...
regards, tom lane