Bruce Hyatt <brucejhyatt@yahoo.com> writes:
> VARCHAR is slower too? There's no check on VARCHAR, is there?
Well, all the textual operators/functions are actually declared to take
and return type TEXT, so when you are working with VARCHAR columns the
expressions have no-op cast nodes in them ("RelabelType" nodes), even
if it's an unconstrained-length VARCHAR. I'm not sure whether the
execution cost of these would be measurable in real applications, but
it's not zero. A bigger problem is that sometimes the planner gets
confused by the RelabelTypes and fails to find as good a plan as it
finds for a pure-TEXT query. Now if you run into that kind of problem
it's a bug and should be reported, but nonetheless you'll get stuck with
bad plans until it's fixed ...
regards, tom lane
