Neil Conway <neilc@samurai.com> writes:
> ... One question is whether we should make use of
> exec_eval_expr() by representing the RAISE parameter list as a list of
> expressions (each of which would likely be simple enough to evaluate via
> ExecEvalExpr()), or whether we want to extend exec_eval_expr() to handle
> expressions that yield multiple attributes.
I'd lean to the former myself --- which actually does suggest that this
patch is not ready for application yet, because it banks on the
assumption that "x,y,z" should be treated as a single expression.
Now that I think about it, the amount of overhead in that assumption is
pretty high: there's tuple construction and deconstruction involved,
no matter how simple the individual datatypes are. So I'd definitely
prefer to see it changed.
regards, tom lane
