Re: safer node casting

On 2017-01-26 20:29:06 -0500, Tom Lane wrote:
> Andres Freund <andres(at)anarazel(dot)de> writes:
> > On 2016-12-31 12:08:22 -0500, Peter Eisentraut wrote:
> >> This is inspired by the dynamic_cast operator in C++, but follows the
> >> syntax of the well-known makeNode() macro.
>
> > The analogy to dynamic_cast goes only so far, because we don't actually
> > support inheritance. I.e. in c++ we could successfully cast SeqScanState to a
> > PlanState, ScanState and SeqScanState - but with our model only
> > SeqScanState can be checked.
>
> Yeah, I was thinking about that earlier --- this can only be used to cast
> to a concrete node type, not one of the "abstract" types like Plan * or
> Expr *. Not sure if that's worth worrying about though; I don't think
> I've ever seen actual bugs in PG code from casting the wrong thing in that
> direction. For the most part, passing the wrong thing would end up firing
> a default: case in a switch, or some such, so we already do have some
> defenses for that direction.

Yea, I'm not actually worried about it - I was more generally remarking
on the analogy made by Peter. For a second I was considering bringing up
the analogy in a comment or such, and this was one of a number of
arguments that made me disregard that idea.

Andres