Tom Lane wrote:
> +1 on using the parser location mechanism and avoiding the
> terminology problem altogether.
I figured we would let the parser only point to the UNION or VALUES or
whatever word. It would be fairly cumbersome to drag the individual
expression positions down into select_common_value() for full
precision.
> I fear though that we're not set up
> to have multiple locations in one error report. Will it be
> sufficient if we point at one of the two offending expressions? (I'd
> guess pointing at the second makes the most sense, if feasible.)
I don't think that would help. In the example I was looking at 90
expression and I had no idea in most cases what their results types
are, so if it tells me that the 15th expression somewhere doesn't
match, I would need to know which is the other mismatching one.
--
Peter Eisentraut
http://developer.postgresql.org/~petere/