Tom Lane wrote:
> Richard Huxton <dev@archonet.com> writes:
>> Any developers care to comment?
>
> plpgsql isn't very bright about tuple descriptors containing dropped
> columns. The immediate problem is that compatible_tupdesc() doesn't
> think a tupdesc containing one column matches one containing the same
> column plus a dropped column; but fixing that would just allow
> subsequent processing to crash :-(. It needs a fair amount of work.
> I think the other PLs are no better.
OK, I think (after a bit of playing) I see what you mean.
If I do the following:
CREATE TYPE tt AS (id int);
CREATE OR REPLACE FUNCTION sel_test2() RETURNS tt AS $$
...
Then it all works OK, because the SELECT inside the function is
returning one int and so is the function.
--
Richard Huxton
Archonet Ltd
