On Sat, 21 Jan 2006, Tony Caduto wrote:
> Tom Lane wrote:
>
> > Please go back and read the very extensive discussions of this question
> > a couple years ago in the pghackers archives (around the time we were
> > implementing DROP COLUMN, iirc). There is no "just" about it.
>
> So it's a real big deal to add another column to pg_attribute and
> populate it once with the same value as attnum when a column is added to
> a table? The attorder wouldn't really have to do anything except allow
> the developer to sort on that instead of attnum, and then we could just
> change the values in attorder to get the visual ordering we want.
If the effect you're looking for is entirely visual reordering with no
logical effects, then the server really doesn't need to be involved at
that level.
However, that's a very weak definition of column reordering because it
potentially hides the actual order used in those few places that SQL
cares. I think a client that tells me the columns are a, b, c but then
gives me an error on "insert into table values (aval, bval, cval)" because
the actual logical order is different is probably fairly broken.
