I've just noticed that there are some severe bugs in the column-aliasing
feature in the presence of dropped columns. For instance
regression=# create table t1(f1 int, f2 int, f3 int);
CREATE TABLE
regression=# select * from t1 as t(a,b);a | b | f3 -- okay, as expected
---+---+----
(0 rows)
regression=# alter table t1 drop column f2;
ALTER TABLE
regression=# select * from t1 as t(a,b);a | f3 -- what happened to b?
---+----
(0 rows)
regression=# select b from t1 as t(a,b);
ERROR: column "........pg.dropped.2........" of relation "t1" does not exist
regression=#
Of course what's happening there is that the alias b got attached to the
dropped column :-(
I have some fixes that appear to take care of this particular issue but
there's more to do yet --- aliases associated with JOINs over tables
with dropped columns aren't right either. The changes to deal with that
look sufficiently complex that I don't think it's appropriate to try to
back-patch into 7.4 ... I'm just gonna commit the fixes to HEAD.
regards, tom lane