Peter Eisentraut <peter_e@gmx.net> writes:
> Shouldn't this work?
> create table test ( a int, unique (oid) );
> ERROR: CREATE TABLE: column "oid" named in key does not exist
Now it does.
regression=# create table test ( a int, unique (oid) );
NOTICE: CREATE TABLE/UNIQUE will create implicit index 'test_oid_key' for table 'test'
CREATE
regression=# drop table test;
DROP
regression=# create table test ( a int, unique (oid) ) without oids;
ERROR: CREATE TABLE: column "oid" named in key does not exist
regression=# create table test ( a int ) without oids;
CREATE
regression=# alter table test add unique (oid);
ERROR: ALTER TABLE: column "oid" named in key does not exist
regression=# drop table test;
DROP
regression=# create table test ( a int );
CREATE
regression=# alter table test add unique (oid);
NOTICE: ALTER TABLE/UNIQUE will create implicit index 'test_oid_key' for table 'test'
CREATE
regression=#
> And shouldn't the last one say "ALTER"?
The reason that happens is that parser/analyze.c transforms the command
into an ALTER TABLE step that adds a constraint (a no-op in this case)
plus a CREATE INDEX step. The commandTag emitted by the last step is
what psql shows. This could possibly be fixed, but it seems not worth
the trouble.
regards, tom lane
