Heikki Linnakangas wrote:
> Zdenek Kotala wrote:
>> Jasen Betts wrote:
>>> template1=# create temp table foo ( x text);
>>> CREATE TABLE
>>> template1=# alter table foo add column y text ;
>>> ALTER TABLE
>>> template1=# alter table foo add column id serial;
>>> NOTICE: ALTER TABLE will create implicit sequence "foo_id_seq" for
>>> serial
>>> colum
>>> n "foo.id"
>>> ERROR: relation "public.foo" does not exist
>>> template1=#
>>
>> It does not work on 8.2.4 as well. It seems PG lost information about
>> schema and try to use default schema. Following command works well:
>>
>> alter table pg_temp.foo add column id serial;
>>
>> It could be use as workaround.
>
> 8.1 creates the sequence in wrong schema:
>
> postgres=# create temp table foo ( x text);
> CREATE TABLE
> postgres=# alter table foo add column id serial;
> NOTICE: ALTER TABLE will create implicit sequence "foo_id_seq" for
> serial column "foo.id"
> ALTER TABLE
> postgres=# \d
> List of relations
> Schema | Name | Type | Owner
> -----------+------------+----------+----------
> pg_temp_1 | foo | table | hlinnaka
> public | foo_id_seq | sequence | hlinnaka
> (2 rows)
>
> The problem seems to be in transformColumnDefinition, where the schema
> of the to-be-created sequence is determined from the relation name
> given. The default creation schema is used, if the user didn't specify
> the schame of the table explicitly, but since it's an ALTER TABLE, it
> really should use the schema of the existing table.
Correct.
> Patch against 8.2 attached, seems to apply to 8.1 and CVS head though I
> haven't tested them.. This is not my area of expertise, so I'm not 100%
> sure this is the right way to fix it.
I looked on it, but I think let parser to fill namespace information in
ctx->relation structure should be better then do it in this place. There
is also unfilled istemp flag.
Zdenek