There is no guarantee that a given sequence is used only for one column
in one table, as I understand it. So renaming it could screw you up badly.
If we made 'serial-ness' first class, and hid the sequence completely
from view, this would make more sense.
Or am I smoking crack?
andrew
Jonathan Gardner wrote:
>I've always wanted to be a PoatgreSQL hacker, and I am going to try this
>change out first. Bruce said that it's kind of low on the priority list, so
>hopefully I won't be holding anyone up if I take a while to get it right.
>
>The bug is that when you craete a table with a "SERIAL" column, and/or a
>"PRIMARY KEY", and then change that table's name via "ALTER TABLE", the
>related sequence and primary key index do not change their names
>accordingly.
>
>I think the change is simple -- just update the names of the related
>sequences and indexes when the table name changes. Of course, the entire
>operation will have to be done in a transaction block.
>
>I'm playing with the CVS version of PostgreSQL right now -- compiling it and
>testing it. In the meantime, I am coming up with some unit tests to
>determine whether I succeed or not.
>
>Any comments about the project and its scope?
>
>
>
>
