Andrew,
The Tom's message was in Dec/2006. We are in 2010.
Sincerelly, I'm not afraid to seem naive, but I believe that a column that inherits the persistent semantics of attnum solves the 99.9% problem with column reordering of legacy software.
The exceptions seems to be:
1) software that address buffers based on attnum. Tipically a core/hacker software.
2) INSERTs without naming the columns. This could be solved when attnum become changeable with a server ou database variable allow_attnum_changes with false default value.
The problem addressed by Tom about the need of a primary key for attributes is almost the same of the current solutions to reorder the columns:
a) recreate the table
b) "shift" the columns.
Nilson
On Tue, Jul 27, 2010 at 6:45 PM, Andrew Dunstan
<andrew@dunslane.net> wrote:
Nilson wrote: Quoting "wiki.postgresql.org/wiki/Alter_column_position <http://wiki.postgresql.org/wiki/Alter_column_position>" :
"The idea of allowing re-ordering of column position is not one the postgresql developers are against, it is more a case where no one has stepped forward to do the work."
Well, a hard journey starts with a single step.
Why not, in the next release that requires to run initdb, add a *attraw* column (a better name is welcome) in the catalog that stores the physical position of column forever, i.e., the same semantics of *attnum*?
Then, in a future release - 9.1 for example - the postgres team can make *attnum* changeable using something like ALTER COLUMN POSITION?
Pros:
- Requires only a couple of changes in main postgreSQL code. It seems to be very simple.
- Allows a smooth and decentralized rewrite of the whole code that may needs the *attraw *attribute - postgreSQL, contribs, pgAdmin, drivers, tools etc. This will give time to developers of that code to detect the impact of semantics change, make the arrangements necessary and also allow the release of production level software using the new feature before *attnum *becomes changeable.
So, when *attnum *becomes read/write, all that software will be ready.
Cons
- More 4 bytes in each row of the catalog.
Nilson
cheers
andrew