Re: Smaller data types use same disk space

On Thu, Jul 26, 2012 at 9:12 AM, Merlin Moncure <mmoncure(at)gmail(dot)com> wrote:
> On Thu, Jul 26, 2012 at 11:02 AM, Mike Christensen <mike(at)kitchenpc(dot)com> wrote:
>> I don't really think you'd need to decouple the internal column order
>> from what the user sees. A REORDER COLUMNS command should re-build
>> the table with the columns in the specified order. Internally, it
>> should be no different from making a new table, copying all the data
>> over, then deleting the old table. If there's any optimizations that
>> can be done (such as making this faster on large tables), those could
>> be done in future versions. I'd just like to changing column order
>> easier without remaking the table or renaming columns and changing
>> their data types (as suggested by Marc)
>
> That's a controversial point: doing it that way makes reordering of
> large tables highly impractical. A column map turns that into a
> catalog update which can be done at any time. I would argue that you
> can have it both ways: implement the map and have table rebuilding
> operations (like TRUNCATE and CLUSTER) opportunistically do the
> physical swap.

Yea, it all comes down to level of effort. It sounds like adding the
ability to store columns in a different order than they're displayed
would have a huge development cost associated with it, however it
would have some other advantages. For example, perhaps Postgres could
always optimize the column order on disk automatically (similar to how
a compiler does)..

However, once the syntax is defined for how you re-order a column, the
underlying implementation could be changed and improved in future
versions without breaking changes.

Mike