Peter Eisentraut <peter_e@gmx.net> writes:
> On 12/21/12 7:26 AM, Andres Freund wrote:
>> Hm. Wouldn't it be better to create a pg_node_tree[] and use that in
>> pg_attribute? Its already in the variable length part of pg_proc
>> anyway...
> That sounds like a good idea. I don't know why they are currently
> stored as a list.
They're stored as a list because that's what's convenient for use by the
parser/planner. I believe that a change like this would be quite
inconvenient on that end, and that that is not where we want to put the
inconvenience. I'm also concerned about possibly breaking any
third-party code that's already coping with the current representation.
regards, tom lane