Greg Stark <gsstark@mit.edu> wrote:
> Creating new catalog entries for [temp tables] gives up -- what I
> think is the whole point of their design -- their lack of DDL
> overhead.
As long as we're brainstorming... Would it make any sense for temp
tables to be created as in-memory tuplestores up to the point that we
hit the temp_buffers threshold? Creating and deleting a whole set of
disk files per temp table is part of what makes them so heavy.
(There's still the issue of dealing with the catalogs, of course....)
-Kevin