On Mon Jan 17 13:27:33 2011, ECARROLL wrote:
Show quoted text> I think a lot of this code could be greatly simplified if
> moved from digging into the catalog with multiple SELECTS
> to using one single query.
The reason we don't use information_schema, as far as I can
remember and tell, is:
1. It doesn't cover all the cases
2. We can't guarantee it will exist
3. It's much slower
Number one is not much of an argument, as we can address things on a
case by case (or sub by sub) basis. Number 2 is weak but valid: the
information_schema can be dropped, but pg_class cannot :) Number three
is an important one: in my tests (some time ago, granted), using the
catalogs directly was always faster than the views in
information_schema, which while creating some smaller queries up front,
create some pretty long and convoluted SQL once you unwind all the
views.
There may have been other reasons, but this is probably a topic for the
dbdpg list to argue about: dbd-pg@perl.org
