Rich Cullingford <rculling@sysd.com> writes:
> In trying to write a utility to drop/restore table indexes (as an aid to
> a fast reload of the associated tables), I ran into the definition of
> the field pg_index.indkey as an int2vector.
> How can I deal with this type in plpgsql?
You should be able to index into it (indkey[0], etc), but I cannot think
of any good reason to be looking at the details of the pg_index row for
your purposes. It is certainly not reasonable to think of writing
plpgsql code to reconstruct 7.4 expressional index definitions from
what's in pg_index :-(. I would suggest using the pg_get_indexdef
function instead. Perhaps something like
regression=# select pg_get_indexdef(indexrelid) from pg_index where indrelid = 'tenk1'::regclass;
pg_get_indexdef
-----------------------------------------------------------
CREATE INDEX tenk1_hundred ON tenk1 USING btree (hundred)
CREATE INDEX tenk1_unique2 ON tenk1 USING btree (unique2)
CREATE INDEX tenk1_unique1 ON tenk1 USING btree (unique1)
(3 rows)
regression=#
pg_get_indexdef() exists at least as far back as 7.0, and probably
further (pg_dump has relied on it for a long time).
regards, tom lane