I wrote: > You could get back the column name(s) by joining to pg_attribute,
Oh, another possible approach is to use pg_describe_object, which'd be interesting if you also want to handle non-column dependencies. For example,
regression=# drop table t; DROP TABLE regression=# create table t(f1 text); CREATE TABLE regression=# create index ti on t (fipshash(f1)); CREATE INDEX regression=# select * from pg_depend where classid = 'pg_class'::regclass and objid = 'ti'::regclass; classid | objid | objsubid | refclassid | refobjid | refobjsubid | deptype ---------+-------+----------+------------+----------+-------------+--------- 1259 | 48632 | 0 | 1259 | 48627 | 0 | a 1259 | 48632 | 0 | 1259 | 48627 | 1 | a 1259 | 48632 | 0 | 1255 | 16501 | 0 | n (3 rows)
regression=# select pg_describe_object(refclassid, refobjid, refobjsubid) from pg_depend where classid = 'pg_class'::regclass and objid = 'ti'::regclass; pg_describe_object ------------------------- table t column f1 of table t function fipshash(text) (3 rows)