On 2018-02-17 11:39:57 -0500, Tom Lane wrote: > BTW, I was wondering if it'd be a good idea to try to forestall future > oversights of this sort by adding a test query in, say, misc_sanity.sql. > Something like
+many > relname | attname | atttypid > -------------------------+-----------------+-------------- > pg_aggregate | agginitval | text > pg_aggregate | aggminitval | text Seems like it should have a toast table, it's not too hard to imagine some form of aggregate to have a large initial condition. > pg_attribute | attacl | aclitem[] > pg_attribute | attfdwoptions | text[] > pg_attribute | attoptions | text[] Seems like it should have a toast table, but I think other people differed. > pg_authid | rolpassword | text that seems not not to require one. > pg_class | relacl | aclitem[] > pg_class | reloptions | text[] > pg_class | relpartbound | pg_node_tree I still think this should have one, but we disagreed. Only argument against that I can see is complexity around rewrites. > pg_collation | collversion | text unnecessary. > pg_database | datacl | aclitem[] > pg_default_acl | defaclacl | aclitem[] hm. > pg_event_trigger | evttags | text[] unnecessary? > pg_extension | extcondition | text[] > pg_extension | extconfig | oid[] > pg_extension | extversion | text Possibly add? > pg_foreign_data_wrapper | fdwacl | aclitem[] > pg_foreign_data_wrapper | fdwoptions | text[] Possibly add? > pg_foreign_server | srvacl | aclitem[] > pg_foreign_server | srvoptions | text[] > pg_foreign_server | srvtype | text > pg_foreign_server | srvversion | text > pg_foreign_table | ftoptions | text[] Add? That's a fair number of potentially longer fields. > pg_index | indexprs | pg_node_tree > pg_index | indpred | pg_node_tree Imo we should add one here, but honestly I can recall only one or two complaints. > pg_init_privs | initprivs | aclitem[] Only if we decide to make other aclitem containing fields toastable. > pg_largeobject | data | bytea We deal with this in other ways. > pg_partitioned_table | partexprs | pg_node_tree Probably makes sense. > pg_pltemplate | tmplacl | aclitem[] > pg_pltemplate | tmplhandler | text > pg_pltemplate | tmplinline | text > pg_pltemplate | tmpllibrary | text > pg_pltemplate | tmplvalidator | text Hard to imagine this being required, unless we just want to make aclitem[] toastable as a rule. > pg_policy | polqual | pg_node_tree > pg_policy | polroles | oid[] > pg_policy | polwithcheck | pg_node_tree Yes. > pg_replication_origin | roname | text unnecessary. > pg_subscription | subconninfo | text > pg_subscription | subpublications | text[] > pg_subscription | subsynccommit | text I'd say yes, with a few alternative hosts connection info can become quite long. > pg_tablespace | spcacl | aclitem[] > pg_tablespace | spcoptions | text[] Hm? > pg_ts_dict | dictinitoption | text probably not. > I think in most of these cases we've consciously decided not to toast-ify, > but maybe some of them need a second look. Greetings, Andres Freund
