With this patch, all instances of tableoid, cmin, cmax etc. will get a non-NULL value irrespective of whether they appear on nullable side of the join or not.
e.g. select t1.c1, t1.tableoid, t2.c1, t2.tableoid from ft4 t1 left join ft5 t2 on (t1.c1 = t2.c1); run in contrib_regression gives output c1 | tableoid | c1 | tableoid -----+----------+----+---------- 2 | 54282 | | 54285 4 | 54282 | | 54285 6 | 54282 | 6 | 54285 8 | 54282 | | 54285 10 | 54282 | | 54285 12 | 54282 | 12 | 54285 but the same query run on local tables select t1.c1, t1.tableoid, t2.c1, t2.tableoid from "S 1"."T 3" t1 left join "S 1"."T 4" t2 on (t1.c1 = t2.c1); gives output c1 | tableoid | c1 | tableoid -----+----------+----+---------- 2 | 54258 | | 4 | 54258 | | 6 | 54258 | 6 | 54266 8 | 54258 | | 10 | 54258 | | 12 | 54258 | 12 | 54266 BTW, why do we want to set the column values with invalid values, and not null? Wouldn't setting them NULL will be a better way? On Tue, Apr 5, 2016 at 12:11 PM, Etsuro Fujita <fujita.ets...@lab.ntt.co.jp> wrote: > On 2016/03/29 15:37, Etsuro Fujita wrote: > >> I added two helper functions: GetFdwScanTupleExtraData and >> FillFdwScanTupleSysAttrs. The FDW author could use the former to get >> info about system attributes other than ctids and oids in fdw_scan_tlist >> during BeginForeignScan, and the latter to set values for these system >> attributes during IterateForeignScan (InvalidTransactionId for >> xmins/xmaxs, InvalidCommandId for cmins/cmaxs, and valid values for >> tableoids). Attached is a proposed patch for that. I also slightly >> simplified the changes to make_tuple_from_result_row and >> conversion_error_callback made by the postgres_fdw join pushdown patch. >> What do you think about that? >> > > I revised comments a little bit. Attached is an updated version of the > patch. I think this issue should be fixed in advance of the PostgreSQL > 9.6beta1 release. > > Best regards, > Etsuro Fujita > -- Best Wishes, Ashutosh Bapat EnterpriseDB Corporation The Postgres Database Company
