Do not release the lock on the source table.

If we want to release the weaker lock before requestin stronger one (to avoid
deadlock), we need to spend a lot of effort checking for concurrent catalog
changes. This effort is probably not worth the risk.

Author: Antonin Houska

concurrent.c

@@ -83,7 +83,6 @@ pg_rewrite_process_concurrent_changes(EState *estate,
 									  struct PartitionTupleRouting *proute,
 									  LogicalDecodingContext *ctx,
 									  XLogRecPtr end_of_wal,
-									  CatalogState *cat_state,
 									  ScanKey ident_key,
 									  int ident_key_nentries,
 									  Relation ident_index,
@@ -122,9 +121,6 @@ pg_rewrite_process_concurrent_changes(EState *estate,
 		if (dstate->nchanges == 0)
 			continue;
 
-		/* Make sure the changes are still applicable. */
-		pg_rewrite_check_catalog_changes(cat_state, lock_held);
-
 		/*
 		 * XXX Consider if it's possible to check *must_complete and stop
 		 * processing partway through. Partial cleanup of the tuplestore seems