At 2016-03-24 12:31:16 -0300, alvherre@2ndquadrant.com wrote:
>
> In other words I think the conclusion here is that we must use
> qualified_name in the new production rather than switching the old
> production to any_name.
Makes sense.
> I think I would like to see code implement both alternatives to see
> which one is least ugly. Maybe a third idea will manifest itself upon
> seeing those.
Here's the first one. ExecAlterObjectDependsStmt() looks like this:
+ObjectAddress
+ExecAlterObjectDependsStmt(AlterObjectDependsStmt *stmt)
+{
+ ObjectAddress address;
+ ObjectAddress extAddr;
+ Relation rel = NULL;
+
+ /*
+ * If the parser handed us a RangeVar, we add the relation's name to
+ * stmt->objname so that we can pass it to get_object_address().
+ */
+ if (stmt->relation)
+ {
+ stmt->objname = lcons(makeString(stmt->relation->relname), stmt->objname);
+ if (stmt->relation->schemaname)
+ stmt->objname = lcons(makeString(stmt->relation->schemaname), stmt->objname);
+ if (stmt->relation->catalogname)
+ stmt->objname = lcons(makeString(stmt->relation->catalogname), stmt->objname);
+ }
+
+ address = get_object_address(stmt->objectType, stmt->objname, stmt->objargs,
+ &rel, AccessExclusiveLock, false);
+
+ if (rel)
+ heap_close(rel, NoLock);
+
+ extAddr = get_object_address(OBJECT_EXTENSION, stmt->extname, NULL,
+ &rel, AccessExclusiveLock, false);
+
+ recordDependencyOn(&address, &extAddr, DEPENDENCY_AUTO_EXTENSION);
+
+ return address;
+}
(This works fine for both functions and triggers, I tested it.)
Complete patch attached for reference.
I'll post the get_object_address_rv() variant tomorrow, but comments are
welcome in the meantime.
-- Abhijit