I looked over
0001-Add-expression-dependencies-on-composite-type-whole-.patch.  That
seems useful independent of the things you have following.

Even though it is presented more as a preparatory change, it appears to
have noticeable user-facing impact, which we should analyze.  For
example, in the regression test, the command

    alter table tt14t drop column f3;

is now rejected, rightly, but the command

    alter table tt14t drop column f3 cascade;

ends up dropping the whole view, which might be a bit surprising.  We
don't support dropping columns from views, so there might be no
alternative, but I wonder what else is changing because of this.  I
think that might deserve a bit more analysis and perhaps some test cases.

--- a/src/backend/catalog/dependency.c
+++ b/src/backend/catalog/dependency.c
@@ -206,6 +206,9 @@ static bool object_address_present_add_flags(const
ObjectAddress *object,
 static bool stack_address_present_add_flags(const ObjectAddress *object,
                                                                int flags,

ObjectAddressStack *stack);
+static void add_type_addresses(Oid typeid, bool include_components,
ObjectAddresses *addrs);
+static void add_type_component_addresses(Oid typeid, ObjectAddresses
+static void add_class_component_addresses(Oid relid, ObjectAddresses
 static void DeleteInitPrivs(const ObjectAddress *object);

For some reason, the function add_object_address() is singular, and
these new functions are plural  Clearly, they take a plural argument, so
your version seems more correct, but I wonder if we should keep that

+                        * e.g. not to the right thing for column-less
tables.  Note that

Small typo there.  (to -> do)

@@ -1639,6 +1641,9 @@ find_expr_references_walker(Node *node,

                add_object_address(OCLASS_PROC, funcexpr->funcid, 0,
+               /* dependency on type itself already exists via function */
+               add_type_component_addresses(funcexpr->funcresulttype,
                /* fall through to examine arguments */
        else if (IsA(node, OpExpr))

Why shouldn't the function itself also depend on the components of its
return type?

How are argument types handled?

Peter Eisentraut              http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:

Reply via email to