Re: Fixing findDependentObjects()'s dependency on scan order (regressions in DROP diagnostic messages) - Mailing list pgsql-hackers

I wrote:
> I propose that we handle this case by adding a new DEPFLAG_IS_SUBOBJECT
> flag to the column object's flags, denoting that we know the whole table
> will be dropped later.  The only effect of this flag is to suppress
> reporting of the column object in reportDependentObjects.

Here's a proposed patch for that bit.  As expected, it seems to eliminate
the variation in number-of-cascaded-drops-reported under
ignore_system_indexes.  I do not see any regression outputs change
otherwise.

            regards, tom lane

diff --git a/src/backend/catalog/dependency.c b/src/backend/catalog/dependency.c
index 3529084..6b8c735 100644
*** a/src/backend/catalog/dependency.c
--- b/src/backend/catalog/dependency.c
*************** typedef struct
*** 102,107 ****
--- 102,108 ----
  #define DEPFLAG_INTERNAL    0x0008    /* reached via internal dependency */
  #define DEPFLAG_EXTENSION    0x0010    /* reached via extension dependency */
  #define DEPFLAG_REVERSE        0x0020    /* reverse internal/extension link */
+ #define DEPFLAG_SUBOBJECT    0x0040    /* subobject of another deletable object */


  /* expansible list of ObjectAddresses */
*************** reportDependentObjects(const ObjectAddre
*** 886,891 ****
--- 887,896 ----
          if (extra->flags & DEPFLAG_ORIGINAL)
              continue;

+         /* Also ignore sub-objects; we'll report the whole object elsewhere */
+         if (extra->flags & DEPFLAG_SUBOBJECT)
+             continue;
+
          objDesc = getObjectDescription(obj);

          /*
*************** object_address_present_add_flags(const O
*** 2320,2332 ****
                   * DROP COLUMN action even though we know we're gonna delete
                   * the table later.
                   *
                   * Because there could be other subobjects of this object in
                   * the array, this case means we always have to loop through
                   * the whole array; we cannot exit early on a match.
                   */
                  ObjectAddressExtra *thisextra = addrs->extras + i;

!                 thisextra->flags |= flags;
              }
          }
      }
--- 2325,2343 ----
                   * DROP COLUMN action even though we know we're gonna delete
                   * the table later.
                   *
+                  * What we can do, though, is mark this as a subobject so that
+                  * we don't report it separately, which is confusing because
+                  * it's unpredictable whether it happens or not.  But do so
+                  * only if flags != 0 (flags == 0 is a read-only probe).
+                  *
                   * Because there could be other subobjects of this object in
                   * the array, this case means we always have to loop through
                   * the whole array; we cannot exit early on a match.
                   */
                  ObjectAddressExtra *thisextra = addrs->extras + i;

!                 if (flags)
!                     thisextra->flags |= (flags | DEPFLAG_SUBOBJECT);
              }
          }
      }
*************** stack_address_present_add_flags(const Ob
*** 2374,2380 ****
                   * object_address_present_add_flags(), we should propagate
                   * flags for the whole object to each of its subobjects.
                   */
!                 stackptr->flags |= flags;
              }
          }
      }
--- 2385,2392 ----
                   * object_address_present_add_flags(), we should propagate
                   * flags for the whole object to each of its subobjects.
                   */
!                 if (flags)
!                     stackptr->flags |= (flags | DEPFLAG_SUBOBJECT);
              }
          }
      }