Ignore execution time errors for schemadiff view analysis

The UnsupportedConstruct error is returned for execution of select
queries inside vtgate, but when we analyze views we don't care about
those.

Views here are executed at the MySQL level though and always inside a
single shard, so we don't want to return these errors.

The test example here shows a case where it thinks it's in sharded mode
because of a missing dependency, but we only want to show that we're
missing a table dependency.

Signed-off-by: Dirkjan Bussink <[email protected]>

go/vt/schemadiff/schema.go

@@ -1085,6 +1085,11 @@ func (s *Schema) ValidateViewReferences() error {
 					View:   view.Name(),
 					Column: e.Column.Name.String(),
 				}
+			case *semantics.UnsupportedConstruct:
+				// These are error types from semantic analysis for executing queries. When we
+				// have a view, we don't have Vitess execute these queries but MySQL does, so
+				// we don't want to return errors for these.
+				return nil
 			}
 			return err
 		}

go/vt/schemadiff/schema_test.go

@@ -570,6 +570,20 @@ func TestInvalidSchema(t *testing.T) {
 			`,
 			expectErr: &DuplicateForeignKeyConstraintNameError{Table: "t2", Constraint: "const_id"},
 		},
+		{
+			schema: `
+CREATE TABLE users (id INT PRIMARY KEY, name VARCHAR(255));
+CREATE VIEW user_earnings_ranking AS
+SELECT
+    u.id AS user_id,
+    e.total_earnings AS total_earnings,
+    ROW_NUMBER() OVER (
+        ORDER BY e.total_earnings DESC, u.id ASC
+    ) AS ranking
+FROM users AS u JOIN earnings AS e ON e.user_id = u.id;
+`,
+			expectErr: &ViewDependencyUnresolvedError{View: "user_earnings_ranking"},
+		},
 	}
 	for _, ts := range tt {
 		t.Run(ts.schema, func(t *testing.T) {