Fixed a couple bugs in column default normalization: 1) editing the expression tree in place so the change is invisible to debug logging, 2) normalizing some literals that should manage their own round-trip serialization

Author: zachmu

sql/analyzer/resolve_column_defaults.go

@@ -438,13 +438,27 @@ func normalizeDefault(ctx *sql.Context, colDefault *sql.ColumnDefaultValue) (sql
 		return colDefault, transform.SameTree, nil
 	}
 	typ := colDefault.Type()
-	if types.IsTime(typ) || types.IsTimespan(typ) || types.IsEnum(typ) || types.IsSet(typ) || types.IsJSON(typ) {
+	if skipDefaultNormalizationForType(typ) {
 		return colDefault, transform.SameTree, nil
 	}
 	val, err := colDefault.Eval(ctx, nil)
 	if err != nil {
 		return colDefault, transform.SameTree, nil
 	}
-	colDefault.Expr = expression.NewLiteral(val, typ)
-	return colDefault, transform.NewTree, nil
+	
+	newDefault, err := colDefault.WithChildren(expression.NewLiteral(val, typ))
+	if err != nil {
+		return nil, transform.SameTree, err
+	}
+	return newDefault, transform.NewTree, nil
+}
+
+// skipDefaultNormalizationForType returns true if the default value for the given type should not be normalized for 
+// serialization before being passed to the integrator for table creation
+func skipDefaultNormalizationForType(typ sql.Type) bool {
+	// Extended types handle their own serialization concerns
+	if _, ok := typ.(types.ExtendedType); ok {
+		return true
+	}
+	return types.IsTime(typ) || types.IsTimespan(typ) || types.IsEnum(typ) || types.IsSet(typ) || types.IsJSON(typ)
 }