[Bug tree-optimization/78586] [7 Regression] Wrong code caused by printf-return-value

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78586

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
--- gcc/gimple-ssa-sprintf.c.jj 2016-11-28 23:50:20.000000000 +0100
+++ gcc/gimple-ssa-sprintf.c    2016-11-29 15:03:29.201488577 +0100
@@ -1159,16 +1159,15 @@ format_integer (const conversion_spec &s
          if (TYPE_UNSIGNED (argtype))
            argmax = build_all_ones_cst (argtype);
          else
-           argmax = fold_build2 (LSHIFT_EXPR, argtype, integer_one_node,
+           argmax = fold_build2 (LSHIFT_EXPR, argtype,
+                                 build_int_cst (argtype, 1),
                                  build_int_cst (integer_type_node,
                                                 argprec - 1));
        }
       else
-       {
-         argmax = fold_build2 (LSHIFT_EXPR, dirtype, integer_one_node,
-                               build_int_cst (integer_type_node,
-                                              typeprec - 1));
-       }
+       argmax = fold_build2 (LSHIFT_EXPR, dirtype,
+                             build_int_cst (dirtype, 1),
+                             build_int_cst (integer_type_node, typeprec - 1));
       res.argmin = argmin;
       res.argmax = argmax;
     }

fixes it, that is a rather obvious bug.  But, I want to think about whether we
can't/shouldn't just use TYPE_{MIN,MAX}_VALUE (TYPE_DOMAIN (...)) instead.