In the PR, Joseph says that in C char8_t is not a distinct type. So
we should behave as if it can alias anything, like ordinary char.
In C, unsigned_char_type_node == char8_type_node, so with this patch
we return 0 instead of -1. And the following comment says:
/* The C standard guarantees that any object may be accessed via an
lvalue that has narrow character type (except char8_t). */
if (t == char_type_node
|| t == signed_char_type_node
|| t == unsigned_char_type_node)
return 0;
Which appears to be wrong, so I'm adjusting that as well.
PR c/111884
gcc/c-family/ChangeLog:
* c-common.cc (c_common_get_alias_set): Return -1 for char8_t only
in C++.
gcc/testsuite/ChangeLog:
* c-c++-common/alias-1.c: New test.
(cherry picked from commit
281699fbff6262766674ab13087d37db751cd40a)
if (!TYPE_P (t))
return -1;
- /* Unlike char, char8_t doesn't alias. */
- if (flag_char8_t && t == char8_type_node)
+ /* Unlike char, char8_t doesn't alias in C++. (In C, char8_t is not
+ a distinct type.) */
+ if (flag_char8_t && t == char8_type_node && c_dialect_cxx ())
return -1;
/* The C standard guarantees that any object may be accessed via an
- lvalue that has narrow character type (except char8_t). */
+ lvalue that has narrow character type. */
if (t == char_type_node
|| t == signed_char_type_node
|| t == unsigned_char_type_node)
--- /dev/null
+/* PR c/111884 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -Wall" } */
+/* { dg-additional-options "-std=c++20" { target c++ } } */
+/* { dg-additional-options "-std=c2x" { target c } } */
+
+int f(int i)
+{
+ int f = 1;
+ return i[(unsigned char *)&f];
+}
+
+int g(int i)
+{
+ int f = 1;
+ return i[(signed char *)&f];
+}
+
+int h(int i)
+{
+ int f = 1;
+ return i[(char *)&f];
+}