Hi!
This was a regression introduced by r16-1893 (and its backports) for C++,
though for C it had false positive warning for years. Fixed by r16-2000
(and its backports).
Tested on x86_64-linux, committed to trunk as obvious.
2025-07-11 Jakub Jelinek <ja...@redhat.com>
PR c++/120954
* c-c++-common/Warray-bounds-11.c: New test.
--- gcc/testsuite/c-c++-common/Warray-bounds-11.c.jj 2025-07-11
13:12:37.438580284 +0200
+++ gcc/testsuite/c-c++-common/Warray-bounds-11.c 2025-07-11
13:11:58.215096590 +0200
@@ -0,0 +1,21 @@
+/* PR c++/120954 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -Warray-bounds=2" } */
+
+static const int a[32] = { 11, 12, 13, 14, 15 };
+static const int b[32] = { 21, 22, 23, 24, 25 };
+static const int c[32] = { 31, 32, 33, 34, 35 };
+static const int d[32] = { 111, 112, 113, 114, 115 };
+static const int e[32] = { 121, 122, 123, 124, 125 };
+static const int f[32] = { 131, 132, 133, 134, 135 };
+
+int
+foo (int x, int y)
+{
+ int r = 0;
+ if (x >= 0 && x < 32)
+ r = (y >= 4 ? (y >= 0x65 ? a : b ) : c)[x];
+ else if (x >= 0x100 && x < 0x120)
+ r = (y >= 4 ? (y >= 0x65 ? d : e ) : f)[x - 0x100];
+ return r;
+}
Jakub
