The intermediate expression (unsigned char) '\234' * scale overflows
int on int16 targets, causing the test case to fail there. Fixed by
performing the arithmetic in unsigned type, as suggested by Andrew Pinski.
Regression tested on x86_64-pc-linux-gnu, and on an out-of-tree 16-bit
target with simulator. Manually checked the generated code for pdp11
and xstormy16.
Ok for trunk? (I don't have commit rights so I'd need help committing it.)
gcc/testsuite/
PR testsuite/116806
* gcc.dg/cpp/charconst-3.c: Perform arithmetic in unsigned
type to avoid integer overflow.
Signed-off-by: Mikael Pettersson <mikpeli...@gmail.com>
---
gcc/testsuite/gcc.dg/cpp/charconst-3.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/gcc/testsuite/gcc.dg/cpp/charconst-3.c
b/gcc/testsuite/gcc.dg/cpp/charconst-3.c
index 7c196639131..7b2df8b2998 100644
--- a/gcc/testsuite/gcc.dg/cpp/charconst-3.c
+++ b/gcc/testsuite/gcc.dg/cpp/charconst-3.c
@@ -20,7 +20,7 @@ int main ()
if ('ab' != (int) ((unsigned char) 'a' * scale + (unsigned char) 'b'))
abort ();
- if ('\234b' != (int) ((unsigned char) '\234' * scale + (unsigned char) 'b'))
+ if ('\234b' != (int) ((unsigned char) '\234' * (unsigned int) scale +
(unsigned char) 'b'))
abort ();
if ('b\234' != (int) ((unsigned char) 'b' * scale + (unsigned char) '\234'))
--
2.46.0
