[compiler-rt] Fix a flaky heap-overflow-large.cc test

[kubamracek]

Summary

The regression test in test/asan/TestCases/heap-overflow-large.cc returns an exit code with this statement:

return res ? res : 1;

with the intention of always returning a non-zero exit code, no matter what res is (it's a result of a pretty much random memory read). However, when res is accidentally a multiple of 256, then (at least on Darwin) this will be treated as a zero exit code. Let's change the test to only return one byte of res and not the whole int.

[samsonov]

Nice catch! I'd go for smth. simpler than that, e.g.

unsigned res = x[index];
return (res % 10) + 1;

[kubamracek]

Updating patch.

[kubamracek]

Ah, forgot the unsigned...

[samsonov]

LGTM

[kubamracek]

Landed in r232920.

test/asan/TestCases/heap-overflow-large.cc

int main(int argc, char *argv[]) {
   fprintf(stderr, "main\n");
   int *x = new int[5];
   memset(x, 0, sizeof(x[0]) * 5);
   int index = atoi(argv[1]);
   int res = x[index];
   // CHECK: main
   // CHECK-NOT: CHECK failed
   delete[] x;
-  return res ? res : 1;
+  return (char)res ? (char)res : 1;
 }