[Bug gcov-profile/99441] [GCOV] Wrong coverage with complex "if" condition
njuwy at smail dot nju.edu.cn
gcc-bugzilla@gcc.gnu.org
Sun Mar 7 13:03:49 GMT 2021
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99441
--- Comment #2 from Yang Wang <njuwy at smail dot nju.edu.cn> ---
(In reply to Andrew Pinski from comment #1)
> So line 27 has a short cutting conditional included, so technically it is
> executed 30 times, one for each side of the &&.
$ gcc -O0 --coverage test.c;./a.out;gcov test;cat test.c.gcov
libgcov profiling error:/home/wangyang/coverage/found
bug/gcc/test.gcda:overwriting an existing profile data with a different
timestamp
File 'test.c'
Lines executed:95.83% of 24
Creating 'test.c.gcov'
-: 0:Source:test.c
-: 0:Graph:test.gcno
-: 0:Data:test.gcda
-: 0:Runs:1
-: 1:typedef __SIZE_TYPE__ size_t;
-: 2:typedef __UINTPTR_TYPE__ uintptr_t;
-: 3:
-: 4:struct S {
-: 5: int a;
-: 6: unsigned short b;
-: 7: int c, d, e;
-: 8: long f, g, h;
-: 9: int i, j;
-: 10:};
-: 11:static struct S *k;
-: 12:static size_t l = 1;
-: 13:int m;
-: 14:
15: 15:static int bar(void) {
-: 16: unsigned i;
-: 17: int j;
15: 18: if (k[0].c == 0) {
15: 19: ++m;
15: 20: size_t n = l * 2;
-: 21: struct S *o;
15: 22: k = (struct S *)__builtin_realloc(k, sizeof(struct S) * n);
30: 23: for (i = l; i < n; i++) {
15: 24: void *p = (void *)&k[i];
15: 25: int q = 0;
15: 26: size_t r = sizeof(struct S);
15: 27: if ((((uintptr_t)p) % __alignof__(long)) == 0
15: 28: &&
30: 29: r % sizeof(long) == 0) {
15: 30: long __attribute__((may_alias)) *s = (long *)p;
15: 31: long *t = (long *)((char *)s + r);
120: 32: while (s < t)
105: 33: *s++ = 0;
-: 34: } else
#####: 35: __builtin_memset(p, q, r);
-: 36: }
-: 37: }
15: 38: return 1;
-: 39:}
-: 40:
1: 41:int main() {
1: 42: k = (struct S *)__builtin_malloc(sizeof(struct S));
1: 43: __builtin_memset(k, '\0', sizeof(struct S));
1: 44: k->a = -1;
16: 45: for (int i = 0; i < 15; ++i)
15: 46: bar();
-: 47:}
Thanks for your reply!
I tried to separate line 27 and put it in a different lines. The coverage of
right side(line 29) is 30, and the left side(line 27) is 15, which is a bit
different from the execution logic about "&&", where the left side of "&&"
should not be less than the right side. Is it a bug?
More information about the Gcc-bugs
mailing list