gcc version 4.1.0 20060219 (prerelease) produces spurious warnings when used with -Wall and -O3 together. Previous released versions haven't had this behaviour. It appears to be issuing multiple warnings when string functions are inlined using the __builtin_ versions eg __builtin_strcmp. For example: test-strcmp.c:9: warning: pointer targets in passing argument 1 of ‘strlen’ differ in signedness test-strcmp.c:9: warning: pointer targets in passing argument 1 of ‘__builtin_strcmp’ differ in signedness test-strcmp.c:9: warning: pointer targets in passing argument 1 of ‘__builtin_strcmp’ differ in signedness test-strcmp.c:9: warning: pointer targets in passing argument 1 of ‘__builtin_strcmp’ differ in signedness The inital warning about strlen is great, the rest is just about GCC internals and confusing. Here is a code snippet: #include <string.h> int main(void) { int i; const char *s="Hello"; unsigned char data[40]; strcpy(data, s); i=strcmp(data, s); return 0; } this will show the problem when built with the prerelease version of gcc (4.1.0 20060219 (prerelease)) as follows: gcc -O3 -Wall -o test test.c I'll attach the build log from gcc -v -save-temps later. Thanks Andrew Roberts
Created attachment 10880 [details] gcc -v -save-temps -O3 -Wall output
This is really a glibc bug as far as I can tell. Lets look into what glibc exands the string functions to: i=__extension__ ({ size_t __s1_len, __s2_len; (__builtin_constant_p (data) && __builtin_constant_p (s) && (__s1_len = strlen (data), __s2_len = strlen (s), (!((size_t)(const void *)((data) + 1) - (size_t)(const void *)(data) == 1) || __s1_len >= 4) && (!((size_t)(const void *)((s) + 1) - (size_t)(const void *)(s) == 1) || __s2_len >= 4)) ? __builtin_strcmp (data, s) : (__builtin_constant_p (data) && ((size_t)(const void *)((data) + 1) - (size_t)(const void *)(data) == 1) && (__s1_len = strlen (data), __s1_len < 4) ? (__builtin_constant_p (s) && ((size_t)(const void *)((s) + 1) - (size_t)(const void *)(s) == 1) ? __builtin_strcmp (data, s) : (__extension__ ({ __const unsigned char *__s2 = (__const unsigned char *) (__const char *) (s); register int __result = (((__const unsigned char *) (__const char *) (data))[0] - __s2[0]); if (__s1_len > 0 && __result == 0) { __result = (((__const unsigned char *) (__const char *) (data))[1] - __s2[1]); if (__s1_len > 1 && __result == 0) { __result = (((__const unsigned char *) (__const char *) (data))[2] - __s2[2]); if (__s1_len > 2 && __result == 0) __result = (((__const unsigned char *) (__const char *) (data))[3] - __s2[3]); } } __result; }))) : (__builtin_constant_p (s) && ((size_t)(const void *)((s) + 1) - (size_t)(const void *)(s) == 1) && (__s2_len = strlen (s), __s2_len < 4) ? (__builtin_constant_p (data) && ((size_t)(const void *)((data) + 1) - (size_t)(const void *)(data) == 1) ? __builtin_strcmp (data, s) : (__extension__ ({ __const unsigned char *__s1 = (__const unsigned char *) (__const char *) (data); register int __result = __s1[0] - ((__const unsigned char *) (__const char *) (s))[0]; if (__s2_len > 0 && __result == 0) { __result = (__s1[1] - ((__const unsigned char *) (__const char *) (s))[1]); if (__s2_len > 1 && __result == 0) { __result = (__s1[2] - ((__const unsigned char *) (__const char *) (s))[2]); if (__s2_len > 2 && __result == 0) __result = (__s1[3] - ((__const unsigned char *) (__const char *) (s))[3]); } } __result; }))) : __builtin_strcmp (data, s)))); }); Yes that mess. This is not a gcc bug, there is nothing GCC can do better. Anyways with ppc-darwin, I only get the following warnings (not using the preprocessed source): t.c:8: warning: pointer targets in passing argument 1 of 'strcpy' differ in signedness t.c:9: warning: pointer targets in passing argument 1 of 'strcmp' differ in signedness So this is a bug in the glibc headers which exposed the other warnings.