According to the Microsoft docs the API char * strerror(int errnum)
"maps errnum to an error-message string and returns a pointer to the string"
On the practice it turned out that there is one important aspect which is not covered by the description:
This is the code which represents this behavior:
printf("---------- print separately [OK] ----------\n");
printf("strerror(1)=(%s), ptr=(%p)\n", strerror(1), strerror(1));
printf("strerror(2)=(%s), ptr=(%p)\n", strerror(2), strerror(2));
printf("---------- print together [BUG] ----------\n");
printf("strerror(1)=(%s), strerror(2)=(%s)\n", strerror(1), strerror(2));
printf("ptr_1=(%p), ptr_2=(%p)\n", strerror(1), strerror(2));
printf("---------- print together, get pointers separately [BUG] ----------\n");
const char * s1 = strerror(1);
const char * s2 = strerror(2);
printf("strerror(1)=(%s), strerror(2)=(%s)\n", s1, s2);
printf("ptr_1=(%p), ptr_2=(%p)\n", s1, s2);
printf("---------- --------------------------- ----------\n");
Output:
---------- print separately [OK] ----------
strerror(1)=(Operation not permitted), ptr=(000002756DE6BD90)
strerror(2)=(No such file or directory), ptr=(000002756DE6BD90)
---------- print together [BUG] ----------
strerror(1)=(Operation not permitted), strerror(2)=(Operation not permitted)
ptr_1=(000002756DE6BD90), ptr_2=(000002756DE6BD90)
---------- print together, get pointers separately [BUG] ----------
strerror(1)=(No such file or directory), strerror(2)=(No such file or directory)
ptr_1=(000002756DE6BD90), ptr_2=(000002756DE6BD90)
---------- --------------------------- ----------
Compiler optimization flags were not used. The same code works properly on Linux with gcc.
UPDATE: The questions are answered.
Q: Is this behavior a BUG?
Q: Is there another better API which works as strerror on Linux?
The memory returned by strerror can be modified by by subsequent calls to the function.
Instead, use strerror_s which allows you to supply a buffer to write the error string to.
char s1[200], s2[200];
strerror_s(s1, sizeof s1, 1);
strerror_s(s2, sizeof s2, 2);
This isn't a bug. Per the POSIX documentation for strerror:
The returned string pointer might be invalidated or the string content might be overwritten by a subsequent call to
strerror()
So if you call it twice in a single statement, you shouldn't expect more than one of the values to be valid (and it would be perfectly acceptable for your code to crash or do other weird things).
The underlying reason is probably because the Microsoft C library is using a single fixed buffer for the strerror return value, so each call overwrites the previous held value. This is specifically allowed by the standard, for ease of implementation.