glibc has two strerror_r functions, declared in <string.h> depending on feature macros. But none of them is POSIX compliant. The one which is in effect when _GNU_SOURCE=1 has the declaration char *strerror_r (int, char *, size_t). It has a different return type and therefore a different calling convention than the POSIX function. Test case: ================================================= #define _GNU_SOURCE 1 #include <string.h> #include <stdio.h> int main() { char buf[100]; char *s = strerror_r (-2, buf, sizeof (buf)); printf ("result: %s\n", s); return 0; } ================================================= The one which is in effect when _GNU_SOURCE is undefined and _POSIX_C_SOURCE=200112L has the declaration int strerror_r (int, char *, size_t), which is POSIX compatible, but it has a different return value convention in case of failure. POSIX:2001 and POSIX:2008 say that in case of failure the strerror_r function should return "an error number"; this is the same wording as for pthread_create, pthread_mutex_lock, etc. However, the glibc function returns -1 and sets errno to the error number instead. Test case: ======================================================================== #define _POSIX_C_SOURCE 200112L #include <string.h> #include <errno.h> #include <stdio.h> int main() { char buf[100] = "______"; int r; r = strerror_r (-2, buf, sizeof (buf)); if (r == 0) printf ("result: success %s\n", buf); else if (r > 0) printf ("result: failure %d %s\n", r, buf); else printf ("result: non-POSIX failure %d %d %s\n", r, errno, buf); r = strerror_r (EACCES, buf, 3); if (r == 0) printf ("result: success %s\n", buf); else if (r > 0) printf ("result: failure %d %s\n", r, buf); else printf ("result: non-POSIX failure %d %d %s\n", r, errno, buf); return 0; } ========================================================================== Expected result: result: failure 22 ______ result: failure 34 ______ Actual result: result: non-POSIX failure -1 22 ______ result: non-POSIX failure -1 34 ______
Created attachment 5117 [details] test case
The fix could either be to change the behaviour of __xpg_strerror_r so that it returns the error number instead of assigning it to errno, or to introduce a new function __posix_strerror_r and enable it in <string.h> under the appropriate conditions.
What do other OSes do? What's the behavior on Solaris? Common practice IIRC was at least to return -1. Might just be that the incorrect wording was copied into the strerror_r spec.
According to http://www.freebsd.org/cgi/man.cgi *BSD and Solaris both return the error number.
> What do other OSes do? What's the behavior on Solaris? Here are the results of the test program: Solaris 10: result: failure 22 Unknown error result: failure 34 Unknown error FreeBSD 6.4, OpenBSD 4.4: result: failure 22 Unknown error: -2 result: failure 34 Pe HP-UX 11.31: result: failure 22 ______ result: failure 34 ______ AIX 6.1: result: success Error -2 occurred. result: success Pe Tru64 5.1: result: non-POSIX failure -1 22 Error -2 occurred. result: success Pe You can see that the major Unices use the POSIX compliant return value convention. Tru64 is the only one to use the (-1,errno) return value.
I've changed the __xpg_strerror_r implementations.
See also http://austingroupbugs.net/view.php?id=655 Looks like strerror_r is being retired because of the confusion.
*** Bug 260998 has been marked as a duplicate of this bug. *** Seen from the domain http://volichat.com Page where seen: http://volichat.com/adult-chat-rooms Marked for reference. Resolved as fixed @bugzilla.