https://bugzilla.kernel.org/show_bug.cgi?id=219847 Bug ID: 219847 Summary: mbsnrtowcs(3) man page behavior with glibc incorrect (and POSIX.1-2024 incompatible) Product: Documentation Version: unspecified Hardware: All OS: Linux Status: NEW Severity: normal Priority: P3 Component: man-pages Assignee: documentation_man-pages@xxxxxxxxxxxxxxxxxxxx Reporter: explorer09@xxxxxxxxx Regression: No mbsnrtowcs(3) man page has a part saying: "According to POSIX.1, if the input buffer ends with an incomplete character, it is unspecified whether conversion stops at the end of the previous character (if any), or at the end of the input buffer. The glibc implementation adopts the former behavior." (https://man7.org/linux/man-pages/man3/mbsnrtowcs.3.html) (Source: https://web.git.kernel.org/pub/scm/docs/man-pages/man-pages.git/tree/man/man3/mbsnrtowcs.3) The problem: It is POSIX.1-2008 and POSIX.1-2017 that leave it unspecified where the conversion stops. POSIX.1-2024 now requires the _latter_ behavior, and the reason they cited about the change is, strangely, glibc. But this man page says that glibc uses the former behavior. (https://pubs.opengroup.org/onlinepubs/9799919799/functions/mbsrtowcs.html) (https://www.austingroupbugs.net/view.php?id=616) For my curiosity, I did test with the code included in the Austin Group Issue report (also pasted below, with my personal modifications), in Devuan GNU/Linux 5 (glibc 2.36-9+deb12u9). Glibc's behavior is close to the latter, but I would rather like to clarify the behavior as follows: "If the input buffer (up to the `nmc` limit) ends with an incomplete character, conversion stops at the `nmc` byte index of the input buffer. However, if a null byte ('\0') is encountered in the input buffer before the `nmc` limit, then the incomplete sequence is treated as invalid instead, and `*src` would point to the start of that invalid byte sequence." (The behavior of treating the incomplete sequence before '\0' makes the behavior of `mbsnrtowcs(dest, src, SIZE_MAX, size, ps)` identical to `mbsrtowcs(dest, src, size, ps)` so mbsrtowcs(3) can be directly implemented using mbsnrtowcs(3).) My wording isn't great, so please revise the wording when you can. ```c #include <wchar.h> #include <locale.h> #include <string.h> #include <stdio.h> wchar_t wcs[100]; char mbs[100]; int main() { mbstate_t state; const char *s; setlocale(LC_CTYPE, "en_US.UTF-8"); // U+754C U+7DDA memset(&state, 0, sizeof(state)); memcpy(mbs, "\xe7\x95\x8c\xe7\xb7\x9a", 7); s = mbs; printf("%u ", (unsigned)mbsnrtowcs(wcs, &s, 5, 100, &state)); printf("%u\n", (unsigned)(s - mbs)); // Output: "1 5" // (If conversion stops at character boundary, the output would be "1 3".) memset(&state, 0, sizeof(state)); memcpy(mbs, "\xe7\x95\x8c\xe7\xb7", 6); s = mbs; printf("%u ", (unsigned)mbsnrtowcs(wcs, &s, 6, 100, &state)); printf("%u\n", (unsigned)(s - mbs)); // Output: "4294967295 3" } ``` -- You may reply to this email to add a comment. You are receiving this mail because: You are watching the assignee of the bug.
