wcscmp: Work around two ISO C compliance bugs on several platforms.

* lib/wchar.in.h (wcscmp): Consider REPLACE_WCSCMP.
* lib/wcscmp-impl.h (wcscmp): Don't assume that the two wide characters
are in the range 0..INT_MAX.
* m4/wcscmp.m4 (gl_FUNC_WCSCMP): Test whether wcscmp works for all wide
characters. Set REPLACE_WCSCMP.
* m4/wchar_h.m4 (gl_WCHAR_H_DEFAULTS): Initialize REPLACE_WCSCMP.
* modules/wchar (Makefile.am): Substitute REPLACE_WCSCMP.
* modules/wcscmp (Status, Notice): Un-obsolete this module.
(configure.ac): Consider REPLACE_WCSCMP.
* doc/posix-functions/wcscmp.texi: Mention the two bugs.

8 files changed, 99 insertions, 13 deletions

diff --git a/ChangeLog b/ChangeLog
index 17596c4b23..bc02f2d5f5 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,19 @@
 2023-04-18  Bruno Haible  <bruno@clisp.org>
 
+	wcscmp: Work around two ISO C compliance bugs on several platforms.
+	* lib/wchar.in.h (wcscmp): Consider REPLACE_WCSCMP.
+	* lib/wcscmp-impl.h (wcscmp): Don't assume that the two wide characters
+	are in the range 0..INT_MAX.
+	* m4/wcscmp.m4 (gl_FUNC_WCSCMP): Test whether wcscmp works for all wide
+	characters. Set REPLACE_WCSCMP.
+	* m4/wchar_h.m4 (gl_WCHAR_H_DEFAULTS): Initialize REPLACE_WCSCMP.
+	* modules/wchar (Makefile.am): Substitute REPLACE_WCSCMP.
+	* modules/wcscmp (Status, Notice): Un-obsolete this module.
+	(configure.ac): Consider REPLACE_WCSCMP.
+	* doc/posix-functions/wcscmp.texi: Mention the two bugs.
+
+2023-04-18  Bruno Haible  <bruno@clisp.org>
+
 	wmemcmp: Add tests.
 	* tests/test-wmemcmp.c: New file, based on tests/unistr/test-cmp.h.
 	* modules/wmemcmp-tests: New file.
diff --git a/doc/posix-functions/wcscmp.texi b/doc/posix-functions/wcscmp.texi
index 4c4de8c6dc..bc64d28f56 100644
--- a/doc/posix-functions/wcscmp.texi
+++ b/doc/posix-functions/wcscmp.texi
@@ -8,6 +8,14 @@ Gnulib module: wcscmp
 
 Portability problems fixed by Gnulib:
 @itemize
+@item
+This function compares the wide characters as if they were unsigned, although
+@code{wchar_t} is signed, on some platforms:
+glibc 2.14.1 on x86 or x86_64, musl libc 1.2.3, macOS 12.5, FreeBSD 13.2, NetBSD 9.0, OpenBSD 7.2, Solaris 11.4.
+@item
+This function may return a wrong result if the two arguments are of different
+length, on some platforms:
+AIX 7.2 in 64-bit mode.
 @end itemize
 
 Portability problems not fixed by Gnulib:
diff --git a/lib/wchar.in.h b/lib/wchar.in.h
index 6a5b18d39d..c347256368 100644
--- a/lib/wchar.in.h
+++ b/lib/wchar.in.h
@@ -938,11 +938,21 @@ _GL_WARN_ON_USE (wcsncat, "wcsncat is unportable - "
 
 /* Compare S1 and S2.  */
 #if @GNULIB_WCSCMP@
-# if !@HAVE_WCSCMP@
+# if @REPLACE_WCSCMP@
+#  if !(defined __cplusplus && defined GNULIB_NAMESPACE)
+#   undef wcscmp
+#   define wcscmp rpl_wcscmp
+#  endif
+_GL_FUNCDECL_RPL (wcscmp, int, (const wchar_t *s1, const wchar_t *s2)
+                               _GL_ATTRIBUTE_PURE);
+_GL_CXXALIAS_RPL (wcscmp, int, (const wchar_t *s1, const wchar_t *s2));
+# else
+#  if !@HAVE_WCSCMP@
 _GL_FUNCDECL_SYS (wcscmp, int, (const wchar_t *s1, const wchar_t *s2)
                                _GL_ATTRIBUTE_PURE);
-# endif
+#  endif
 _GL_CXXALIAS_SYS (wcscmp, int, (const wchar_t *s1, const wchar_t *s2));
+# endif
 # if __GLIBC__ >= 2
 _GL_CXXALIASWARN (wcscmp);
 # endif
diff --git a/lib/wcscmp-impl.h b/lib/wcscmp-impl.h
index ba96db4f77..bc17c75d94 100644
--- a/lib/wcscmp-impl.h
+++ b/lib/wcscmp-impl.h
@@ -24,8 +24,9 @@ wcscmp (const wchar_t *s1, const wchar_t *s2)
       wchar_t wc2 = *s2++;
       if (wc1 != (wchar_t)'\0' && wc1 == wc2)
         continue;
-      /* Note that wc1 and wc2 each have at most 31 bits.  */
-      return (int)wc1 - (int)wc2;
+      /* ISO C requires wcscmp to work with all wchar_t values.
+         We cannot assume that wc1 and wc2 are in the range 0..INT_MAX.  */
+      return _GL_CMP (wc1, wc2);
              /* > 0 if wc1 > wc2, < 0 if wc1 < wc2,
                 = 0 if wc1 and wc2 are both '\0'.  */
     }
diff --git a/m4/wchar_h.m4 b/m4/wchar_h.m4
index dfd154f8e9..b9fa7cec84 100644
--- a/m4/wchar_h.m4
+++ b/m4/wchar_h.m4
@@ -7,7 +7,7 @@ dnl with or without modifications, as long as this notice is preserved.
 
 dnl Written by Eric Blake.
 
-# wchar_h.m4 serial 58
+# wchar_h.m4 serial 59
 
 AC_DEFUN_ONCE([gl_WCHAR_H],
 [
@@ -253,6 +253,7 @@ AC_DEFUN([gl_WCHAR_H_DEFAULTS],
   REPLACE_WCWIDTH=0;    AC_SUBST([REPLACE_WCWIDTH])
   REPLACE_WCSWIDTH=0;   AC_SUBST([REPLACE_WCSWIDTH])
   REPLACE_WCSFTIME=0;   AC_SUBST([REPLACE_WCSFTIME])
+  REPLACE_WCSCMP=0;     AC_SUBST([REPLACE_WCSCMP])
   REPLACE_WCSSTR=0;     AC_SUBST([REPLACE_WCSSTR])
   REPLACE_WCSTOK=0;     AC_SUBST([REPLACE_WCSTOK])
   REPLACE_WMEMCMP=0;    AC_SUBST([REPLACE_WMEMCMP])
diff --git a/m4/wcscmp.m4 b/m4/wcscmp.m4
index 61fd800e4d..a3000ed275 100644
--- a/m4/wcscmp.m4
+++ b/m4/wcscmp.m4
@@ -1,4 +1,4 @@
-# wcscmp.m4 serial 2
+# wcscmp.m4 serial 3
 dnl Copyright (C) 2011-2023 Free Software Foundation, Inc.
 dnl This file is free software; the Free Software Foundation
 dnl gives unlimited permission to copy and/or distribute it,
@@ -10,5 +10,61 @@ AC_DEFUN([gl_FUNC_WCSCMP],
   AC_CHECK_FUNCS_ONCE([wcscmp])
   if test $ac_cv_func_wcscmp = no; then
     HAVE_WCSCMP=0
+  else
+    AC_CACHE_CHECK([whether wcscmp works for all wide characters],
+      [gl_cv_func_wcscmp_works],
+      [AC_RUN_IFELSE(
+         [AC_LANG_SOURCE([[
+            #include <wchar.h>
+            int main ()
+            {
+              int result = 0;
+              { /* This test fails on glibc < 2.15, musl libc 1.2.3, macOS 12.5,
+                   FreeBSD 13.2, NetBSD 9.0, OpenBSD 7.2, Solaris 11.4.  */
+                wchar_t a[2] = { (wchar_t) 0x76547654, 0 };
+                wchar_t b[2] = { (wchar_t) 0x9abc9abc, 0 };
+                int cmp = wcscmp (a, b);
+                if (!((wchar_t)-1 < 0 ? cmp > 0 : cmp < 0))
+                  result |= 1;
+              }
+              { /* This test fails on AIX in 64-bit mode.  */
+                wchar_t c[2] = { (wchar_t) 'x', 0 };
+                wchar_t d[3] = { (wchar_t) 'x', (wchar_t) 0x9abc9abc, 0 };
+                int cmp = wcscmp (c, d);
+                if (!((wchar_t)-1 < 0 ? cmp > 0 : cmp < 0))
+                  result |= 2;
+              }
+              return result;
+            }
+            ]])
+         ],
+         [gl_cv_func_wcscmp_works=yes],
+         [gl_cv_func_wcscmp_works=no],
+         [case "$host_on" in
+            # Guess no on glibc versions < 2.15.
+            *-gnu* | gnu*)
+              AC_EGREP_CPP([Unlucky],
+                [
+#include <features.h>
+#ifdef __GNU_LIBRARY__
+ #if (__GLIBC__ == 2 && __GLIBC_MINOR__ < 15)
+  Unlucky GNU user
+ #endif
+#endif
+                ],
+                [gl_cv_func_wcscmp_works="guessing no"],
+                [gl_cv_func_wcscmp_works="guessing yes"])
+              ;;
+            # Guess no on musl systems.
+            *-musl* | midipix*) gl_cv_func_wcscmp_works="guessing no" ;;
+            # If we don't know, obey --enable-cross-guesses.
+            *) gl_cv_func_wcscmp_works="$gl_cross_guess_normal" ;;
+          esac
+         ])
+      ])
+    case "$gl_cv_func_wcscmp_works" in
+      *yes) ;;
+      *) REPLACE_WCSCMP=1 ;;
+    esac
   fi
 ])
diff --git a/modules/wchar b/modules/wchar
index 180c94309c..88b442b525 100644
--- a/modules/wchar
+++ b/modules/wchar
@@ -142,6 +142,7 @@ wchar.h: wchar.in.h $(top_builddir)/config.status $(CXXDEFS_H) $(ARG_NONNULL_H)
 	      -e 's|@''REPLACE_WCWIDTH''@|$(REPLACE_WCWIDTH)|g' \
 	      -e 's|@''REPLACE_WCSWIDTH''@|$(REPLACE_WCSWIDTH)|g' \
 	      -e 's|@''REPLACE_WCSFTIME''@|$(REPLACE_WCSFTIME)|g' \
+	      -e 's|@''REPLACE_WCSCMP''@|$(REPLACE_WCSCMP)|g' \
 	      -e 's|@''REPLACE_WCSSTR''@|$(REPLACE_WCSSTR)|g' \
 	      -e 's|@''REPLACE_WCSTOK''@|$(REPLACE_WCSTOK)|g' \
 	      -e 's|@''REPLACE_WMEMCMP''@|$(REPLACE_WMEMCMP)|g' \
diff --git a/modules/wcscmp b/modules/wcscmp
index 1b4fde704a..6df70f0543 100644
--- a/modules/wcscmp
+++ b/modules/wcscmp
@@ -1,12 +1,6 @@
 Description:
 wcscmp() function: compare two wide strings.
 
-Status:
-obsolete
-
-Notice:
-This module is obsolete.
-
 Files:
 lib/wcscmp.c
 lib/wcscmp-impl.h
@@ -17,7 +11,8 @@ wchar
 
 configure.ac:
 gl_FUNC_WCSCMP
-gl_CONDITIONAL([GL_COND_OBJ_WCSCMP], [test $HAVE_WCSCMP = 0])
+gl_CONDITIONAL([GL_COND_OBJ_WCSCMP],
+               [test $HAVE_WCSCMP = 0 || test $REPLACE_WCSCMP = 1])
 gl_WCHAR_MODULE_INDICATOR([wcscmp])
 
 Makefile.am: