Update.

	* string/bits/string2.h: Fix aliasing problems.
	* sysdeps/i386/i486/bits/string.h: Likewise.
	* sysdeps/i386/bits/string.h: Likewise.

1998-12-28  Geoff Keating  <geoffk@ozemail.com.au>

	* sysdeps/powerpc/dl-machine.c: Handle shared library profiling.
	* elf/rtld.c (dl_main): Don't call malloc() between
	re-relocating the dynamic linker and running the application's crt0.

1999-07-21  Paul D. Smith  <psmith@gnu.org>

	* sysdeps/generic/glob.c: Move getlogin{,_r} prototypes below
	glob.h to get __P() macro.

	* posix/fnmatch.c (internal_fnmatch): Use K&R definition syntax,
	not ANSI.
	(__strchrnul): This won't exist outside GLIBC, so create one.

1999-07-25  Jakub Jelinek   <jj@ultra.linux.cz>

	* sysdeps/unix/sysv/linux/sparc/bits/types.h: Always define __qaddr_t.
	__ino64_t should be 32bit unsigned type on sparc32.
	Define __off64_t to __quad_t instead of __int64_t.
	Make __pic_pid_t unsigned on sparc32.

1999-07-25  Andreas Jaeger  <aj@arthur.rhein-neckar.de>

	* localedata/tst-rpmatch.sh: Use "&" instead of "§" to avoid a bug
	in bash 2.03.

1999-07-25  Ulrich Drepper  <drepper@cygnus.com>

8 files changed, 417 insertions, 240 deletions

diff --git a/ChangeLog b/ChangeLog
index 72c8ad625f..0097e04a4f 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,38 @@
 1999-07-25  Ulrich Drepper  <drepper@cygnus.com>
 
+	* string/bits/string2.h: Fix aliasing problems.
+	* sysdeps/i386/i486/bits/string.h: Likewise.
+	* sysdeps/i386/bits/string.h: Likewise.
+
+1998-12-28  Geoff Keating  <geoffk@ozemail.com.au>
+
+	* sysdeps/powerpc/dl-machine.c: Handle shared library profiling.
+	* elf/rtld.c (dl_main): Don't call malloc() between
+	re-relocating the dynamic linker and running the application's crt0.
+
+1999-07-21  Paul D. Smith  <psmith@gnu.org>
+
+	* sysdeps/generic/glob.c: Move getlogin{,_r} prototypes below
+	glob.h to get __P() macro.
+
+	* posix/fnmatch.c (internal_fnmatch): Use K&R definition syntax,
+	not ANSI.
+	(__strchrnul): This won't exist outside GLIBC, so create one.
+
+1999-07-25  Jakub Jelinek   <jj@ultra.linux.cz>
+
+	* sysdeps/unix/sysv/linux/sparc/bits/types.h: Always define __qaddr_t.
+	__ino64_t should be 32bit unsigned type on sparc32.
+	Define __off64_t to __quad_t instead of __int64_t.
+	Make __pic_pid_t unsigned on sparc32.
+
+1999-07-25  Andreas Jaeger  <aj@arthur.rhein-neckar.de>
+
+	* localedata/tst-rpmatch.sh: Use "&" instead of "§" to avoid a bug
+	in bash 2.03.
+
+1999-07-25  Ulrich Drepper  <drepper@cygnus.com>
+
 	* iconvdata/euc-kr.c (euckr_from_ucs4): Set cp[0] to '\0' in case
 	of an error.  Patch by Changwoo Ryu <cwryu@debian.org>.
 
diff --git a/elf/rtld.c b/elf/rtld.c
index 60293800c5..72c81fb00f 100644
--- a/elf/rtld.c
+++ b/elf/rtld.c
@@ -1024,6 +1024,14 @@ of this helper program; chances are you did not intend to run this program.\n\
        _dl_relocate_object might need to call `mprotect' for DT_TEXTREL.  */
     _dl_sysdep_start_cleanup ();
 
+    /* Now enable profiling if needed.  Like the previous call,
+       this has to go here because the calls it makes should use the
+       rtld versions of the functions (particularly calloc()), but it
+       needs to have _dl_profile_map set up by the relocator.  */
+    if (_dl_profile_map != NULL)
+      /* We must prepare the profiling.  */
+      _dl_start_profile (_dl_profile_map, _dl_profile_output);
+
     if (_dl_rtld_map.l_opencount > 0)
       {
 	/* There was an explicit ref to the dynamic linker as a shared lib.
@@ -1082,11 +1090,6 @@ of this helper program; chances are you did not intend to run this program.\n\
   _dl_unload_cache ();
 #endif
 
-  /* Now enable profiling if needed.  */
-  if (_dl_profile_map != NULL)
-    /* We must prepare the profiling.  */
-    _dl_start_profile (_dl_profile_map, _dl_profile_output);
-
   /* Once we return, _dl_sysdep_start will invoke
      the DT_INIT functions and then *USER_ENTRY.  */
 }
@@ -1299,11 +1302,7 @@ process_envvars (enum mode *modep, int *lazyp)
 
 	  /* Which shared object shall be profiled.  */
 	  if (memcmp (&envline[3], "PROFILE", 7) == 0)
-	    {
-	      _dl_profile = &envline[11];
-	      if (*_dl_profile == '\0')
-		_dl_profile = NULL;
-	    }
+	    _dl_profile = &envline[11];
 	  break;
 
 	case 8:
@@ -1388,6 +1387,12 @@ process_envvars (enum mode *modep, int *lazyp)
 	unsetenv ("LD_PRELOAD");
       if (library_path != NULL)
 	unsetenv ("LD_LIBRARY_PATH");
+      if (_dl_origin_path != NULL)
+	unsetenv ("LD_ORIGIN_PATH");
+      if (debug_output != NULL)
+	unsetenv ("LD_DEBUG_OUTPUT");
+      if (_dl_profile != NULL)
+	unsetenv ("LD_PROFILE");
 
       for (cnt = 0;
 	   cnt < sizeof (unsecure_envvars) / sizeof (unsecure_envvars[0]);
@@ -1395,6 +1400,10 @@ process_envvars (enum mode *modep, int *lazyp)
 	unsetenv (unsecure_envvars[cnt]);
     }
 
+  /* The name of the object to profile cannot be empty.  */
+  if (_dl_profile != NULL && *_dl_profile == '\0')
+    _dl_profile = NULL;
+
   /* If we have to run the dynamic linker in debugging mode and the
      LD_DEBUG_OUTPUT environment variable is given, we write the debug
      messages to this file.  */
diff --git a/posix/fnmatch.c b/posix/fnmatch.c
index 9617d336f4..b1e1b94afa 100644
--- a/posix/fnmatch.c
+++ b/posix/fnmatch.c
@@ -127,14 +127,35 @@ extern char *getenv ();
 extern int errno;
 # endif
 
+/* This function doesn't exist on most systems.  */
+
+# if !defined HAVE___STRCHRNUL && !defined _LIBC
+static char *
+__strchrnul (s, c)
+     const char *s;
+     int c;
+{
+  char *result = strchr (s, c);
+  if (result == NULL)
+    result = strchr (s, '\0');
+  return result;
+}
+# endif
+
 /* Match STRING against the filename pattern PATTERN, returning zero if
    it matches, nonzero if not.  */
+static int internal_fnmatch __P ((const char *pattern, const char *string,
+				  int no_leading_period, int flags))
+     internal_function;
 static int
 #ifdef _LIBC
 internal_function
 #endif
-internal_fnmatch (const char *pattern, const char *string,
-		  int no_leading_period, int flags)
+internal_fnmatch (pattern, string, no_leading_period, flags)
+     const char *pattern;
+     const char *string;
+     int no_leading_period;
+     int flags;
 {
   register const char *p = pattern, *n = string;
   register unsigned char c;
diff --git a/string/bits/string2.h b/string/bits/string2.h
index 43cb6d5eff..f27ea6269d 100644
--- a/string/bits/string2.h
+++ b/string/bits/string2.h
@@ -110,51 +110,68 @@ __STRING2_COPY_TYPE (8);
 
 #  define __memset_gc(s, c, n) \
   ({ void *__s = (s);							      \
-     __uint32_t *__ts = (__uint32_t *) __s;				      \
+     union {								      \
+       unsigned int __ui;						      \
+       unsigned short int __usi;					      \
+       unsigned char __uc;						      \
+     } *__u = __s;							      \
      __uint8_t __c = (__uint8_t) (c);					      \
-     									      \
+									      \
      /* This `switch' statement will be removed at compile-time.  */	      \
      switch (n)								      \
        {								      \
        case 15:								      \
-	 *__ts++ = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
+	 __u = (void *) __u + 4;					      \
        case 11:								      \
-	 *__ts++ = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
+	 __u = (void *) __u + 4;					      \
        case 7:								      \
-	 *__ts++ = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
+	 __u = (void *) __u + 4;					      \
        case 3:								      \
-	 *((__uint16_t *) __ts)++ = __c * 0x0101;			      \
-	 *((__uint8_t *) __ts) = __c;					      \
+	 __u->__usi = (unsigned short int) __c * 0x0101;		      \
+	 __u = (void *) __u + 2;					      \
+	 __u->__uc = (unsigned char) __c;				      \
 	 break;								      \
 									      \
        case 14:								      \
-	 *__ts++ = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
+	 __u = (void *) __u + 4;					      \
        case 10:								      \
-	 *__ts++ = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
+	 __u = (void *) __u + 4;					      \
        case 6:								      \
-	 *__ts++ = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
+	 __u = (void *) __u + 4;					      \
        case 2:								      \
-	 *((__uint16_t *) __ts) = __c * 0x0101;				      \
+	 __u->__usi = (unsigned short int) __c * 0x0101;		      \
 	 break;								      \
 									      \
        case 13:								      \
-	 *__ts++ = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
+	 __u = (void *) __u + 4;					      \
        case 9:								      \
-	 *__ts++ = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
+	 __u = (void *) __u + 4;					      \
        case 5:								      \
-	 *__ts++ = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
+	 __u = (void *) __u + 4;					      \
        case 1:								      \
-	 *((__uint8_t *) __ts) = __c;					      \
+	 __u->__uc = (unsigned char) __c;				      \
 	 break;								      \
 									      \
        case 16:								      \
-	 *__ts++ = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
+	 __u = (void *) __u + 4;					      \
        case 12:								      \
-	 *__ts++ = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
+	 __u = (void *) __u + 4;					      \
        case 8:								      \
-	 *__ts++ = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
+	 __u = (void *) __u + 4;					      \
        case 4:								      \
-	 *__ts = __c * 0x01010101;					      \
+	 __u->__ui = __c * 0x01010101;					      \
        case 0:								      \
 	 break;								      \
        }								      \
@@ -169,7 +186,7 @@ __STRING2_COPY_TYPE (8);
 
 /* GCC optimizes memset(s, 0, n) but not bzero(s, n).
    The optimization is broken before EGCS 1.1.  */
-#if __GNUC_PREREQ (2, 91)
+# if __GNUC_PREREQ (2, 91)
 #  define __bzero(s, n) __builtin_memset (s, '\0', n)
 # endif
 
@@ -207,48 +224,60 @@ __mempcpy_small (void *__dest1,
 		 __uint32_t __src0_4, __uint32_t __src4_4,
 		 size_t __srclen)
 {
-  char *__dest = (char *) __dest1;
+  union {
+    __uint32_t __ui;
+    __uint16_t __usi;
+    unsigned char __uc;
+    unsigned char __c;
+  } *__u = __dest1;
   switch (__srclen)
     {
     case 1:
-      *__dest++ = __src0_1;
+      __u->__c = __src0_1;
+      __u = (void *) __u + 1;
       break;
     case 2:
-      *((__uint16_t *) __dest) = __src0_2;
-      __dest += 2;
+      __u->__usi = __src0_2;
+      __u = (void *) __u + 2;
       break;
     case 3:
-      *((__uint16_t *) __dest) = __src0_2;
-      __dest += 2;
-      *__dest++ = __src2_1;
+      __u->__usi = __src0_2;
+      __u = (void *) __u + 2;
+      __u->__c = __src2_1;
+      __u = (void *) __u + 1;
       break;
     case 4:
-      *((__uint32_t *) __dest) = __src0_4;
-      __dest += 4;
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
       break;
     case 5:
-      *((__uint32_t *) __dest) = __src0_4;
-      __dest += 4;
-      *__dest++ = __src4_1;
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
+      __u->__c = __src4_1;
+      __u = (void *) __u + 1;
       break;
     case 6:
-      *((__uint32_t *) __dest) = __src0_4;
-      *((__uint16_t *) (__dest + 4)) = __src4_2;
-      __dest += 6;
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
+      __u->__usi = __src4_2;
+      __u = (void *) __u + 2;
       break;
     case 7:
-      *((__uint32_t *) __dest) = __src0_4;
-      *((__uint16_t *) (__dest + 4)) = __src4_2;
-      __dest += 6;
-      *__dest++ = __src6_1;
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
+      __u->__usi = __src4_2;
+      __u = (void *) __u + 2;
+      __u->__c = __src6_1;
+      __u = (void *) __u + 1;
       break;
     case 8:
-      *((__uint32_t *) __dest) = __src0_4;
-      *((__uint32_t *) (__dest + 4)) = __src4_4;
-      __dest += 8;
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
+      __u->__ui = __src4_4;
+      __u = (void *) __u + 4;
       break;
     }
-  return (void *) __dest;
+  return (void *) __u;
 }
 #  else
 #   define __mempcpy_args(src) \
@@ -287,41 +316,50 @@ __STRING_INLINE void *__mempcpy_small (void *, char, __STRING2_COPY_ARR2,
 				       __STRING2_COPY_ARR7,
 				       __STRING2_COPY_ARR8, size_t);
 __STRING_INLINE void *
-__mempcpy_small (void *__dest1, char __src1,
+__mempcpy_small (void *__dest, char __src1,
 		 __STRING2_COPY_ARR2 __src2, __STRING2_COPY_ARR3 __src3,
 		 __STRING2_COPY_ARR4 __src4, __STRING2_COPY_ARR5 __src5,
 		 __STRING2_COPY_ARR6 __src6, __STRING2_COPY_ARR7 __src7,
 		 __STRING2_COPY_ARR8 __src8, size_t __srclen)
 {
-  char *__dest = (char *) __dest1;
+  union {
+    char __c;
+    __STRING2_COPY_ARR2 __sca2;
+    __STRING2_COPY_ARR3 __sca3;
+    __STRING2_COPY_ARR4 __sca4;
+    __STRING2_COPY_ARR5 __sca5;
+    __STRING2_COPY_ARR6 __sca6;
+    __STRING2_COPY_ARR7 __sca7;
+    __STRING2_COPY_ARR8 __sca8;
+  } *__u = __dest;
   switch (__srclen)
     {
     case 1:
-      *__dest = __src1;
+      __u->__c = __src1;
       break;
     case 2:
-      __extension__ *((__STRING2_COPY_ARR2 *) __dest) = __src2;
+      __extension__ __u->__sca2 = __src2;
       break;
     case 3:
-      __extension__ *((__STRING2_COPY_ARR3 *) __dest) = __src3;
+      __extension__ __u->__sca3 = __src3;
       break;
     case 4:
-      __extension__ *((__STRING2_COPY_ARR4 *) __dest) = __src4;
+      __extension__ __u->__sca4 = __src4;
       break;
     case 5:
-      __extension__ *((__STRING2_COPY_ARR5 *) __dest) = __src5;
+      __extension__ __u->__sca5 = __src5;
       break;
     case 6:
-      __extension__ *((__STRING2_COPY_ARR6 *) __dest) = __src6;
+      __extension__ __u->__sca6 = __src6;
       break;
     case 7:
-      __extension__ *((__STRING2_COPY_ARR7 *) __dest) = __src7;
+      __extension__ __u->__sca7 = __src7;
       break;
     case 8:
-      __extension__ *((__STRING2_COPY_ARR8 *) __dest) = __src8;
+      __extension__ __u->__sca8 = __src8;
       break;
     }
-  return (void *) (__dest + __srclen);
+  return __extension__ ((void *) __u + __srclen);
 }
 #  endif
 # endif
@@ -362,37 +400,48 @@ __strcpy_small (char *__dest,
 		__uint32_t __src0_4, __uint32_t __src4_4,
 		size_t __srclen)
 {
+  union {
+    __uint32_t __ui;
+    __uint16_t __usi;
+    unsigned char __uc;
+  } *__u = (void *) __dest;
   switch (__srclen)
     {
     case 1:
-      *__dest = '\0';
+      __u->__uc = '\0';
       break;
     case 2:
-      *((__uint16_t *) __dest) = __src0_2;
+      __u->__usi = __src0_2;
       break;
     case 3:
-      *((__uint16_t *) __dest) = __src0_2;
-      *(__dest + 2) = '\0';
+      __u->__usi = __src0_2;
+      __u = (void *) __u + 2;
+      __u->__uc = '\0';
       break;
     case 4:
-      *((__uint32_t *) __dest) = __src0_4;
+      __u->__ui = __src0_4;
       break;
     case 5:
-      *((__uint32_t *) __dest) = __src0_4;
-      *(__dest + 4) = '\0';
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
+      __u->__uc = '\0';
       break;
     case 6:
-      *((__uint32_t *) __dest) = __src0_4;
-      *((__uint16_t *) (__dest + 4)) = __src4_2;
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
+      __u->__usi = __src4_2;
       break;
     case 7:
-      *((__uint32_t *) __dest) = __src0_4;
-      *((__uint16_t *) (__dest + 4)) = __src4_2;
-      *(__dest + 6) = '\0';
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
+      __u->__usi = __src4_2;
+      __u = (void *) __u + 2;
+      __u->__uc = '\0';
       break;
     case 8:
-      *((__uint32_t *) __dest) = __src0_4;
-      *((__uint32_t *) (__dest + 4)) = __src4_4;
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
+      __u->__usi = __src4_4;
       break;
     }
   return __dest;
@@ -439,31 +488,41 @@ __strcpy_small (char *__dest,
 		__STRING2_COPY_ARR6 __src6, __STRING2_COPY_ARR7 __src7,
 		__STRING2_COPY_ARR8 __src8, size_t __srclen)
 {
+  union {
+    char __c;
+    __STRING2_COPY_ARR2 __sca2;
+    __STRING2_COPY_ARR3 __sca3;
+    __STRING2_COPY_ARR4 __sca4;
+    __STRING2_COPY_ARR5 __sca5;
+    __STRING2_COPY_ARR6 __sca6;
+    __STRING2_COPY_ARR7 __sca7;
+    __STRING2_COPY_ARR8 __sca8;
+  } *__u = __dest;
   switch (__srclen)
     {
     case 1:
-      *__dest = '\0';
+      __u->__c = '\0';
       break;
     case 2:
-      __extension__ *((__STRING2_COPY_ARR2 *) __dest) = __src2;
+      __extension__ __u->__sca2 = __src2;
       break;
     case 3:
-      __extension__ *((__STRING2_COPY_ARR3 *) __dest) = __src3;
+      __extension__ __u->__sca3 = __src3;
       break;
     case 4:
-      __extension__ *((__STRING2_COPY_ARR4 *) __dest) = __src4;
+      __extension__ __u->__sca4 = __src4;
       break;
     case 5:
-      __extension__ *((__STRING2_COPY_ARR5 *) __dest) = __src5;
+      __extension__ __u->__sca5 = __src5;
       break;
     case 6:
-      __extension__ *((__STRING2_COPY_ARR6 *) __dest) = __src6;
+      __extension__ __u->__sca6 = __src6;
       break;
     case 7:
-      __extension__ *((__STRING2_COPY_ARR7 *) __dest) = __src7;
+      __extension__ __u->__sca7 = __src7;
       break;
     case 8:
-      __extension__ *((__STRING2_COPY_ARR8 *) __dest) = __src8;
+      __extension__ __u->__sca8 = __src8;
       break;
   }
   return __dest;
@@ -500,47 +559,55 @@ __stpcpy_small (char *__dest,
 		__uint32_t __src0_4, __uint32_t __src4_4,
 		size_t __srclen)
 {
+  union {
+    unsigned int __ui;
+    unsigned short int __usi;
+    unsigned char __uc;
+  } *__u = (void *) __dest;
   switch (__srclen)
     {
     case 1:
-      *__dest = '\0';
+      __u->__uc = '\0';
       break;
     case 2:
-      *((__uint16_t *) __dest) = __src0_2;
-      ++__dest;
+      __u->__usi = __src0_2;
+      __u = (void *) __u + 1;
       break;
     case 3:
-      *((__uint16_t *) __dest) = __src0_2;
-      __dest += sizeof (__uint16_t);
-      *__dest = '\0';
+      __u->__usi = __src0_2;
+      __u = (void *) __u + 2;
+      __u->__uc = '\0';
       break;
     case 4:
-      *((__uint32_t *) __dest) = __src0_4;
-      __dest += 3;
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 3;
       break;
     case 5:
-      *((__uint32_t *) __dest) = __src0_4;
-      __dest += 4;
-      *__dest = '\0';
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
+      __u->__uc = '\0';
       break;
     case 6:
-      *((__uint32_t *) __dest) = __src0_4;
-      *((__uint16_t *) (__dest + 4)) = __src4_2;
-      __dest += 5;
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
+      __u->__usi = __src4_2;
+      __u = (void *) __u + 1;
       break;
     case 7:
-      *((__uint32_t *) __dest) = __src0_4;
-      *((__uint16_t *) (__dest + 4)) = __src4_2;
-      __dest += 6;
-      *__dest = '\0';
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
+      __u->__usi = __src4_2;
+      __u = (void *) __u + 2;
+      __u->__uc = '\0';
       break;
     case 8:
-      *((__uint32_t *) __dest) = __src0_4;
-      *((__uint32_t *) (__dest + 4)) = __src4_4;
-      __dest += 7;
+      __u->__ui = __src0_4;
+      __u = (void *) __u + 4;
+      __u->__ui = __src4_4;
+      __u = (void *) __u + 3;
       break;
     }
-  return __dest;
+  return &__u->__uc;
 }
 #  else
 #  define __stpcpy_args(src) \
@@ -584,31 +651,41 @@ __stpcpy_small (char *__dest,
 		__STRING2_COPY_ARR6 __src6, __STRING2_COPY_ARR7 __src7,
 		__STRING2_COPY_ARR8 __src8, size_t __srclen)
 {
+  union {
+    char __c;
+    __STRING2_COPY_ARR2 __sca2;
+    __STRING2_COPY_ARR3 __sca3;
+    __STRING2_COPY_ARR4 __sca4;
+    __STRING2_COPY_ARR5 __sca5;
+    __STRING2_COPY_ARR6 __sca6;
+    __STRING2_COPY_ARR7 __sca7;
+    __STRING2_COPY_ARR8 __sca8;
+  } *__u = __dest;
   switch (__srclen)
     {
     case 1:
-      *__dest = '\0';
+      __u->__uc = '\0';
       break;
     case 2:
-      __extension__ *((__STRING2_COPY_ARR2 *) __dest) = __src2;
+      __extension__ __u->__sca2 = __src2;
       break;
     case 3:
-      __extension__ *((__STRING2_COPY_ARR3 *) __dest) = __src3;
+      __extension__ __u->__sca3 = __src3;
       break;
     case 4:
-      __extension__ *((__STRING2_COPY_ARR4 *) __dest) = __src4;
+      __extension__ __u->__sca4 = __src4;
       break;
     case 5:
-      __extension__ *((__STRING2_COPY_ARR5 *) __dest) = __src5;
+      __extension__ __u->__sca5 = __src5;
       break;
     case 6:
-      __extension__ *((__STRING2_COPY_ARR6 *) __dest) = __src6;
+      __extension__ __u->__sca6 = __src6;
       break;
     case 7:
-      __extension__ *((__STRING2_COPY_ARR7 *) __dest) = __src7;
+      __extension__ __u->__sca7 = __src7;
       break;
     case 8:
-      __extension__ *((__STRING2_COPY_ARR8 *) __dest) = __src8;
+      __extension__ __u->__sca8 = __src8;
       break;
   }
   return __dest + __srclen - 1;
diff --git a/sysdeps/generic/glob.c b/sysdeps/generic/glob.c
index 9b134c5a0d..1ab5d8b629 100644
--- a/sysdeps/generic/glob.c
+++ b/sysdeps/generic/glob.c
@@ -155,12 +155,6 @@ extern void abort (), exit ();
 
 #endif	/* Standard headers.  */
 
-#ifdef HAVE_GETLOGIN_R
-extern int getlogin_r __P ((char *, size_t));
-#else
-extern char *getlogin __P ((void));
-#endif
-
 #ifndef	ANSI_STRING
 
 # ifndef bzero
@@ -286,6 +280,12 @@ extern char *alloca ();
 # undef	GLOB_PERIOD
 #endif
 #include <glob.h>
+
+#ifdef HAVE_GETLOGIN_R
+extern int getlogin_r __P ((char *, size_t));
+#else
+extern char *getlogin __P ((void));
+#endif
 
 static
 #if __GNUC__ - 0 >= 2
diff --git a/sysdeps/i386/bits/string.h b/sysdeps/i386/bits/string.h
index 1ea0571680..5f4f58a6db 100644
--- a/sysdeps/i386/bits/string.h
+++ b/sysdeps/i386/bits/string.h
@@ -54,59 +54,65 @@ __STRING_INLINE void *
 __memcpy_c (void *__dest, __const void *__src, size_t __n)
 {
   register unsigned long int __d0, __d1, __d2;
+  union {
+    unsigned int __ui;
+    unsigned short int __usi;
+    unsigned char __uc;
+  } *__u = __dest;
   switch (__n)
     {
     case 0:
       return __dest;
     case 1:
-      *(unsigned char *) __dest = *(const unsigned char *) __src;
+      __u->__uc = *(const unsigned char *) __src;
       return __dest;
     case 2:
-      *(unsigned short int *) __dest = *(const unsigned short int *) __src;
+      __u->__usi = *(const unsigned short int *) __src;
       return __dest;
     case 3:
-      *(unsigned short int *) __dest = *(const unsigned short int *) __src;
-      *(2 + (unsigned char *) __dest) = *(2 + (const unsigned char *) __src);
+      __u->__usi = *(const unsigned short int *) __src;
+      __u = (void *) __u + 2;
+      __u->__uc = *(2 + (const unsigned char *) __src);
       return __dest;
     case 4:
-      *(unsigned long int *) __dest = *(const unsigned long int *) __src;
+      __u->__ui = *(const unsigned int *) __src;
       return __dest;
-    case 6:	/* for ethernet addresses */
-      *(unsigned long int *) __dest = *(const unsigned long int *) __src;
-      *(2 + (unsigned short int *) __dest) =
-	*(2 + (const unsigned short int *) __src);
+    case 6:
+      __u->__ui = *(const unsigned int *) __src;
+      __u = (void *) __u + 4;
+      __u->__usi = *(2 + (const unsigned short int *) __src);
       return __dest;
     case 8:
-      *(unsigned long int *) __dest = *(const unsigned long int *) __src;
-      *(1 + (unsigned long int *) __dest) =
-	*(1 + (const unsigned long int *) __src);
+      __u->__ui = *(const unsigned int *) __src;
+      __u = (void *) __u + 4;
+      __u->__ui = *(1 + (const unsigned int *) __src);
       return __dest;
     case 12:
-      *(unsigned long int *) __dest = *(const unsigned long int *) __src;
-      *(1 + (unsigned long int *) __dest) =
-	*(1 + (const unsigned long int *) __src);
-      *(2 + (unsigned long int *) __dest) =
-	*(2 + (const unsigned long int *) __src);
+      __u->__ui = *(const unsigned int *) __src;
+      __u = (void *) __u + 4;
+      __u->__ui = *(1 + (const unsigned int *) __src);
+      __u = (void *) __u + 4;
+      __u->__ui = *(2 + (const unsigned int *) __src);
       return __dest;
     case 16:
-      *(unsigned long int *) __dest = *(const unsigned long int *) __src;
-      *(1 + (unsigned long int *) __dest) =
-	*(1 + (const unsigned long int *) __src);
-      *(2 + (unsigned long int *) __dest) =
-	*(2 + (const unsigned long int *) __src);
-      *(3 + (unsigned long int *) __dest) =
-	*(3 + (const unsigned long int *) __src);
+      __u->__ui = *(const unsigned int *) __src;
+      __u = (void *) __u + 4;
+      __u->__ui = *(1 + (const unsigned int *) __src);
+      __u = (void *) __u + 4;
+      __u->__ui = *(2 + (const unsigned int *) __src);
+      __u = (void *) __u + 4;
+      __u->__ui = *(3 + (const unsigned int *) __src);
       return __dest;
     case 20:
-      *(unsigned long int *) __dest = *(const unsigned long int *) __src;
-      *(1 + (unsigned long int *) __dest) =
-	*(1 + (const unsigned long int *) __src);
-      *(2 + (unsigned long int *) __dest) =
-	*(2 + (const unsigned long int *) __src);
-      *(3 + (unsigned long int *) __dest) =
-	*(3 + (const unsigned long int *) __src);
-      *(4 + (unsigned long int *) __dest) =
-	*(4 + (const unsigned long int *) __src);
+      __u->__ui = *(const unsigned int *) __src;
+      __u = (void *) __u + 4;
+      __u->__ui = *(1 + (const unsigned int *) __src);
+      __u = (void *) __u + 4;
+      __u->__ui = *(2 + (const unsigned int *) __src);
+      __u = (void *) __u + 4;
+      __u->__ui = *(3 + (const unsigned int *) __src);
+      __u = (void *) __u + 4;
+      __u->__ui = *(4 + (const unsigned int *) __src);
       return __dest;
     }
 #define __COMMON_CODE(x) \
@@ -115,7 +121,7 @@ __memcpy_c (void *__dest, __const void *__src, size_t __n)
      "rep; movsl"							      \
      x									      \
      : "=&c" (__d0), "=&D" (__d1), "=&S" (__d2)				      \
-     : "0" (__n / 4), "1" (__dest), "2" (__src)				      \
+     : "0" (__n / 4), "1" (&__u->__uc), "2" (__src)			      \
      : "memory");
 
   switch (__n % 4)
@@ -184,22 +190,28 @@ __STRING_INLINE void *
 __memset_cc (void *__s, unsigned long int __pattern, size_t __n)
 {
   register unsigned long int __d0, __d1;
+  union {
+    unsigned int __ui;
+    unsigned short int __usi;
+    unsigned char __uc;
+  } *__u = __s;
   switch (__n)
     {
     case 0:
       return __s;
     case 1:
-      *(unsigned char *) __s = __pattern;
+      __u->__uc = __pattern;
       return __s;
     case 2:
-      *(unsigned short int *) __s = __pattern;
+      __u->__usi = __pattern;
       return __s;
     case 3:
-      *(unsigned short int *) __s = __pattern;
-      *(2 + (unsigned char *) __s) = __pattern;
+      __u->__usi = __pattern;
+      __u = __extension__ ((void *) __u + 1);
+      __u->__uc = __pattern;
       return __s;
     case 4:
-      *(unsigned long *) __s = __pattern;
+      __u->__ui = __pattern;
       return __s;
 	}
 #define __COMMON_CODE(x) \
@@ -208,7 +220,7 @@ __memset_cc (void *__s, unsigned long int __pattern, size_t __n)
      "rep; stosl"							      \
      x									      \
      : "=&c" (__d0), "=&D" (__d1)					      \
-     : "a" (__pattern), "0" (__n / 4), "1" (__s)			      \
+     : "a" (__pattern), "0" (__n / 4), "1" (&__u->__uc)			      \
      : "memory")
 
   switch (__n % 4)
diff --git a/sysdeps/i386/i486/bits/string.h b/sysdeps/i386/i486/bits/string.h
index 39edc6ec56..8857a7178a 100644
--- a/sysdeps/i386/i486/bits/string.h
+++ b/sysdeps/i386/i486/bits/string.h
@@ -211,12 +211,17 @@ memcmp (__const void *__s1, __const void *__s2, size_t __n)
 
 #define __memset_gc(s, c, n) \
   ({ void *__s = (s);							      \
-     unsigned int *__ts = (unsigned int *) __s;				      \
+     union {								      \
+       unsigned int __ui;						      \
+       unsigned short int __usi;					      \
+       unsigned char __uc;						      \
+     } *__u = __s;							      \
      unsigned int __c = ((unsigned int) ((unsigned char) (c))) * 0x01010101;  \
 									      \
      /* We apply a trick here.  `gcc' would implement the following	      \
-	assignments using absolute operands.  But this uses to much	      \
-	memory (7, instead of 4 bytes).  */				      \
+	assignments using immediate operands.  But this uses to much	      \
+	memory (7, instead of 4 bytes).  So we force the value in a	      \
+	registers.  */							      \
      if (n == 3 || n >= 5)						      \
        __asm__ __volatile__ ("" : "=r" (__c) : "0" (__c));		      \
 									      \
@@ -224,44 +229,57 @@ memcmp (__const void *__s1, __const void *__s2, size_t __n)
      switch (n)								      \
        {								      \
        case 15:								      \
-	 *__ts++ = __c;							      \
+	 __u->__ui = __c;						      \
+	 __u = (void *) __u + 4;					      \
        case 11:								      \
-	 *__ts++ = __c;							      \
+	 __u->__ui = __c;						      \
+	 __u = (void *) __u + 4;					      \
        case 7:								      \
-	 *__ts++ = __c;							      \
+	 __u->__ui = __c;						      \
+	 __u = (void *) __u + 4;					      \
        case 3:								      \
-	 *((unsigned short int *) __ts)++ = (unsigned short int) __c;	      \
-	 *((unsigned char *) __ts) = (unsigned char) __c;		      \
+	 __u->__usi = (unsigned short int) __c;				      \
+	 __u = (void *) __u + 2;					      \
+	 __u->__uc = (unsigned char) __c;				      \
 	 break;								      \
 									      \
        case 14:								      \
-	 *__ts++ = __c;							      \
+	 __u->__ui = __c;						      \
+	 __u = (void *) __u + 4;					      \
        case 10:								      \
-	 *__ts++ = __c;							      \
+	 __u->__ui = __c;						      \
+	 __u = (void *) __u + 4;					      \
        case 6:								      \
-	 *__ts++ = __c;							      \
+	 __u->__ui = __c;						      \
+	 __u = (void *) __u + 4;					      \
        case 2:								      \
-	 *((unsigned short int *) __ts) = (unsigned short int) __c;	      \
+	 __u->__usi = (unsigned short int) __c;				      \
 	 break;								      \
 									      \
        case 13:								      \
-	 *__ts++ = __c;							      \
+	 __u->__ui = __c;						      \
+	 __u = (void *) __u + 4;					      \
        case 9:								      \
-	 *__ts++ = __c;							      \
+	 __u->__ui = __c;						      \
+	 __u = (void *) __u + 4;					      \
        case 5:								      \
-	 *__ts++ = __c;							      \
+	 __u->__ui = __c;						      \
+	 __u = (void *) __u + 4;					      \
        case 1:								      \
-	 *((unsigned char *) __ts) = (unsigned char) __c;		      \
+	 __u->__uc = (unsigned char) __c;				      \
 	 break;								      \
 									      \
        case 16:								      \
-	 *__ts++ = __c;							      \
+	 __u->__ui = __c;						      \
+	 __u = (void *) __u + 4;					      \
        case 12:								      \
-	 *__ts++ = __c;							      \
+	 __u->__ui = __c;						      \
+	 __u = (void *) __u + 4;					      \
        case 8:								      \
-	 *__ts++ = __c;							      \
+	 __u->__ui = __c;						      \
+	 __u = (void *) __u + 4;					      \
        case 4:								      \
-	 *__ts = __c;							      \
+	 __u->__ui = __c;						      \
        case 0:								      \
 	 break;								      \
        }								      \
@@ -494,48 +512,50 @@ __strlen_g (__const char *__str)
 		  : __strcpy_g (dest, src)))
 
 #define __strcpy_small(dest, src, srclen) \
-  (__extension__ ({ unsigned char *__dest = (unsigned char *) (dest);	      \
+  (__extension__ ({ char *__dest = (dest);				      \
+		    union {						      \
+		      unsigned int __ui;				      \
+		      unsigned short int __usi;				      \
+		      unsigned char __uc;				      \
+		      char __c;						      \
+		    } *__u = (void *) __dest;				      \
 		    switch (srclen)					      \
 		      {							      \
 		      case 1:						      \
-			*__dest = '\0';					      \
+			__u->__uc = '\0';				      \
 			break;						      \
 		      case 2:						      \
-			*((unsigned short int *) __dest) =		      \
-			  __STRING_SMALL_GET16 (src, 0);		      \
+			__u->__usi = __STRING_SMALL_GET16 (src, 0);	      \
 			break;						      \
 		      case 3:						      \
-			*((unsigned short int *) __dest) =		      \
-			  __STRING_SMALL_GET16 (src, 0);		      \
-			*(__dest + 2) = '\0';				      \
+			__u->__usi = __STRING_SMALL_GET16 (src, 0);	      \
+			__u = (void *) __u + 2;				      \
+			__u->__uc = '\0';				      \
 			break;						      \
 		      case 4:						      \
-			*((unsigned int *) __dest) =			      \
-			  __STRING_SMALL_GET32 (src, 0);		      \
+			__u->__ui = __STRING_SMALL_GET32 (src, 0);	      \
 			break;						      \
 		      case 5:						      \
-			*((unsigned int *) __dest) =			      \
-			  __STRING_SMALL_GET32 (src, 0);		      \
-			*(__dest + 4) = '\0';				      \
+			__u->__ui = __STRING_SMALL_GET32 (src, 0);	      \
+			__u = (void *) __u + 4;				      \
+			__u->__uc = '\0';				      \
 			break;						      \
 		      case 6:						      \
-			*((unsigned int *) __dest) =			      \
-			  __STRING_SMALL_GET32 (src, 0);		      \
-			*((unsigned short int *) (__dest + 4)) =	      \
-			  __STRING_SMALL_GET16 (src, 4);		      \
+			__u->__ui = __STRING_SMALL_GET32 (src, 0);	      \
+			__u = (void *) __u + 4;				      \
+			__u->__usi = __STRING_SMALL_GET16 (src, 4);	      \
 			break;						      \
 		      case 7:						      \
-			*((unsigned int *) __dest) =			      \
-			  __STRING_SMALL_GET32 (src, 0);		      \
-			*((unsigned short int *) (__dest + 4)) =	      \
-			  __STRING_SMALL_GET16 (src, 4);		      \
-			*(__dest + 6) = '\0';				      \
+			__u->__ui = __STRING_SMALL_GET32 (src, 0);	      \
+			__u = (void *) __u + 4;				      \
+			__u->__usi = __STRING_SMALL_GET16 (src, 4);	      \
+			__u = (void *) __u + 2;				      \
+			__u->__uc = '\0';				      \
 			break;						      \
 		      case 8:						      \
-			*((unsigned int *) __dest) =			      \
-			  __STRING_SMALL_GET32 (src, 0);		      \
-			*((unsigned int *) (__dest + 4)) =		      \
-			  __STRING_SMALL_GET32 (src, 4);		      \
+			__u->__ui = __STRING_SMALL_GET32 (src, 0);	      \
+			__u = (void *) __u + 4;				      \
+			__u->__ui = __STRING_SMALL_GET32 (src, 4);	      \
 			break;						      \
 		      }							      \
 		    (char *) __dest; }))
@@ -583,56 +603,56 @@ __strcpy_g (char *__dest, __const char *__src)
 # define stpcpy(dest, src) __stpcpy (dest, src)
 
 # define __stpcpy_small(dest, src, srclen) \
-  (__extension__ ({ unsigned char *__dest = (unsigned char *) (dest);	      \
+  (__extension__ ({ union {						      \
+		      unsigned int __ui;				      \
+		      unsigned short int __usi;				      \
+		      unsigned char __uc;				      \
+		      char __c;						      \
+		    } *__u = (void *) (dest);				      \
 		    switch (srclen)					      \
 		      {							      \
 		      case 1:						      \
-			*__dest = '\0';					      \
+			__u->__uc = '\0';				      \
 			break;						      \
 		      case 2:						      \
-			*((unsigned short int *) __dest) =		      \
-			  __STRING_SMALL_GET16 (src, 0);		      \
-			++__dest;					      \
+			__u->__usi = __STRING_SMALL_GET16 (src, 0);	      \
+			__u = (void *) __u + 1;				      \
 			break;						      \
 		      case 3:						      \
-			*((unsigned short int *) __dest)++ =		      \
-			  __STRING_SMALL_GET16 (src, 0);		      \
-			*__dest = '\0';					      \
+			__u->__usi = __STRING_SMALL_GET16 (src, 0);	      \
+			__u = (void *) __u + 2;				      \
+			__u->__uc = '\0';				      \
 			break;						      \
 		      case 4:						      \
-			*((unsigned int *) __dest) =			      \
-			  __STRING_SMALL_GET32 (src, 0);		      \
-			__dest += 3;					      \
+			__u->__ui = __STRING_SMALL_GET32 (src, 0);	      \
+			__u = (void *) __u + 3;				      \
 			break;						      \
 		      case 5:						      \
-			*((unsigned int *) __dest)++ =			      \
-			  __STRING_SMALL_GET32 (src, 0);		      \
-			*__dest = '\0';					      \
+			__u->__ui = __STRING_SMALL_GET32 (src, 0);	      \
+			__u = (void *) __u + 4;				      \
+			__u->__uc = '\0';				      \
 			break;						      \
 		      case 6:						      \
-			*((unsigned int *) __dest) =			      \
-			  __STRING_SMALL_GET32 (src, 0);		      \
-			*((unsigned short int *) (__dest + 4)) =	      \
-			  __STRING_SMALL_GET16 (src, 4);		      \
-			__dest += 5;					      \
+			__u->__ui = __STRING_SMALL_GET32 (src, 0);	      \
+			__u = (void *) __u + 4;				      \
+			__u->__usi = __STRING_SMALL_GET16 (src, 4);	      \
+			__u = (void *) __u + 1;				      \
 			break;						      \
 		      case 7:						      \
-			*((unsigned int *) __dest) =			      \
-			  __STRING_SMALL_GET32 (src, 0);		      \
-			*((unsigned short int *) (__dest + 4)) =	      \
-			  __STRING_SMALL_GET16 (src, 4);		      \
-			__dest += 6;					      \
-			*__dest = '\0';					      \
+			__u->__ui = __STRING_SMALL_GET32 (src, 0);	      \
+			__u = (void *) __u + 4;				      \
+			__u->__usi = __STRING_SMALL_GET16 (src, 4);	      \
+			__u = (void *) __u + 2;				      \
+			__u->__uc = '\0';				      \
 			break;						      \
 		      case 8:						      \
-			*((unsigned int *) __dest) =			      \
-			  __STRING_SMALL_GET32 (src, 0);		      \
-			*((unsigned int *) (__dest + 4)) =		      \
-			  __STRING_SMALL_GET32 (src, 4);		      \
-			__dest += 7;					      \
+			__u->__ui = __STRING_SMALL_GET32 (src, 0);	      \
+			__u = (void *) __u + 4;				      \
+			__u->__ui = __STRING_SMALL_GET32 (src, 4);	      \
+			__u = (void *) __u + 3;				      \
 			break;						      \
 		      }							      \
-		    (char *) __dest; }))
+		    (char *) __u; }))
 
 __STRING_INLINE char *__mempcpy_by4 (char *__dest, __const char *__src,
 				     size_t __srclen);
diff --git a/sysdeps/powerpc/dl-machine.c b/sysdeps/powerpc/dl-machine.c
index cf918fab03..9e158c9865 100644
--- a/sysdeps/powerpc/dl-machine.c
+++ b/sysdeps/powerpc/dl-machine.c
@@ -176,6 +176,11 @@ __elf_machine_runtime_setup (struct link_map *map, int lazy, int profile)
 				  ? _dl_prof_resolve
 				  : _dl_runtime_resolve);
 
+      if (profile && _dl_name_match_p (_dl_profile, map))
+	/* This is the object we are looking for.  Say that we really
+	   want profiling and the timers are started.  */
+	_dl_profile_map = map;
+
       if (lazy)
 	for (i = 0; i < num_plt_entries; i++)
 	{