glibc 2.3.1: fix for the i386 inline strings code

Nobody has answered to me for a while, so I'm resending this politely...

This is a trivial patch for the inlined i386 strings' header. Without it, say, X11 can't be compiled with -D__USE_STRING_INLINES - a problems will be caused by XtNewString and X11's own bzero. These problems are cured by the first and the second chunks of the patch, correspondingly.

So, please, apply this.

@@ -40,10 +40,10 @@
 
 /* The macros are used in some of the optimized implementations below.  */
 #define __STRING_SMALL_GET16(src, idx) \
-  (((src)[idx + 1] << 8) | (src)[idx])
+  ((((char*)(src))[idx + 1] << 8) | ((char*)(src))[idx])
 #define __STRING_SMALL_GET32(src, idx) \
-  ((((src)[idx + 3] << 8 | (src)[idx + 2]) << 8				      \
-    | (src)[idx + 1]) << 8 | (src)[idx])
+  (((((char*)(src))[idx + 3] << 8 | ((char*)(src))[idx + 2]) << 8	      \
+    | ((char*)(src))[idx + 1]) << 8 | ((char*)(src))[idx])
 
 
 /* Copy N bytes of SRC to DEST.  */

@@ -229,7 +229,7 @@ memcmp (__const void *__s1, __const void
 	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)						      \
+     if ((n) == 3 || (n) >= 5)						      \
        __asm__ __volatile__ ("" : "=r" (__c) : "0" (__c));		      \
 									      \
      /* This `switch' statement will be removed at compile-time.  */	      \

The exact problem with XtNewString is that we have the "void pointer dereference" error when it is used with void* argument: XtNewString is a macros which uses strcpy, whose inline variant, in turn, uses __STRING_SMALL_GET(16|32), which don't cast their (str) argument to (char*). The patch inserts such a casting. But this is not the X-specific problem, I just met it when was compiling X. It's a general error - string functions should work happily with void* arguments, so __STRING_SMALL_GET macros' must not assume their (str) argument can't be of type void*...

Again, the second problem is not X-specific, but X-exposed (for me). X's bzero is defined as memset, and when we have something like

        bzero(..., col = screen->max_col + 1)

the inlined memset, which is a macros, dies, because there is no brackets around the second bzero's argument, where it is used in the macros. And, again, it is a general error - the brackets must be there. The second chunk of the patch cures this.

On Tue, Jan 07, 2003 at 08:43:01AM +0500, Denis Zaitsev wrote:
[...]

To be more specific: __STRING_SMALL_GET expects the (str) argument to be an `unsigned char *'. Everything else will make it do funny things when (signed) characters have values >= 128. Since __strcpy_a_small() and __stpcpy_a_small() pass the pointer unmodified, they're broken as well.

 Michael "Tired" Riepe <[email protected]>
 "All I wanna do is have a little fun before I die"

You are right, indeed... There must be the (unsigned char*) cast.

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I've applied the patch after correcting the type used in the casts. Thanks,

- --
- --------------. ,-. 444 Castro Street
Ulrich Drepper \ ,-----------------' \ Mountain View, CA 94041 USA
Red Hat --' drepper at redhat.com ---------------------------
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (GNU/Linux)

iD8DBQE+I7Gd2ijCOnn/RHQRAsVBAJ956F4GTF/fv1RpbU9nImKsPvgfxwCghrPN UWiP7thvgVDX78LkRQleBks=
=V2JO
-----END PGP SIGNATURE-----