bug in strxfrm

[Bruno Haible <haible at ilog dot fr>]

The inline function utf8_encode, used by strxfrm, currently returns 0
(instead of the result's length) if the argument is >= 0x80.

Here is a patch which
  - fixes the bug,
  - further optimizes this function by using simple arithmetic instead of
    memory accesses.


2000-07-30  Bruno Haible  <haible@clisp.cons.org>

	* string/strxfrm.c (encoding_mask, encoding_byte): Remove.
	(utf8_encode): Use simple shifts instead. Fix return value.

*** glibc-20000729/string/strxfrm.c.bak	Wed Jan 26 23:17:05 2000
--- glibc-20000729/string/strxfrm.c	Sun Jul 30 10:29:50 2000
***************
*** 46,69 ****
  
  
  #ifndef WIDE_CHAR_VERSION
- /* These are definitions used by some of the functions for handling
-    UTF-8 encoding below.  */
- static const uint32_t encoding_mask[] =
- {
-   ~0x7ff, ~0xffff, ~0x1fffff, ~0x3ffffff
- };
- 
- static const unsigned char encoding_byte[] =
- {
-   0xc0, 0xe0, 0xf0, 0xf8, 0xfc
- };
- 
  
  /* We need UTF-8 encoding of numbers.  */
  static inline int
  utf8_encode (char *buf, int val)
  {
-   char *startp = buf;
    int retval;
  
    if (val < 0x80)
--- 47,57 ----
***************
*** 76,86 ****
        int step;
  
        for (step = 2; step < 6; ++step)
! 	if ((val & encoding_mask[step - 2]) == 0)
  	  break;
        retval = step;
  
!       *buf = encoding_byte[step - 2];
        --step;
        do
  	{
--- 64,74 ----
        int step;
  
        for (step = 2; step < 6; ++step)
! 	if ((val & (~(uint32_t)0 << (5 * step + 1))) == 0)
  	  break;
        retval = step;
  
!       *buf = (unsigned char) (~0xff >> step);
        --step;
        do
  	{
***************
*** 91,97 ****
        *buf |= val;
      }
  
!   return buf - startp;
  }
  #endif
  
--- 79,85 ----
        *buf |= val;
      }
  
!   return retval;
  }
  #endif