RE: Unable to bootstrap gcc snapshot of 20001211 on Dynix/ptx

[biswapesh dot chattopadhyay at bt dot com]

> According to the Single Unix Specification, bcopy must handle 
> overlapping
> memory.
> 

There definitely is a bug in bcopy in Dynix/ptx 4.4.8 then, and I've got a
testcase ready. Here is the code:
-----------------------------------------------------------------
#include <string.h>
#ifdef USE_BCOPY
#define BUF "Native bcopy"
#include <strings.h>
#else
#ifndef USE_MEMMOVE
/* This is the GNU implementation found in libinerty/bcopy.c */
void bcopy (src, dest, len)
  register char *src, *dest;
 int len;
 {
  if (dest < src)
     while (len--)
       *dest++ = *src++;
      else
         {
           char *lasts = src + (len-1);
           char *lastd = dest + (len-1);
              while (len--)
               *(char *)lastd-- = *(char *)lasts--;
                }
 }
#define BUF "GNU bcopy"
#else
#define bcopy(A,B,C) memmove((B), (A), (C))
#define BUF "Memmove"
#endif
#endif

int main()
{
  char *str = strdup("Test string");
  bcopy(str, str + 2, 6);
  printf("%s: %s\n", BUF, str);
}
-------------------------------------------------------------------------

Here is the output:
----------------------------------------
/usr/home/swbxc01/cpp>gcc -DUSE_BCOPY -o test1 test1.c
/usr/home/swbxc01/cpp>./test1
Native bcopy: TeTeststing
/usr/home/swbxc01/cpp>gcc -o test1 test1.c
/usr/home/swbxc01/cpp>./test1
GNU bcopy: TeTest sing
/usr/home/swbxc01/cpp>gcc -DUSE_MEMMOVE -o test1 test1.c
/usr/home/swbxc01/cpp>./test1
Memmove: TeTest sing
--------------------------------------------

So, GNU bcopy and memmove do the correct thing, ptx native bcopy comes up
with a diferent (wrong) string. Janis, can you test this in ptx 4.6.0 to see
if the bug has been corrected ? This might explain why you were able to
compile gcc without the cp/class.c BCOPY patch.

It might be a good idea to incorporate a test somewhat like this one in
autoconf to check for buggy bcopy and disable (substitute it with the
libiberty version/memmove) accordingly. Another idea might be not to use
bcopy for overlapping memory regions (I went through the instances of bcopy
and cp/class.c seems to be the only candidate.)

Rgds.
Biswa