RE: Windows realloc & GNU

[Conrad T. Pino]

> From: Derek Price [mailto:derek@ximbiot.com]
> Sent: Tuesday, September 20, 2005 14:33
> 
> Conrad T. Pino wrote:
> 
> >Yes but the "free" function detects the write past the end and displays
> >a dialog box with "Abort", "Retry", "Ignore" options.  Nothing odd occurs
> >when "free" isn't called.  Looks clean to me.
> 
> That isn't right.  Since realloc(0,0) should return a valid pointer, you
> need to be able to pass it to free() without a problem.  Some code may
> be making this assumption.  Use the GNULIB replacement realloc in
> lib/realloc.c.

It IS a valid pointer that "free" will accept gladly provided NO WRITES
are performed OUTSIDE the allocated range.  The "free" function checks
a guard byte at the end of the block to detect this error:

        char *p;

        p = realloc( NULL, 1 );
        p[0] = 1; /* this is fine and */
        free(p); /* free will be quite */

        p = realloc( NULL, 1 );
        p[0] = 1; /* this is fine but */
        p[1] = 1; /* this is a no no and */
        free(p); /* free will complain */

        p = realloc( NULL, 0 );
        p[0] = 1; /* this is a no no and */
        free(p); /* free will complain */

> Regards,

Ditto,

> Derek

Conrad