Re: xalloc.h (x2nrealloc): Don't always double the buffer size.

[Paul Eggert]

Jim Meyering <[EMAIL PROTECTED]> writes:

> -      if (((size_t) -1) / 2 / s < n)
> +      if ((2 * (((size_t) -1 - 1) / 3)) / s < n)

That's not quite right.  As an extreme case, suppose S is
SIZE_MAX/4 + 1 and N is 2.  Then (2 * (((size_t) -1 - 1) / 3)) / S
evaluates to 2 and N will appear to be in range here, but:

> +      n = n + n / 2 + 1;

will cause N to become 4, and N * S will then overflow.

I installed this:

2007-02-03  Paul Eggert  <[EMAIL PROTECTED]>

        * lib/xalloc.h (x2nrealloc): Fix an unlikely bug in the overflow
        checking code.  Set N = ceil (1.5 * N) rather than to a slightly
        larger value.

--- lib/xalloc.h        1 Feb 2007 23:53:04 -0000       1.33
+++ lib/xalloc.h        4 Feb 2007 02:20:07 -0000
@@ -141,7 +141,7 @@ xnrealloc (void *p, size_t n, size_t s)

    In the following implementation, nonzero sizes are increased by a
    factor of approximately 1.5 so that repeated reallocations have
-   O(N log N) overall cost rather than O(N**2) cost, but the
+   O(N) overall cost rather than O(N**2) cost, but the
    specification for this function does not guarantee that rate.

    Here is an example of use:
@@ -204,9 +204,13 @@ x2nrealloc (void *p, size_t *pn, size_t 
     }
   else
     {
-      if ((2 * (((size_t) -1 - 1) / 3)) / s < n)
+      /* Set N = ceil (1.5 * N) so that progress is made if N == 1.
+        Check for overflow, so that N * S stays in size_t range.
+        The check is slightly conservative, but an exact check isn't
+        worth the trouble.  */
+      if ((size_t) -1 / 3 * 2 / s <= n)
        xalloc_die ();
-      n = n + n / 2 + 1;
+      n += (n + 1) / 2;
     }

   *pn = n;