Re: [RFC][PATCH] Add reallocarray function.

[Paul Eggert <eggert at cs dot ucla dot edu>]

RÃdiger Sonderfeld wrote:
> +  /* size_t is unsigned so the behavior on overflow is defined.  */
> +  INTERNAL_SIZE_T bytes = nmemb * elem_size;
> +#define HALF_INTERNAL_SIZE_T                                    \
> +  (((INTERNAL_SIZE_T) 1) << (8 * sizeof (INTERNAL_SIZE_T) / 2))
> +  if (__builtin_expect ((nmemb | elem_size) >= HALF_INTERNAL_SIZE_T, 0))
> +    {
> +      if (elem_size != 0 && bytes / elem_size != nmemb)

Instead of cut and pasting this code from calloc, please refactor so 
that the code is present only once, with the goal of optimizing it when 
the GCC folks get their act together and have a function like the 
__builtin_umul_overflow function that Clang has had since January.  This 
will let calloc and reallocarray do the unsigned multiplication and 
inspect the hardware's overflow bit directly, which is nicer than the 
above hackery.