What does malloc(0) return?

[Doug Gwyn]

In article <4538 at inews.intel.com> bhoughto at hopi.intel.com (Blair P. Houghton) writes:
>In article <16317 at smoke.brl.mil> gwyn at smoke.brl.mil (Doug Gwyn) writes:
>>malloc(0) should return either NULL or a pointer to some storage location.
>>Technically the behavior is undefined, since there are no objects of the
>>specified size (0) according to the standard.
>	"If the size of the space requested is zero, the behavior
>	 is implementation-defined; the value returned shall be
>	 either a null pointer or a unique pointer."
>		 (ANSI X3.159-1989, sec. 4.10.3, p. 155, ll. 18-20)

Oops, thanks for pointing that out.  (I remembered the committee debate
on the issue, but not the outcome.)  Thus the "should" in my response
becomes a "must" and "undefined" should have been "implementation-defined".

I think, but am not sure, that malloc()ed storage is intended to be such
that the address (one past the end) of the requested amount of storage
be meaningful (but its "contents" not necessarily accessible), just as for
explicitly declared arrays.

>I.e., `malloc(0)' must "eat" space if it does not return NULL, ...

The rest of that discussion seemed confusing.  I think it is easier to
simply say that the implementation has the choice (which must be
documented) of EITHER simply returning NULL for a 0-sized request, OR
acting in effect as though the request had been for just 1 byte.  In some
situations the ability to not actually make the additional storage
accessible might be of some use to the implementor; however, since a
strictly conforming program cannot count on 0-sized malloc()s succeeding
even when there is plenty of unallocated storage available, it is hardly
worth making such an optimization for this special case (which hardly any
program will ever tickle).

Generally I would think that reasonable implementations of malloc()
would naturally support successful 0-sized allocations with no special
coding for the 0-sized case.  Certainly this would be true for "block
header" style schemes; for "buddy system" schemes (which I don't
recommend), there would have to be a simple fudge (such as
"if(bytes_wanted==0)++bytes_wanted;" to obtain the required "unique
pointer" property.