Re: C++ and Java CNI: Check Java references

[Geoff Keating <geoffk at geoffk dot org>]

Fergus Henderson <fjh@cs.mu.oz.au> writes:

> On 24-Apr-2001, Andrew Haley <aph@cambridge.redhat.com> wrote:
> >  > It's also common to write expressions such as `&p[0]',
> > 
> > I think it's impossible in this case.  These are Java objects, and
> > C++-style arrays can't occur.
> 
> In C++, any object is equivalent to a length one array of itself.
> 
> So library code may handle the single-object case by calling the code
> for the array case with an array size of one.
> 
> You might, for example, pass a pointer to a Java object to some
> library template routine that uses &p[0] before checking whether
> p is null:
> 
> 	template <class T>
> 	void foo(T *p, int n) {
> 		void *raw_pointer = &p[0];
> 		if (raw_pointer != NULL && n != 0) { ... }
> 	}
> 
> 	template <class T>
> 	void foo(T *p) {
> 		foo(p, 1);
> 	}
> 
> Such code isn't strictly conforming, but it works on all traditional
> implementations, which don't explicitly check for dereferencing null,
> and for which no actual dereference takes place if the dereferencing
> operator is the operand of unary `&'.
> 
> OK, maybe this is a bit far-fetched.  It's unlikely to be a big
> problem in practice, I guess.

In C at least the ISO C99 standard was explicitly modified to allow
this to work, so apparently someone uses it.  The equivalent

void foo(char *p, int n) {
  void *raw_pointer = &p[0];
   if (raw_pointer != NULL && n != 0) { ... }
}

is strictly conforming C.

It's likely this change will move into C++ in the next standard revision.

-- 
- Geoffrey Keating <geoffk@geoffk.org>