[Libvir] RFC: safer memory allocation APIs with compile time checking

[Mark McLoughlin]

On Mon, 2008-04-28 at 03:39 -0400, Daniel Veillard wrote:
> >   #define VIR_REALLOC(ptr)          virRealloc(&(ptr), sizeof(*(ptr)))
> 
>   That i really don't understand. How do you expect to use that realloc
> without passing a new size.
> 
> >   #define VIR_REALLOC_N(ptr, count) virReallocN(&(ptr), sizeof(*(ptr)), (count))
> 
>   That I can understand , but the previous one i can't.
> 
> >   #define VIR_FREE(ptr)             virFree(&(ptr))
> [...]
> > Much less ugly:
> > 
> >       if (VIR_ALLOC_N(guest->arch.defaultInfo.machines, nmachines) < 0)
> >          return -1;
> > 
> >       if (VIR_REALLOC(migrateTrans, caps->host.nmigrateTrans+1) < 0)
> >          return -1;
> 
>   how does sizeof(*(caps->host.nmigrateTrans+1)) increases the size ? 
> Doesn't make sense to me, you take a pointer, increment it, so basically just
> pointing to the next element in the array, but the size of the pointed object
> would be identical and realloc() becomes a noop.
>   
>   The proposal may help clean a lot of things, but VIR_REALLOC I don't
> understand, what did i missed ?

Looks to me like VIR_REALLOC() would be a rarely used API and Dan just
had a typo in the above example - it should have used VIR_REALLOC_N()

I'm having difficulty thinking of how VIR_REALLOC() might be interesting
- e.g. the highly contrived example below.

Cheers,
Mark.

   struct foo
   {
       int i;
   };

   struct bar
   {
       struct foo foo;
       int j;
   }

   struct foo *get_foo(int i)
   {
       struct foo *foo;

       if (!VIR_ALLOC(foo))
           return NULL;

       foo->i = i;

       return foo;
   }

   struct bar *get_bar(int i, int j)
   {
       struct bar *bar;
       struct foo *foo;

       if (!(foo = get_foo(i)))
           return NULL;

       bar = (struct bar *) foo;

       if (!VIR_REALLOC(bar)) {
           VIR_FREE(foo);
           return NULL;
       }

       bar->j = j;

       return bar;      
   }