[Bug c/116016] enhancement: add __builtin_set_counted_by(P->FAM, COUNT) or equivalent

[isanbard at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116016

--- Comment #23 from Bill Wendling <isanbard at gmail dot com> ---
(In reply to qinzhao from comment #22)
> the following is the user documentation I came up based on all the
> discussion so far, let me know any comment and suggestion. (refer to GCC's
> __builtin_clear_padding doc on the prototype of the new builtin:
> https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html#index-
> _005f_005fbuiltin_005fclear_005fpadding)
> 
> 
> Builtin-in Function: void __builtin_set_counted_by (ptr, type expr) 
> 
> The built-in function __builtin_set_counted_by checks whether the array
> object pointed by the pointer PTR has another object associated with it that
> represents the number of elements in the array object through counted_by
> attribute (i.e, the counted-by object). If so, sets the corresponding
> counted-by object to EXPR. If such counted-by object does not exist, do
> nothing.
> 
> The first argument must be a pointer to an array.
> The TYPE of the second argument may be any integral type.

The documentation for the 'counted_by' attribute says this about the field
value:

"The field that represents the number of the elements should have an integer
type. Otherwise, the compiler reports an error and ignores the attribute."

I think we should specify that the type of the second argument must (or should)
be an integral type. Using "may" seems less strict.

> This built-in never evaluates its argument for side effects. If there are

s/argument/arguments/

> any side effects in them, the compiler does not set the counted-by object if
> there is one and issues warnings at the same time. 

Suggested alternative to the second sentence:

If there are any side effects in them, the builtin becomes a no-op and the
compiler issues a diagnostic.

> For example:
> 
> for the following:
> 
>   struct foo1 {
>     int counter1;
>     struct bar1 array[] __attribute__((counted_by(counter)));
>   } *p;
> 
>   struct foo2 {
>     int other;
>     struct bar2 array[];
>   } *q;
> 
>   __builtin_set_counted_by (p->array, COUNT)
> 
> behaves like:
> 
>   p->counter1 = COUNT;
> 
> However, 
> 
>   __builtin_set_counted_by (q->array, COUNT)
> 
> behaves like a no-op since q->array does not have any associated counted-by
> object through counted-by attribute.