Re: [PATCH][cilkplus] fix c++ implicit conversions with cilk_spawn (PR/69024, PR/68997)

[Jason Merrill <jason at redhat dot com>]

On 01/20/2016 12:57 PM, Ryan Burn wrote:
>       case AGGR_INIT_EXPR:
> +      {
> +	int len = 0;
> +	int ii = 0;
> +	extract_free_variables (TREE_OPERAND (t, 1), wd, ADD_READ);
> +	if (TREE_CODE (TREE_OPERAND (t, 0)) == INTEGER_CST)
> +	  {
> +	    len = TREE_INT_CST_LOW (TREE_OPERAND (t, 0));
> +
> +	    for (ii = 3; ii < len; ii++)
> +	      extract_free_variables (TREE_OPERAND (t, ii), wd, ADD_READ);
> +	    extract_free_variables (TREE_TYPE (t), wd, ADD_READ);
> +	  }
> +	break;
> +      }

Please add a comment about skipping operand 2 (the slot).  Would it make 
sense to skip operand 2 (the static chain) for CALL_EXPR, too?

> +is_conversion_operator_function_decl_p (tree t) {

Open brace gets its own line.

> +	tree fn = AGGR_INIT_EXPR_FN (exp);
> +	if (TREE_CODE (fn) == ADDR_EXPR
> +	    && is_conversion_operator_function_decl_p (TREE_OPERAND (fn, 0))

It would be good to have a cp_get_callee_fndecl like the normal 
get_callee_fndecl, but supporting AGGR_INIT_EXPR as well.  That would 
replace the less capable get_function_named_in_call function in 
constexpr.c and various other places that access AGGR_INIT_EXPR_FN 
directly.  Mind doing that, either in this patch or as a follow-up?

Jason