Re: [PATCH] 69517 - [5/6 regression] SEGV on a VLA with excess initializer elements

[Jason Merrill <jason at redhat dot com>]

On 04/10/2016 07:14 PM, Martin Sebor wrote:
> +	  if (TREE_CODE (type) == ARRAY_TYPE
> +	      && variably_modified_type_p (type, NULL_TREE)
> +	      && !processing_template_decl)
> +	    {
> +	      /* Statically check for overflow in VLA bounds and build
> +		 an expression that checks at runtime whether the VLA
> +		 is erroneous due to invalid (runtime) bounds.
> +		 Another expression to check for excess initializers
> +		 is built in build_vec_init.  */

Why do this both in check_initializer and then again in cp_finish_decl 
right after the call to check_initializer?

> +      /* Also check to see if the final array size is zero (the size
> +	 is unsigned so the earlier overflow check detects negative
> +	 values as well.  */
> +      tree zerocheck = fold_build2 (EQ_EXPR, boolean_type_node,
> +				    vlasize, size_zero_node);

I'm not sure whether we want this, given that GCC allows zero-length 
arrays in general.  As I recall, with the C++14 stuff whether we checked 
for zero was dependent on flag_iso, though I wasn't particularly happy 
with that.  If you want to check this unconditionally that's fine.

Jason