[PATCH] reject invalid flexarrays even in anonymous structs (PR 93753)

[Jason Merrill]

On 2/18/20 12:06 AM, Martin Sebor wrote:
> The improved checking of flexible array members committed in r231665
> (and later) deliberately excluded anonymous structs and unions
> (unnamed members of types with no tags) to allow a harmless extension
> that was accepted by G++ prior to the change.  However, being overly
> broad, the exemption prevents flexible array members that are invalid
> for other reasons from being diagnosed in these contexts.  That in
> turn leads to a subset of such cases causing ICEs downstream.
> 
> The attached trivial patch tightens up the conditions under which
> this extension is accepted: it only considers such structs when they
> are members, thus eliminating the ICE.

The comment says,

          Ignore anonymous structs and unions whose members are 

          considered to be members of the enclosing class and thus will 

          be diagnosed when checking it.

Certainly if there is no enclosing class this is not the case, so the 
patch is an improvement.  Do we already properly diagnose an unnamed 
struct used to declare a data member?

Jason