On Tue, Mar 20, 2018 at 9:55 AM, Nathan Sidwell <nat...@acm.org> wrote:
> This fixes the 84962 ICE, which is where we name a member function of an
> anon-struct in the width specifier of a bitfield. The problem is that when
> permissive we allow some of this strangeness. Prior to my lookup changes, we
> only added TYPE_FIELDS and we still do that. Trouble is that now contains
> member functions, and badness happens. The fix is to add the member_vec
> members, if that exists, which takes care of overloaded.
>
> Due to history, we don't check some of the anon-struct restrictions until
> the outer struct is complete. That's partly historical and partly because
> we don't tell the struct parser the difference between
>
> struct X {
> struct { ... }; // anon-struct
> typedef struct { ... } name; // not an anon-struct
Right, because we can't tell until after the struct:
struct { ... } member; // also not an anon-struct
> However, all this stuff is an extension:
> 1) we permit anon-struct, not just anon-union
> 2) we permit things other than non-static-data members
> Maybe we could deprecate #2 at least? I'm not sure what its use case might
> be.
Fine with me. I know #1 was deliberate, but I haven't been aware of #2 before.
Jason
