Variables whose types are anonymous structs, once again

[Matt Austern]

OK, I raised this issue once before.  I know more now than I did then, 
so I'm raising it again.

The mainline C++ compiler gives an error (a pedwarn, which is treated 
as an error by default) if you do this:
   enum { a } x;

The error message, and the comments in the source code, say that the 
compiler is enforcing clause 3.5 paragraph 8: "A name with no linkage 
(notably, the name of a class or enumeration declared in a local scope 
(3.3.2)) shall not be used to declare an entity with linkage."

At a minimum, that comment is wrong.  Whatever it is we're declaring x 
with, it's not "a name with no linkage".  Its type is a anonymous enum. 
  Tautology: since this type has no name, it can't have "a name with no 
linkage".

Now the next question: given that 3.5 paragraph 8 doesn't apply, should 
we accept or reject this code?  There the question gets more 
interesting.

It turns out that a lot of people have found what the standard says 
about this to be pretty murky.  I found three different issues on the 
CWG issues list that related directly to this case, and maybe a half 
dozen that were tangentially related.  Here are the ones that seem most 
directly relevant:
   http://anubis.dkuug.dk/jtc1/sc22/wg21/docs/cwg_closed.html#132
   http://anubis.dkuug.dk/jtc1/sc22/wg21/docs/cwg_active.html#278
   http://anubis.dkuug.dk/jtc1/sc22/wg21/docs/cwg_active.html#389

Note that issue 132 has NAD status, and it was rejected as NAD 
specifically because prohibiting code like what I listed above "would 
have introduced an incompatibility with the C language" because this 
example "represents an idiom that is used in C".

Issue 389, which has "review" status, has a different proposed 
resolution: the proposed resolution would explicitly ban this code.  
(For those who are unfamiliar with WG21 bureaucracy, "review" status 
means that the core working group is leaning towards this solution, 
that they have precise standardese that they're considering, but that 
the committee hasn't yet voted to accept this wording.)

So putting all this together, what I get is that: (1) as it stands 
right now the standard isn't as clear as it might be, but a close 
reading shows that allows "enum {a} x".  (2) The committee is leaning 
towards saying that that's a mistake and that this construct should be 
forbidden.

My feeling: we should do what the standard says, not what we think the 
committee will do in a future version of the standard.  Unless someone 
can come up with an argument that my reading is wrong, I'm going to 
submit a patch to allow this.

			--Matt