[C++ PATCH] P1091R3 - Extending structured bindings to be more like var decls

[Jason Merrill]

On 5/20/19 5:44 PM, Jakub Jelinek wrote:
> Hi!
> 
> The following patch is an attempt at implementing these two papers.
> I must be missing something important, because in P1091R3 the rationale
> talks about the intent to support structured bindings with static,
> thread_local, constexpr, inline and extern, but the actual wording
> changes in the paper are only about static and thread_local, because dcl.dcl
> still says
> If the decl-specifier-seq contains any decl-specifier other than static,
> thread_local, auto, or cv-qualifiers, the program is ill-formed.

Right, the original proposal was for more, but it was limited to static 
and thread_local by the committee.

> Another thing are the lambda captures of the structured bindings.
> The end result is that they are allowed, both capturing values and capturing
> references, except when capturing bitfields, but that is actually what gcc
> has been implementing for quite a while.  The question is if we shouldn't
> reject those for -std=c++17 or -std=c++17 -pedantic-errors, seems e.g.
> clang++ rejects those, but looking at the C++17 standard I don't see any
> wording that would say those are not allowed.  It was only P0588R1 that
> added
> "or captures a structured binding (explicitly or implicitly), the program is
> ill-formed."
> Is P0588R1 a DR that would be applied to older standards?

P0588 was certainly intended to apply to C++17, but I'd treat this as 
overriding it, and not complain.

> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
> 
> 2019-05-20  Jakub Jelinek  <jakub@redhat.com>
> 
> 	P1091R3 - Extending structured bindings to be more like var decls
> 	P1381R1 - Reference capture of structured bindings
> 	* decl.c (cp_maybe_mangle_decomp): Handle TREE_STATIC decls even at
> 	function scope.
> 	(cp_finish_decomp): Copy over various decl properties from decl to
> 	v[i] in the tuple case.
> 	(grokdeclarator): Allow static, thread_local and __thread for C++2a
> 	and use pedwarn instead of error for older standard revisions.
> 	Make other structured binding diagnostic messages more i18n friendly.
> 
> 	* g++.dg/cpp1z/decomp3.C (test): For static, expect only warning
> 	instead of error and only for c++17_down.  Add a thread_local test.
> 	(z2): Add a __thread test.
> 	* g++.dg/cpp2a/decomp1.C: New test.
> 	* g++.dg/cpp2a/decomp1-aux.cc: New file.
> 	* g++.dg/cpp2a/decomp2.C: New test.
> 	* g++.dg/cpp2a/decomp3.C: New test.
> 
> --- gcc/cp/decl.c.jj	2019-05-20 14:52:26.871375569 +0200
> +++ gcc/cp/decl.c	2019-05-20 15:48:33.563599150 +0200
> +      && (DECL_NAMESPACE_SCOPE_P (decl) || TREE_STATIC (decl)))
 > +	      if (TREE_STATIC (v[i]))

When can a structured binding not be TREE_STATIC?

Jason