[Bug c++/95976] New: [[no_unique_address]] on union members has the opposite-of-intended effect

[Bug c++/95976] New: [[no_unique_address]] on union members has the opposite-of-intended effect

[metaprogrammingtheworld at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95976

            Bug ID: 95976
           Summary: [[no_unique_address]] on union members has the
                    opposite-of-intended effect
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: metaprogrammingtheworld at gmail dot com
  Target Milestone: ---

When using [[no_unique_address]] with union members, ironically, each member is
now given a unique address! Without [[no_unique_address]], the layout is as one
would expect. This is particularly unfortunate since it means that a union's
size grows in proportion to the number of members when using
[[no_unique_address]]!

Example:

//////////
struct empty {};

union no_attribute_t
{
  empty _0;
  empty _1;
};

union with_attribute_t
{
  [[no_unique_address]] empty _0;
  [[no_unique_address]] empty _1;
};

constexpr no_attribute_t no_attribute{};
constexpr with_attribute_t with_attribute{};

// This succeeds
static_assert( &no_attribute._0 == &no_attribute._1 );

// This fails
static_assert( &with_attribute._0 == &with_attribute._1 );
//////////

[Bug c++/95976] [[no_unique_address]] on union members has the opposite-of-intended effect

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95976

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
            Version|unknown                     |10.1.0

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
This changed behavior twice, first one with
r9-3261-gbedf03a298f87b43f02eaee57542fbd10228ec08
with which the union size became 0 rather than the expected 1, and then in
r9-5710-g7e574f68fa82e7c5f879fd468291ec8b5ebecc83
which made the union size 2 instead (also incorrect).  From what I can see in
the standard, [[no_unique_address]] is allowed on union members and those
members are potentially-overlapping subobjects, but the union description makes
it clear that all those members start at offset 0.

[Bug c++/95976] [[no_unique_address]] on union members has the opposite-of-intended effect

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95976

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2020-06-30
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |ABI
                 CC|                            |jason at gcc dot gnu.org,
                   |                            |redi at gcc dot gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
One way to fix this would be:
--- gcc/cp/class.c.jj   2020-06-27 12:35:11.935311848 +0200
+++ gcc/cp/class.c      2020-06-30 14:18:08.448067688 +0200
@@ -6575,7 +6575,9 @@ layout_class_type (tree t, tree *virtual
          /* We must also reset the DECL_MODE of the field.  */
          SET_DECL_MODE (field, TYPE_MODE (type));
        }
-      else if (might_overlap && is_empty_class (type))
+      else if (might_overlap
+              && is_empty_class (type)
+              && TREE_CODE (t) != UNION_TYPE)
        {
          DECL_FIELD_ABI_IGNORED (field) = 1;
          layout_empty_base_or_field (rli, field, empty_base_offsets);
Another might be not to change DECL_FIELD_OFFSET in layout_empty_base_or_field
for TREE_CODE (rli->t) == UNION_TYPE, etc.
The question is if [[no_unique_address]] members of a union should affect any
ABI decisions (such as passing the union by value) or not.
Looking at the config/*/* code, in most cases the targets ignore
DECL_FIELD_ABI_IGNORED on UNION_TYPE members, but ia64 does not, and I think
rs6000_special_round_type_align and darwin_rs6000_special_round_type_align
don't ignore it either.

[Bug c++/95976] [[no_unique_address]] on union members has the opposite-of-intended effect

[jason at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95976

Jason Merrill <jason at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |jason at gcc dot gnu.org

[Bug c++/95976] [[no_unique_address]] on union members has the opposite-of-intended effect

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95976

--- Comment #3 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-10 branch has been updated by Jason Merrill
<jason@gcc.gnu.org>:

https://gcc.gnu.org/g:3bb330022ce47a3e8966a9930f392e497c608f59

commit r10-8456-g3bb330022ce47a3e8966a9930f392e497c608f59
Author: Jason Merrill <jason@redhat.com>
Date:   Thu Jul 9 15:11:12 2020 -0400

    c++: [[no_unique_address]] fixes. [PR96105]

    We were wrongly checking is_empty_class on the result of strip_array_types
    rather than the actual field type.  We weren't considering the alignment of
    the data member.  We needed to handle unions the same way as
    layout_nonempty_base_or_field.

    gcc/cp/ChangeLog:

            PR c++/96105
            PR c++/96052
            PR c++/95976
            * class.c (check_field_decls): An array of empty classes is not an
            empty data member.
            (layout_empty_base_or_field): Handle explicit alignment.
            Fix union handling.

    gcc/testsuite/ChangeLog:

            PR c++/96105
            PR c++/96052
            PR c++/95976
            * g++.dg/cpp2a/no_unique_address4.C: New test.
            * g++.dg/cpp2a/no_unique_address5.C: New test.
            * g++.dg/cpp2a/no_unique_address6.C: New test.

[Bug c++/95976] [[no_unique_address]] on union members has the opposite-of-intended effect

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95976

--- Comment #4 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jason Merrill <jason@gcc.gnu.org>:

https://gcc.gnu.org/g:e47dfca5aa473e77fdff95d631dc39de87a41eec

commit r11-2014-ge47dfca5aa473e77fdff95d631dc39de87a41eec
Author: Jason Merrill <jason@redhat.com>
Date:   Thu Jul 9 15:11:12 2020 -0400

    c++: [[no_unique_address]] fixes. [PR96105]

    We were wrongly checking is_empty_class on the result of strip_array_types
    rather than the actual field type.  We weren't considering the alignment of
    the data member.  We needed to handle unions the same way as
    layout_nonempty_base_or_field.

    gcc/cp/ChangeLog:

            PR c++/96105
            PR c++/96052
            PR c++/95976
            * class.c (check_field_decls): An array of empty classes is not an
            empty data member.
            (layout_empty_base_or_field): Handle explicit alignment.
            Fix union handling.

    gcc/testsuite/ChangeLog:

            PR c++/96105
            PR c++/96052
            PR c++/95976
            * g++.dg/cpp2a/no_unique_address4.C: New test.
            * g++.dg/cpp2a/no_unique_address5.C: New test.
            * g++.dg/cpp2a/no_unique_address6.C: New test.

[Bug c++/95976] [[no_unique_address]] on union members has the opposite-of-intended effect

[jason at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95976

Jason Merrill <jason at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
   Target Milestone|---                         |10.2
         Resolution|---                         |FIXED

--- Comment #5 from Jason Merrill <jason at gcc dot gnu.org> ---
Fixed for 10.2/11.