On Wed, Jan 15, 2025 at 11:55:28AM -0500, Jason Merrill wrote:
> On 1/15/25 9:43 AM, Jason Merrill wrote:
> > On 11/25/24 4:46 PM, Marek Polacek wrote:
> > > Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk/14?
> > > 
> > > -- >8 --
> > > In a template, for
> > > 
> > >    union {
> > >      union {
> > >        T d;
> > >      };
> > >    };
> > > 
> > > build_anon_union_vars crates a malformed COMPONENT_REF: we have no
> > > DECL_NAME for the nested anon union so we create something like
> > > "object.".
> > > Most of the front end doesn't seem to care, but if such a tree gets to
> > > potential_constant_expression, it can cause a crash.
> > > 
> > > I suppose we can use any name for the COMPONENT_REF's member.
> > > tsubst_stmt
> > > should build up a proper one in:
> > > 
> > >      if (VAR_P (decl) && !DECL_NAME (decl)
> > >     && ANON_AGGR_TYPE_P (TREE_TYPE (decl)))
> > >        /* Anonymous aggregates are a special case.  */
> > >        finish_anon_union (decl);
> > > 
> > >     PR c++/117153
> > > 
> > > gcc/cp/ChangeLog:
> > > 
> > >     * decl2.cc (build_anon_union_vars): If DECL_NAME is empty, use
> > >     an anonymous name for the second operand of a COMPONENT_REF.
> > > 
> > > gcc/testsuite/ChangeLog:
> > > 
> > >     * g++.dg/other/anon-union6.C: New test.
> > >     * g++.dg/other/anon-union7.C: New test.
> > > ---
> > >   gcc/cp/decl2.cc                          |  5 ++++-
> > >   gcc/testsuite/g++.dg/other/anon-union6.C | 13 +++++++++++++
> > >   gcc/testsuite/g++.dg/other/anon-union7.C | 16 ++++++++++++++++
> > >   3 files changed, 33 insertions(+), 1 deletion(-)
> > >   create mode 100644 gcc/testsuite/g++.dg/other/anon-union6.C
> > >   create mode 100644 gcc/testsuite/g++.dg/other/anon-union7.C
> > > 
> > > diff --git a/gcc/cp/decl2.cc b/gcc/cp/decl2.cc
> > > index bc3a9d0e922..1383cbb660c 100644
> > > --- a/gcc/cp/decl2.cc
> > > +++ b/gcc/cp/decl2.cc
> > > @@ -1959,7 +1959,10 @@ build_anon_union_vars (tree type, tree object)
> > >         if (processing_template_decl)
> > >       ref = build_min_nt_loc (UNKNOWN_LOCATION, COMPONENT_REF, object,
> > > -                DECL_NAME (field), NULL_TREE);
> > > +                (DECL_NAME (field)
> > > +                 ? DECL_NAME (field)
> > > +                 : make_anon_name ()),
> > > +                NULL_TREE);
> > 
> > Why don't we use the field directly instead of its name, even for named
> > fields?
> 
> I guess instantiation_dependent_r has a clue:
> 
> >         /* In a template, finish_class_member_access_expr creates a
> > COMPONENT_REF with an IDENTIFIER_NODE for op1 even if it isn't
> > type-dependent, so that we can check access control at
> > instantiation time (PR 42277).  See also Core issue 1273.  */
> 
> but I don't think we need to care about that in build_anon_union_vars.

Ah, interesting.  Using the field directly works, so I suppose we can
go with this simpler patch.  Thanks.

Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk?

-- >8 --
In a template, for

  union {
    union {
      T d;
    };
  };

build_anon_union_vars crates a malformed COMPONENT_REF: we have no
DECL_NAME for the nested anon union so we create something like "object.".
Most of the front end doesn't seem to care, but if such a tree gets to
potential_constant_expression, it can cause a crash.

We can use FIELD directly for the COMPONENT_REF's member.  tsubst_stmt
should build up a proper one in:

    if (VAR_P (decl) && !DECL_NAME (decl)
        && ANON_AGGR_TYPE_P (TREE_TYPE (decl)))
      /* Anonymous aggregates are a special case.  */
      finish_anon_union (decl);

        PR c++/117153

gcc/cp/ChangeLog:

        * decl2.cc (build_anon_union_vars): Use FIELD for the second operand
        of a COMPONENT_REF.

gcc/testsuite/ChangeLog:

        * g++.dg/other/anon-union6.C: New test.
        * g++.dg/other/anon-union7.C: New test.
---
 gcc/cp/decl2.cc                          |  2 +-
 gcc/testsuite/g++.dg/other/anon-union6.C | 13 +++++++++++++
 gcc/testsuite/g++.dg/other/anon-union7.C | 16 ++++++++++++++++
 3 files changed, 30 insertions(+), 1 deletion(-)
 create mode 100644 gcc/testsuite/g++.dg/other/anon-union6.C
 create mode 100644 gcc/testsuite/g++.dg/other/anon-union7.C

diff --git a/gcc/cp/decl2.cc b/gcc/cp/decl2.cc
index ffe1aa3a324..9e61afd359f 100644
--- a/gcc/cp/decl2.cc
+++ b/gcc/cp/decl2.cc
@@ -2032,7 +2032,7 @@ build_anon_union_vars (tree type, tree object)
 
       if (processing_template_decl)
        ref = build_min_nt_loc (UNKNOWN_LOCATION, COMPONENT_REF, object,
-                               DECL_NAME (field), NULL_TREE);
+                               field, NULL_TREE);
       else
        ref = build_class_member_access_expr (object, field, NULL_TREE,
                                              false, tf_warning_or_error);
diff --git a/gcc/testsuite/g++.dg/other/anon-union6.C 
b/gcc/testsuite/g++.dg/other/anon-union6.C
new file mode 100644
index 00000000000..d1cde30f790
--- /dev/null
+++ b/gcc/testsuite/g++.dg/other/anon-union6.C
@@ -0,0 +1,13 @@
+// PR c++/117153
+// { dg-do compile }
+
+template<typename T>
+void f() {
+  union {
+    union {
+      T d;
+    };
+  };
+  (void) (d + 0);
+}
+template void f<double>();
diff --git a/gcc/testsuite/g++.dg/other/anon-union7.C 
b/gcc/testsuite/g++.dg/other/anon-union7.C
new file mode 100644
index 00000000000..e89334a5d0e
--- /dev/null
+++ b/gcc/testsuite/g++.dg/other/anon-union7.C
@@ -0,0 +1,16 @@
+// PR c++/117153
+// { dg-do compile { target c++11 } }
+
+using U = union { union { int a; }; };
+
+U
+foo ()
+{
+  return {};
+}
+
+void
+g ()
+{
+  [[maybe_unused]] auto u = foo ();
+}

base-commit: 4d18acf8023ba0495007aa7a6f36d5509a51760b
-- 
2.48.1

Reply via email to