On 5/25/26 10:58 AM, Jakub Jelinek wrote:
Hi!
As I've mentioned on Saturday, the CWG3130 patchset fail to bootstrap.
The problem is that we try to call build_value_init on anonymous unions
or structs, which doesn't work well when they don't have a default
constructor.
Now, if some non-trivial construction is needed, type will already have
either a user-provided constructor or at least non-trivial one, in that
case build_value_init_noctor isn't called at all. So, this patch
just zero-initializes the anonymous aggregate members.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
2026-05-25 Jakub Jelinek <[email protected]>
* init.cc (build_value_init_noctor): Zero initialize anonymous
union/struct subobjects. Formatting fix.
--- gcc/cp/init.cc.jj 2026-04-03 10:18:16.808121758 +0200
+++ gcc/cp/init.cc 2026-05-25 11:01:59.409845336 +0200
@@ -445,6 +445,20 @@ build_value_init_noctor (tree type, tsub
&& integer_zerop (DECL_SIZE (field)))
continue;
+ /* Zero-initialize anonymous union and struct members. The
+ default constructor doesn't exist for those and if NSDMIs
+ are used, the containing type's default constructor is
+ non-trivial. */
+ if (ANON_AGGR_TYPE_P (ftype))
+ {
+ value = build_zero_init (ftype, NULL_TREE,
+ /*static_storage_p=*/false);
+ if (value == error_mark_node)
+ return error_mark_node;
+ CONSTRUCTOR_APPEND_ELT (v, field, value);
+ continue;
It seems shorter to share the error check and append with the usual
case, just choosing between build_zero_init and build_value_init? OK
either way.
+ }
+
/* We could skip vfields and fields of types with
user-defined constructors, but I think that won't improve
performance at all; it should be simpler in general just
@@ -461,7 +475,7 @@ build_value_init_noctor (tree type, tsub
if (value == error_mark_node)
return error_mark_node;
- CONSTRUCTOR_APPEND_ELT(v, field, value);
+ CONSTRUCTOR_APPEND_ELT (v, field, value);
/* We shouldn't have gotten here for anything that would need
non-trivial initialization, and gimplify_init_ctor_preeval
Jakub
