https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94609
--- Comment #7 from Iain Buclaw <ibuclaw at gdcproject dot org> ---
I'm initially considering the following:
--- a/gcc/d/types.cc
+++ b/gcc/d/types.cc
@@ -915,7 +915,7 @@ public:
/* For structs with a user defined postblit or a destructor,
also set TREE_ADDRESSABLE on the type and all variants.
This will make the struct be passed around by reference. */
- if (t->sym->postblit || t->sym->dtor)
+ if (t->sym->enclosing || t->sym->postblit || t->sym->dtor)
{
for (tree tv = t->ctype; tv != NULL_TREE; tv = TYPE_NEXT_VARIANT (tv))
TREE_ADDRESSABLE (tv) = 1;
However the front-end for some time now has implemented an isPOD method that
recursively checks all fields too. It would be better to use this instead.
--- a/gcc/d/types.cc
+++ b/gcc/d/types.cc
@@ -915,7 +915,7 @@ public:
/* For structs with a user defined postblit or a destructor,
also set TREE_ADDRESSABLE on the type and all variants.
This will make the struct be passed around by reference. */
- if (t->sym->postblit || t->sym->dtor)
+ if (!t->sym->isPOD ())
{
for (tree tv = t->ctype; tv != NULL_TREE; tv = TYPE_NEXT_VARIANT (tv))
TREE_ADDRESSABLE (tv) = 1;
The test itself now passes with the above, but going to rebuild the library and
run it on the wider testsuite.
