http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54652
Bug #: 54652 Summary: ICE with -g Classification: Unclassified Product: gcc Version: 4.7.1 Status: UNCONFIRMED Severity: normal Priority: P3 Component: c++ AssignedTo: unassig...@gcc.gnu.org ReportedBy: ja...@gcc.gnu.org CC: ja...@gcc.gnu.org typedef unsigned long L __attribute__ ((aligned)); typedef unsigned long L __attribute__ ((aligned)); template <typename T> struct S { union U { L l; } u; }; S <int> s; ICEs with -g, TYPE_NAME seems to be garbage. The problem is that handle_aligned_attribute first (with L as decl): tree tt = TREE_TYPE (decl); *type = build_variant_type_copy (*type); DECL_ORIGINAL_TYPE (decl) = tt; TYPE_NAME (*type) = decl; TREE_USED (*type) = TREE_USED (decl); TREE_TYPE (decl) = *type; then duplicate_decls is called on the L olddecl (with new L as newdecl), where TYPE_NAME (TREE_TYPE (olddecl)) == olddecl and TYPE_NAME (TREE_TYPE (newdecl)) == newdecl, in merge_types newdecl's type wins, and eventually ggc_free (newdecl); So, either something should have changed TYPE_NAME (newtype) from newdecl to olddecl before ggc_free (I guess that would be preferred, but can't it be TYPE_NAME of other related types too?), or we shouldn't ggc_free TYPE_DECLs. Jason, can you please look into this?