[Bug c++/56241] [4.6/4.7/4.8 Regression] ICE in toplev.c:332 on invalid

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56241



--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-02-07 
16:25:21 UTC ---

Created attachment 29386

  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29386

gcc48-pr56241.patch



Fixed thusly, build_constructor doesn't like elts with NULL values, such elts

shouldn't be added at all.



I wonder about several other places in init.c where I believe NULL elt->value

could be added:

          ce.value = build_zero_init_1 (TREE_TYPE (type),

                                         /*nelts=*/NULL_TREE,

                                         static_storage_p, NULL_TREE);

          v->quick_push (ce);

in build_zero_init_1 (perhaps vec_alloc should be moved right before the

quick_push and both calls guarded with ce.value != NULL condition), and

          ce.value = build_value_init (TREE_TYPE (type), complain);

          v->quick_push (ce);

in build_value_init_noctor (likewise).