Hi! >From the comments on gcc-patches and from what the then block does I think only FUNCTION_DECLs were meant to be handled this way, but as the testcase shows we can trigger it even with VAR_DECLs.
Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk? 2014-03-10 Jakub Jelinek <ja...@redhat.com> PR ipa/60457 * ipa.c (symtab_remove_unreachable_nodes): Don't call cgraph_get_create_node on VAR_DECLs. * g++.dg/ipa/pr60457.C: New test. --- gcc/ipa.c.jj 2014-03-03 08:24:32.000000000 +0100 +++ gcc/ipa.c 2014-03-07 17:27:20.193696625 +0100 @@ -354,7 +354,8 @@ symtab_remove_unreachable_nodes (bool be node->aux = (void *)2; else { - if (DECL_ABSTRACT_ORIGIN (node->decl)) + if (TREE_CODE (node->decl) == FUNCTION_DECL + && DECL_ABSTRACT_ORIGIN (node->decl)) { struct cgraph_node *origin_node = cgraph_get_create_node (DECL_ABSTRACT_ORIGIN (node->decl)); --- gcc/testsuite/g++.dg/ipa/pr60457.C.jj 2014-03-07 17:32:26.517037376 +0100 +++ gcc/testsuite/g++.dg/ipa/pr60457.C 2014-03-07 17:32:10.000000000 +0100 @@ -0,0 +1,17 @@ +// PR ipa/60457 +// { dg-do compile } + +template <class T> +struct A +{ +}; + +struct B : A <B> +{ + B (); +}; + +B::B () +{ + const int c[] = { 1, 1 }; +} Jakub