We ICE on the following (invalid) testcase, because the DECL_ARG_TYPE is error_mark_node and useless_type_conversion_p doesn't check that it operates on a type.
Regtested/bootstrapped on x86_64-linux, ok for trunk/4.8/4.7? 2014-01-15 Marek Polacek <pola...@redhat.com> PR middle-end/59827 * cgraph.c (gimple_check_call_args): Don't use DECL_ARG_TYPE if it is error_mark_node. testsuite/ * gcc.dg/pr59827.c: New test. --- gcc/cgraph.c.mp2 2014-01-15 14:24:53.235255954 +0100 +++ gcc/cgraph.c 2014-01-15 14:52:28.818244290 +0100 @@ -3035,6 +3035,7 @@ gimple_check_call_args (gimple stmt, tre break; arg = gimple_call_arg (stmt, i); if (p == error_mark_node + || DECL_ARG_TYPE (p) == error_mark_node || arg == error_mark_node || (!types_compatible_p (DECL_ARG_TYPE (p), TREE_TYPE (arg)) && !fold_convertible_p (DECL_ARG_TYPE (p), arg))) --- gcc/testsuite/gcc.dg/pr59827.c.mp2 2014-01-15 15:06:47.129210564 +0100 +++ gcc/testsuite/gcc.dg/pr59827.c 2014-01-15 15:09:17.295843068 +0100 @@ -0,0 +1,15 @@ +/* PR middle-end/59827 */ +/* { dg-do compile } */ + +int +foo (int p[2][]) /* { dg-error "array type has incomplete element type" } */ +{ + return p[0][0]; +} + +void +bar (void) +{ + int p[2][1]; + foo (p); /* { dg-error "type of formal parameter 1 is incomplete" } */ +} Marek