Hi! The following testcase ICEs during error recovery, because finish_decl overwrites TREE_TYPE (error_mark_node), which better should stay always to be error_mark_node.
Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk? 2021-04-10 Jakub Jelinek <ja...@redhat.com> PR c/99990 * c-decl.c (finish_decl): Don't overwrite TREE_TYPE of error_mark_node. * gcc.dg/pr99990.c: New test. --- gcc/c/c-decl.c.jj 2021-04-08 18:35:11.101520870 +0200 +++ gcc/c/c-decl.c 2021-04-09 14:23:47.101571709 +0200 @@ -5402,7 +5402,7 @@ finish_decl (tree decl, location_t init_ gcc_unreachable (); } - if (DECL_INITIAL (decl)) + if (DECL_INITIAL (decl) && DECL_INITIAL (decl) != error_mark_node) TREE_TYPE (DECL_INITIAL (decl)) = type; relayout_decl (decl); --- gcc/testsuite/gcc.dg/pr99990.c.jj 2021-04-09 14:43:31.479327858 +0200 +++ gcc/testsuite/gcc.dg/pr99990.c 2021-04-09 14:45:18.201134102 +0200 @@ -0,0 +1,12 @@ +/* PR c/99990 */ +/* { dg-do compile } */ +/* { dg-options "" } */ + +#include <stdarg.h> + +void +foo () +{ + va_arg (0, long); /* { dg-error "first argument to 'va_arg' not of type 'va_list'" } */ + void *b[] = 0; /* { dg-error "invalid initializer" } */ +} Jakub