Another error-recovery bug. This time TYPE was error_mark_node but
min_align_of_type accessed it via TYPE_ALIGN. So we better check that we're
operating on a type.
Bootstrapped/regtested on x86_64-linux, ok for trunk?
2016-06-07 Marek Polacek <pola...@redhat.com>
PR c/71418
* c-decl.c (grokdeclarator): Check TYPE_P.
* gcc.dg/noncompile/pr71418.c: New test.
diff --git gcc/c/c-decl.c gcc/c/c-decl.c
index d79802e..ac83e2f 100644
--- gcc/c/c-decl.c
+++ gcc/c/c-decl.c
@@ -6313,7 +6313,7 @@ grokdeclarator (const struct c_declarator *declarator,
}
else if (TREE_CODE (type) == FUNCTION_TYPE)
error_at (loc, "alignment specified for function %qE", name);
- else if (declspecs->align_log != -1)
+ else if (declspecs->align_log != -1 && TYPE_P (type))
{
alignas_align = 1U << declspecs->align_log;
if (alignas_align < min_align_of_type (type))
diff --git gcc/testsuite/gcc.dg/noncompile/pr71418.c
gcc/testsuite/gcc.dg/noncompile/pr71418.c
index e69de29..e3243f3 100644
--- gcc/testsuite/gcc.dg/noncompile/pr71418.c
+++ gcc/testsuite/gcc.dg/noncompile/pr71418.c
@@ -0,0 +1,4 @@
+/* PR c/71418 */
+/* { dg-do compile } */
+
+_Alignas (int) int a[7++]; /* { dg-error "lvalue required" } */
Marek
