[Bug fortran/50410] [8/9/10/11 Regression] ICE in record_reference, pointer variable in data statement

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=50410

--- Comment #42 from anlauf at gcc dot gnu.org ---
(In reply to Dominique d'Humieres from comment #41)
> In my working tree I had the following patch

This still ICEs on comment#23 z1.

Slightly rewriting that case, one gets a reasonable error message:

program p
  type ta
     integer :: a
  end type ta
  type t
     type(ta), pointer :: b
  end type t
  type(t) :: u = t(ta(1))
end

pr50410_23x.f90:8:19:

    8 |   type(t) :: u = t(ta(1))
      |                   1
Error: The element in the structure constructor at (1), for pointer component
'b' should be a POINTER or a TARGET


Digging further, one sees that the error is generated for z1 but not emitted.
The following patch does the magic (not regtested):

diff --git a/gcc/fortran/expr.c b/gcc/fortran/expr.c
index 8daa7bb8d06..0a995ec3ae7 100644
--- a/gcc/fortran/expr.c
+++ b/gcc/fortran/expr.c
@@ -3087,6 +3087,9 @@ gfc_reduce_init_expr (gfc_expr *expr)
     t = gfc_check_init_expr (expr);
   gfc_init_expr_flag = false;

+  if (gfc_error_check())
+    return false;
+
   if (!t || !expr)
     return false;