[Bug fortran/53298] ICE in gfc_conv_scalarized_array_ref for ARRAY + substring

["markeggleston at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53298

--- Comment #14 from markeggleston at gcc dot gnu.org ---
The test case in comment 7 proved trickier to track down.

The ICE occurs in this code:

  /* Components can correspond to fields of different containing
     types, as components are created without context, whereas
     a concrete use of a component has the type of decl as context.
     So, if the type doesn't match, we search the corresponding
     FIELD_DECL in the parent type.  To not waste too much time
     we cache this result in norestrict_decl.
     On the other hand, if the context is a UNION or a MAP (a
     RECORD_TYPE within a UNION_TYPE) always use the given FIELD_DECL.  */

  if (context != TREE_TYPE (decl)
      && !(   TREE_CODE (TREE_TYPE (field)) == UNION_TYPE /* Field is union */
          || TREE_CODE (context) == UNION_TYPE)) /* Field is map */
    {
      tree f2 = c->norestrict_decl;
      if (!f2 || DECL_FIELD_CONTEXT (f2) != TREE_TYPE (decl))
==>     for (f2 = TYPE_FIELDS (TREE_TYPE (decl)); f2; f2 = DECL_CHAIN (f2))
          if (TREE_CODE (f2) == FIELD_DECL
              && DECL_NAME (f2) == DECL_NAME (field))
            break;
      gcc_assert (f2);
      c->norestrict_decl = f2;
      field = f2;
    }


The ICE occurs at the line marked with ==>, the assignment f2 = TYPE_FIELDS
(TREE_TYPE (decl)) hides a call to a routine that returns a null pointer which
is then used causing the ICE.

I speculated that the code dealing with f2 should not have been executed. I
noted that in the comments "On the other hand, if the context is a UNION or a
MAP (a RECORD_TYPE within a UNION_TYPE) always use the given FIELD_DECL."  This
does not match the code that follows.

The following change:

trans-expr.c

@@ -2474,8 +2474,8 @@ gfc_conv_component_ref (gfc_se * se, gfc_ref * ref)
      RECORD_TYPE within a UNION_TYPE) always use the given FIELD_DECL.  */

   if (context != TREE_TYPE (decl)
-      && !(   TREE_CODE (TREE_TYPE (field)) == UNION_TYPE /* Field is union */
-           || TREE_CODE (context) == UNION_TYPE))         /* Field is map */
+      && (   TREE_CODE (context) == UNION_TYPE /* Field is union */
+          || TREE_CODE (context) == MAP_TYPE)) /* Field is map */
     {
       tree f2 = c->norestrict_decl;
       if (!f2 || DECL_FIELD_CONTEXT (f2) != TREE_TYPE (decl))

matches the comment and also fixes the ICE for the test case in comment 7.

make -j 8 check-fortran was looking good until test case failures were reported
for:

gfortran.dg/finalize_35.f90
gfortran.dg/finalize_36.f90