[Bug fortran/117901] [ 15 regression] class_transformational_1.f90 with -O3 and -fcheck=bounds gives ICE in make_ssa_name_fn

[pault at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117901

--- Comment #4 from Paul Thomas <pault at gcc dot gnu.org> ---
I held off pushing because I had the scent of the problem. The "fix" of comment
#3 was a dead giveaway. It turns out to be nothing to do with my fix for 102689
but is a latent bug exposed by the testcase. The fix, which regtests fine, is
in trans-array.cc:

@@ -8562,7 +8573,14 @@ gfc_conv_expr_descriptor (gfc_se *se, gfc_expr *expr)

       /* Set the dtype.  */
       tmp = gfc_conv_descriptor_dtype (parm);
-      if (se->unlimited_polymorphic)
+      if (UNLIMITED_POLY (expr) && GFC_ARRAY_TYPE_P (TREE_TYPE (desc)))
+       {
+         dtype = expr->symtree->n.sym->backend_decl;
+         dtype = GFC_DECL_SAVED_DESCRIPTOR (dtype);
+         dtype = gfc_class_data_get (dtype);
+         dtype = gfc_conv_descriptor_dtype (dtype);
+       }
+      else if (se->unlimited_polymorphic)
        dtype = gfc_get_dtype (TREE_TYPE (desc), &loop.dimen);
       else if (expr->ts.type == BT_ASSUMED)
        {

I will now clean up the patch and produce a suitably reduced testcase that
isolates the problem.

Paul