[Patch, fortran] PR25532 ICE in gfc_conv_component_ref, at fortran/trans-expr.c:269

[Paul Thomas <paulthomas2 at wanadoo dot fr>]

Erik and Tobi,

> In the meantime, I will sort out a solution.

Now that the Christmas guests have all departed, I can unleash the patch 
for the regression PR25532, as promised:-

The problem and its solution were relatively easy to sort out.  The 
derived type component of the nested derived type, in the submitted 
testscase, was not being referenced in a contained procedure.  
Therefore, its backend declaration was not being copied and the above 
ICE resulted.  The testsuite program is the reduced testcase contributed 
by Erik Edelmann.  I have checked that Martin Reineke's testcase 
compiles correctly.

Regtested on FC3/Athlon1700.

I will commit tomorrow morning to trunk and 4.1, if there are no no 
objections; the patch is nearly "obvious" and remedies an omission 
without changing the principle of the original patch.

Paul

2005-12-28  Paul Thomas  <pault@gcc.gnu.org>

   PR fortran/25532
   * trans-types.c (copy_dt_decls_ifequal): Copy declarations for
   components of derived type components by recursing into
   gfc_get_derived_type.

2005-12-28  Paul Thomas  <pault@gcc.gnu.org>

   PR fortran/25532
   *gfortran.dg/host_used_types_1.f90: Check that host associated
   derived type components of derived types are properly declared
   in contained procedures.

Index: gcc/fortran/trans-types.c
===================================================================
*** gcc/fortran/trans-types.c	(revision 109038)
--- gcc/fortran/trans-types.c	(working copy)
*************** copy_dt_decls_ifequal (gfc_symbol *from,
*** 1414,1421 ****
    to_cm = to->components;
    from_cm = from->components;
  
    for (; to_cm; to_cm = to_cm->next, from_cm = from_cm->next)
!     to_cm->backend_decl = from_cm->backend_decl;
  
    return 1;
  }
--- 1414,1430 ----
    to_cm = to->components;
    from_cm = from->components;
  
+   /* Copy the component declarations.  If a component is itself
+      a derived type, we need a copy of its component delarations.
+      This is done by recursing into gfc_get_derived_type.  This
+      ensures that the component's component declarations have
+      been built.  */
    for (; to_cm; to_cm = to_cm->next, from_cm = from_cm->next)
!     {
!       to_cm->backend_decl = from_cm->backend_decl;
!       if (from_cm->ts.type == BT_DERIVED)
! 	gfc_get_derived_type (to_cm->ts.derived);
!     }
  
    return 1;
  }

! { dg-do compile }
! Tests the fix for PR25532, which was a regression introduced by
! the fix for PR20244.
!
! Contributed by Erik Edelmann  <eedelman@gcc.gnu.org>
module ModelParams
        implicit none

        type ReionizationParams
             real   :: fraction
        end type ReionizationParams

        type CAMBparams
             type(ReionizationParams) :: Reion
         end type CAMBparams

        type(CAMBparams) CP
end module ModelParams


module ThermoData
    use ModelParams
    implicit none

contains

    subroutine inithermo()
        use ModelParams
        if (0 < CP%Reion%fraction) then
        end if
    end subroutine inithermo

! The bug expressed itself in this subroutine because the component type
! information was not being copied from the parent namespace.
    subroutine SetTimeSteps
        if (0 < CP%Reion%fraction) then
        end if
    end subroutine SetTimeSteps

end module ThermoData