[Patch, fortran] PR20903 - types shouldn't propagate into interfaces

[Paul Thomas <paulthomas2 at wanadoo dot fr>]

:ADDPATCH fortran:

To quote the standard:

"An interface body specifies all of the procedure's characteristics and 
these
shall be consistent with those specified in the procedure definition, except
that the interface may specify a procedure that is not pure if the procedure
is defined to be pure(12.3.2)."

At present, gfortran violates this by allowing derived types to be 
declared outside of the scope of the interface body.  The patch does not 
need any further description.  The testcase checks both the error and 
that nothing is broken; specifically, it checks that internally declared 
and use associated derived types are accepted.

Regtested on FC5/Athlon.  OK for trunk and 4.1?

Paul

2006-07-11  Paul Thomas  <pault@gcc.gnu.org>

   PR fortran/20903
   * decl.c (variable_decl): Add error if a derived type is not
   from the current namespace if the namespace is an interface
   body.

2006-07-11  Paul Thomas  <pault@gcc.gnu.org>

   PR fortran/20903
   * gfortran.dg/interface_derived_type_1.f90: New test.

Index: gcc/fortran/decl.c
===================================================================
*** gcc/fortran/decl.c	(revision 115317)
--- gcc/fortran/decl.c	(working copy)
*************** variable_decl (int elem)
*** 1176,1181 ****
--- 1176,1195 ----
        goto cleanup;
      }
  
+   /* An interface body specifies all of the procedure?s characteristics and these
+      shall be consistent with those specified in the procedure definition, except
+      that the interface may specify a procedure that is not pure if the procedure
+      is defined to be pure(12.3.2).  */
+   if (current_ts.type == BT_DERIVED
+ 	&& gfc_current_ns->proc_name->attr.if_source == IFSRC_IFBODY
+ 	&& current_ts.derived->ns != gfc_current_ns)
+     {
+       gfc_error ("the type of '%s' at %C has not been declared within the "
+ 		 "interface", name, &sym->declared_at);
+       m = MATCH_ERROR;
+       goto cleanup;
+     }
+ 
    /* In functions that have a RESULT variable defined, the function
       name always refers to function calls.  Therefore, the name is
       not allowed to appear in specification statements.  */

! { dg-do compile }
! Test the fix for PR20903, in which derived types could be host associated within
! interface bodies.
!
! Contributed by Joost VandeVondele <jv244@cam.ac.uk>
! 
module test
  implicit none
  type fcnparms
    integer :: i
  end type fcnparms
contains
  subroutine sim_1(func1,params)
    interface
      function func1(fparams)
        type(fcnparms) :: fparams ! { dg-error "not been declared within the interface" }
        real :: func1
      end function func1
    end interface
    type(fcnparms)     :: params
   end subroutine sim_1

  subroutine sim_2(func2,params)
    interface
      function func2(fparams)     ! This is OK because of the derived type decl.
        type fcnparms
          integer :: i
        end type fcnparms
        type(fcnparms)  :: fparams
        real :: func2
      end function func2
    end interface
    type(fcnparms)      :: params ! This is OK, of course
   end subroutine sim_2
end module  test

module type_decl
  implicit none
  type fcnparms
    integer :: i
  end type fcnparms
end module type_decl

subroutine sim_3(func3,params)
  use type_decl
  interface
    function func3(fparams)
      use type_decl
      type(fcnparms)   :: fparams ! This is OK - use associated
      real :: func3
    end function func3
  end interface
  type(fcnparms)       :: params  !         -ditto-
end subroutine sim_3

2006-07-11  Paul Thomas  <pault@gcc.gnu.org>

	PR fortran/20903
	* decl.c (variable_decl): Add error if a derived type is not
	from the current namespace if the namespace is an interface
	body.

2006-07-11  Paul Thomas  <pault@gcc.gnu.org>

	PR fortran/20903
	* gfortran.dg/interface_derived_type_1.f90: New test.