[PATCH, Fortran] Re-fix PR 31292

[PATCH, Fortran] Re-fix PR 31292

[Steve Kargl]

[-- Attachment #1: Type: text/plain, Size: 432 bytes --]

The attached patch was built and regression tested
on x86_64-*-freebsd.  This patch partially un-fixes
the patch applied to fix PR fortran/31292.  I've
re-open that PR to track this change.


2009-08-28  Steven G. Kargl <kargl@gcc.gnu.org>

	* gfortran.dg/module_procedure_1.f90: New test.

2009-08-28  Steven G. Kargl <kargl@gcc.gnu.org>

	* fortran/decl.c: Module procedure can appear in PROGRAM and
	subprogram scope.

-- 
Steve

[-- Attachment #2: modproc.diff --]
[-- Type: text/x-diff, Size: 2088 bytes --]

Index: gcc/testsuite/gfortran.dg/module_procedure_1.f90
===================================================================
--- gcc/testsuite/gfortran.dg/module_procedure_1.f90	(revision 0)
+++ gcc/testsuite/gfortran.dg/module_procedure_1.f90	(revision 0)
@@ -0,0 +1,53 @@
+! { dg-do run }
+! Modified program from http://groups.google.com/group/\
+! comp.lang.fortran/browse_frm/thread/423e4392dc965ab7#
+!
+module myoperator
+   contains
+      function dadd(arg1,arg2)
+         integer ::dadd(2)
+         integer, intent(in) :: arg1(2), arg2(2)
+         dadd(1)=arg1(1)+arg2(1)
+         dadd(2)=arg1(2)+arg2(2)
+      end function dadd
+end module myoperator
+
+program test_interface
+
+   use myoperator
+
+   implicit none
+
+   interface operator (.myadd.)
+      module procedure dadd
+   end interface
+
+   integer input1(2), input2(2), mysum(2)
+
+   input1 = (/0,1/)
+   input2 = (/3,3/)
+   mysum = input1 .myadd. input2
+   if (mysum(1) /= 3 .and. mysum(2) /= 4) call abort 
+
+   call test_sub(input1, input2)
+
+end program test_interface 
+
+subroutine test_sub(input1, input2)
+
+   use myoperator
+
+   implicit none
+
+   interface operator (.myadd.)
+      module procedure dadd
+   end interface
+
+   integer, intent(in) :: input1(2), input2(2)
+   integer mysum(2)
+
+   mysum = input1 .myadd. input2
+   if (mysum(1) /= 3 .and. mysum(2) /= 4) call abort 
+
+end subroutine test_sub 
+! { dg-final { cleanup-modules "myoperator" } }
Index: gcc/fortran/decl.c
===================================================================
--- gcc/fortran/decl.c	(revision 151144)
+++ gcc/fortran/decl.c	(working copy)
@@ -6481,7 +6481,10 @@ gfc_match_modproc (void)
 
   module_ns = gfc_current_ns->parent;
   for (; module_ns; module_ns = module_ns->parent)
-    if (module_ns->proc_name->attr.flavor == FL_MODULE)
+    if (module_ns->proc_name->attr.flavor == FL_MODULE
+	|| module_ns->proc_name->attr.flavor == FL_PROGRAM
+	|| (module_ns->proc_name->attr.flavor == FL_PROCEDURE
+	    && !module_ns->proc_name->attr.contained))
       break;
 
   if (module_ns == NULL)

Re: [PATCH, Fortran] Re-fix PR 31292

[Jerry DeLisle]

Steve Kargl wrote:
> The attached patch was built and regression tested
> on x86_64-*-freebsd.  This patch partially un-fixes
> the patch applied to fix PR fortran/31292.  I've
> re-open that PR to track this change.
ok if passes regression testing

Jerry

Re: [PATCH, Fortran] Re-fix PR 31292

[Tobias Burnus]

Am 28.08.2009 23:34, schrieb Steve Kargl:
> 2009-08-28  Steven G. Kargl <kargl@gcc.gnu.org>
>
> 	* gfortran.dg/module_procedure_1.f90: New test.
>
> 2009-08-28  Steven G. Kargl <kargl@gcc.gnu.org>
>
> 	* fortran/decl.c: Module procedure can appear in PROGRAM and
> 	subprogram scope.
>   

I think your patch is OK, however, one needs to add a bit more to fully
fix the problem, see test case below.

Tobias


module m
contains
  subroutine modSub()
  end subroutine modSub
end module m

use m
implicit none
  intrinsic sin
  interface gen2
    module procedure sin ! Invalid per C1208, but not detected
!     procedure sin ! OK - valid and accepted
  end interface gen2
  interface gen
    module  procedure modSub ! OK (valid and now accepted)
    module procedure intSub ! Invalid per C1208, gives
         ! misleading error "INTERNAL-PROC procedure at (1) is already
         ! declared as MODULE-PROC procedure"
!    procedure intSub ! OK - valid and accepted
  end interface
contains
  subroutine intSub(a)
    integer :: a
  end subroutine intSub
end

C1208 (R1206) If MODULE appears in a procedure-stmt, each procedure-name
in that statement shall be accessible in the current scope as a
module procedure.

Re: [PATCH, Fortran] Re-fix PR 31292

[Steve Kargl]

On Mon, Aug 31, 2009 at 08:00:10AM +0200, Tobias Burnus wrote:
> Am 28.08.2009 23:34, schrieb Steve Kargl:
> > 2009-08-28  Steven G. Kargl <kargl@gcc.gnu.org>
> >
> > 	* gfortran.dg/module_procedure_1.f90: New test.
> >
> > 2009-08-28  Steven G. Kargl <kargl@gcc.gnu.org>
> >
> > 	* fortran/decl.c: Module procedure can appear in PROGRAM and
> > 	subprogram scope.
> >   
> 
> I think your patch is OK, however, one needs to add a bit more to fully
> fix the problem, see test case below.
> 

I'll hold off on committing my patch until I investigate
some the issues you point out below.

> module m
> contains
>   subroutine modSub()
>   end subroutine modSub
> end module m
> 
> use m
> implicit none
>   intrinsic sin

If the above explicit 'intrinsic sin' statement is missing, then ...

>   interface gen2
>     module procedure sin ! Invalid per C1208, but not detected

is this still invalid?  I would assume the answer is "yes"
because sin is implicitly given the intrinsic attribute.

> 
> C1208 (R1206) If MODULE appears in a procedure-stmt, each procedure-name
> in that statement shall be accessible in the current scope as a
> module procedure.

-- 
Steve

Re: [PATCH, Fortran] Re-fix PR 31292

[Steve Kargl]

On Mon, Aug 31, 2009 at 02:58:18PM -0700, Steve Kargl wrote:
> On Mon, Aug 31, 2009 at 08:00:10AM +0200, Tobias Burnus wrote:
> > Am 28.08.2009 23:34, schrieb Steve Kargl:
> > > 2009-08-28  Steven G. Kargl <kargl@gcc.gnu.org>
> > >
> > > 	* gfortran.dg/module_procedure_1.f90: New test.
> > >
> > > 2009-08-28  Steven G. Kargl <kargl@gcc.gnu.org>
> > >
> > > 	* fortran/decl.c: Module procedure can appear in PROGRAM and
> > > 	subprogram scope.
> > >   
> > 
> > I think your patch is OK, however, one needs to add a bit more to fully
> > fix the problem, see test case below.
> > 
> 
> I'll hold off on committing my patch until I investigate
> some the issues you point out below.
> 
> > module m
> > contains
> >   subroutine modSub()
> >   end subroutine modSub
> > end module m
> > 
> > use m
> > implicit none
> >   intrinsic sin
> 
> If the above explicit 'intrinsic sin' statement is missing, then ...
> 
> >   interface gen2
> >     module procedure sin ! Invalid per C1208, but not detected
> 
> is this still invalid?  I would assume the answer is "yes"
> because sin is implicitly given the intrinsic attribute.
> 
> > 
> > C1208 (R1206) If MODULE appears in a procedure-stmt, each procedure-name
> > in that statement shall be accessible in the current scope as a
> > module procedure.
> 

How's this

troutmask:sgk[213] gfc4x -c t2.f90
t2.f90:8.23:

      module procedure sin  ! Invalid per C1208, but not detected
                       1
Error: Intrinsic procedure at (1) cannot be a MODULE PROCEDURE


-- 
Steve

Re: [PATCH, Fortran] Re-fix PR 31292

[Steve Kargl]

[-- Attachment #1: Type: text/plain, Size: 924 bytes --]

On Fri, Aug 28, 2009 at 02:50:13PM -0700, Jerry DeLisle wrote:
> Steve Kargl wrote:
> >The attached patch was built and regression tested
> >on x86_64-*-freebsd.  This patch partially un-fixes
> >the patch applied to fix PR fortran/31292.  I've
> >re-open that PR to track this change.
> ok if passes regression testing
> 

Here's the patch that I've committed.

2009-09-10  Steven G. Kargl  <kargl@gcc.gnu.org>

	PR fortran/31292
	* gfortran.dg/module_procedure_1.f90: New test.
	* gfortran.dg/module_procedure_2.f90: Ditto.
	* gfortran.dg/generic_14.f90: Move dg-error to new location.

2009-09-10  Steven G. Kargl  <kargl@gcc.gnu.org>

	PR fortran/31292
	* fortran/decl.c(gfc_match_modproc): Check that module procedures
	from a module can USEd in module procedure statements in other
	program units.  Update locus for better error message display.
	Detect intrinsic procedures in module procedure statements.

-- 
Steve

[-- Attachment #2: modproc2.diff --]
[-- Type: text/x-diff, Size: 4191 bytes --]

Index: testsuite/gfortran.dg/module_procedure_2.f90
===================================================================
--- testsuite/gfortran.dg/module_procedure_2.f90	(revision 0)
+++ testsuite/gfortran.dg/module_procedure_2.f90	(revision 0)
@@ -0,0 +1,8 @@
+! { dg-do compile }
+program test
+   implicit none
+   intrinsic sin
+   interface gen2
+      module procedure sin  ! { dg-error "cannot be a MODULE PROCEDURE" }
+   end interface gen2
+end program test
Index: testsuite/gfortran.dg/module_procedure_1.f90
===================================================================
--- testsuite/gfortran.dg/module_procedure_1.f90	(revision 0)
+++ testsuite/gfortran.dg/module_procedure_1.f90	(revision 0)
@@ -0,0 +1,53 @@
+! { dg-do run }
+! Modified program from http://groups.google.com/group/\
+! comp.lang.fortran/browse_frm/thread/423e4392dc965ab7#
+!
+module myoperator
+   contains
+      function dadd(arg1,arg2)
+         integer ::dadd(2)
+         integer, intent(in) :: arg1(2), arg2(2)
+         dadd(1)=arg1(1)+arg2(1)
+         dadd(2)=arg1(2)+arg2(2)
+      end function dadd
+end module myoperator
+
+program test_interface
+
+   use myoperator
+
+   implicit none
+
+   interface operator (.myadd.)
+      module procedure dadd
+   end interface
+
+   integer input1(2), input2(2), mysum(2)
+
+   input1 = (/0,1/)
+   input2 = (/3,3/)
+   mysum = input1 .myadd. input2
+   if (mysum(1) /= 3 .and. mysum(2) /= 4) call abort 
+
+   call test_sub(input1, input2)
+
+end program test_interface 
+
+subroutine test_sub(input1, input2)
+
+   use myoperator
+
+   implicit none
+
+   interface operator (.myadd.)
+      module procedure dadd
+   end interface
+
+   integer, intent(in) :: input1(2), input2(2)
+   integer mysum(2)
+
+   mysum = input1 .myadd. input2
+   if (mysum(1) /= 3 .and. mysum(2) /= 4) call abort 
+
+end subroutine test_sub 
+! { dg-final { cleanup-modules "myoperator" } }
Index: testsuite/gfortran.dg/generic_14.f90
===================================================================
--- testsuite/gfortran.dg/generic_14.f90	(revision 151602)
+++ testsuite/gfortran.dg/generic_14.f90	(working copy)
@@ -85,18 +85,18 @@ end module f
 
 module g
   implicit none
-  external wrong_b            ! { dg-error "has no explicit interface" }
+  external wrong_b
   interface gen_wrong_5
-    module procedure wrong_b  ! wrong, see above
+    module procedure wrong_b  ! { dg-error "has no explicit interface" }
   end interface gen_wrong_5
 end module g
 
 module h
   implicit none
-  external wrong_c            ! { dg-error "has no explicit interface" }
+  external wrong_c
   real wrong_c
   interface gen_wrong_6
-    module procedure wrong_c  ! wrong, see above
+    module procedure wrong_c  ! { dg-error "has no explicit interface" }
   end interface gen_wrong_6
 end module h
 
Index: fortran/decl.c
===================================================================
--- fortran/decl.c	(revision 151602)
+++ fortran/decl.c	(working copy)
@@ -6485,7 +6485,10 @@ gfc_match_modproc (void)
 
   module_ns = gfc_current_ns->parent;
   for (; module_ns; module_ns = module_ns->parent)
-    if (module_ns->proc_name->attr.flavor == FL_MODULE)
+    if (module_ns->proc_name->attr.flavor == FL_MODULE
+	|| module_ns->proc_name->attr.flavor == FL_PROGRAM
+	|| (module_ns->proc_name->attr.flavor == FL_PROCEDURE
+	    && !module_ns->proc_name->attr.contained))
       break;
 
   if (module_ns == NULL)
@@ -6497,6 +6500,7 @@ gfc_match_modproc (void)
 
   for (;;)
     {
+      locus old_locus = gfc_current_locus;
       bool last = false;
 
       m = gfc_match_name (name);
@@ -6517,6 +6521,13 @@ gfc_match_modproc (void)
       if (gfc_get_symbol (name, module_ns, &sym))
 	return MATCH_ERROR;
 
+      if (sym->attr.intrinsic)
+	{
+	  gfc_error ("Intrinsic procedure at %L cannot be a MODULE "
+		     "PROCEDURE", &old_locus);
+	  return MATCH_ERROR;
+	}
+
       if (sym->attr.proc != PROC_MODULE
 	  && gfc_add_procedure (&sym->attr, PROC_MODULE,
 				sym->name, NULL) == FAILURE)
@@ -6526,6 +6537,7 @@ gfc_match_modproc (void)
 	return MATCH_ERROR;
 
       sym->attr.mod_proc = 1;
+      sym->declared_at = old_locus;
 
       if (last)
 	break;

Re: [PATCH, Fortran] Re-fix PR 31292

[Steve Kargl]

On Mon, Aug 31, 2009 at 08:00:10AM +0200, Tobias Burnus wrote:
> Am 28.08.2009 23:34, schrieb Steve Kargl:
> > 2009-08-28  Steven G. Kargl <kargl@gcc.gnu.org>
> >
> > 	* gfortran.dg/module_procedure_1.f90: New test.
> >
> > 2009-08-28  Steven G. Kargl <kargl@gcc.gnu.org>
> >
> > 	* fortran/decl.c: Module procedure can appear in PROGRAM and
> > 	subprogram scope.
> >   
> 
> I think your patch is OK, however, one needs to add a bit more to fully
> fix the problem, see test case below.
> 
> Tobias
> 
> module m
> contains
>   subroutine modSub()
>   end subroutine modSub
> end module m
> 
> use m
> implicit none
>   intrinsic sin
>   interface gen2
>     module procedure sin ! Invalid per C1208, but not detected
> !     procedure sin ! OK - valid and accepted
>   end interface gen2

The patch and one of the testcases that I committed now
catches the C1208 problem mentioned above.

>   interface gen
>     module  procedure modSub ! OK (valid and now accepted)
>     module procedure intSub ! Invalid per C1208, gives
>          ! misleading error "INTERNAL-PROC procedure at (1) is already
>          ! declared as MODULE-PROC procedure"
> !    procedure intSub ! OK - valid and accepted
>   end interface
> contains
>   subroutine intSub(a)
>     integer :: a
>   end subroutine intSub
> end

I did not try to fix this problem because it can't lead to
wrong-code.  That is, an error message is generated although
it is somewhat difficult to untangle its meaning.  If you
feel strongly that this needs some additional attention, then
submit a new PR.

-- 
Steve