[Patch, Fortran] PR 42051,43896: [OOP] ICE in gfc_conv_variable, at fortran/trans-expr.c:551

[Patch, Fortran] PR 42051,43896: [OOP] ICE in gfc_conv_variable, at fortran/trans-expr.c:551

[Janus Weil]

[-- Attachment #1: Type: text/plain, Size: 1093 bytes --]

Hi all,

here is an OOP patch by Paul, which I am taking up and posting here because

a) it is simple
b) it has been lying around for a while
c) it fixes two bugs (both of which have been reported by real-world
users. In fact they are duplicates.)
d) Paul seems to have little time for gfortran lately :(
e) I simply want to get this baby in finally :)

[Btw, the original test case in PR 42051 still does not work, but this
is apparently due to some other bug, possibly related or equal to PR
44064.]

The patch has been regtested on x86_64-unknown-linux-gnu. Ok for trunk?

Actually, since it's Paul's patch, I guess I could just approve it
myself (I do indeed think it's fine). Therefore I'll just commit it
tomorrow on behalf of Paul if no one objects by then.

Cheers,
Janus



2010-06-11  Paul Thomas  <pault@gcc.gnu.org>

	PR fortran/42051
	PR fortran/43896
	* trans-expr.c (gfc_conv_derived_to_class): Handle array-valued
	functions with CLASS formal arguments.


2010-06-11  Paul Thomas  <pault@gcc.gnu.org>

	PR fortran/42051
	PR fortran/43896
	* gfortran.dg/class_23.f03: New test.

[-- Attachment #2: pr42051.diff --]
[-- Type: application/octet-stream, Size: 659 bytes --]

Index: gcc/fortran/trans-expr.c
===================================================================
--- gcc/fortran/trans-expr.c	(revision 160588)
+++ gcc/fortran/trans-expr.c	(working copy)
@@ -2492,12 +2492,14 @@ gfc_conv_derived_to_class (gfc_se *parmse, gfc_exp
   ss = gfc_walk_expr (e);
   if (ss == gfc_ss_terminator)
     {
+      parmse->ss = NULL;
       gfc_conv_expr_reference (parmse, e);
       tmp = fold_convert (TREE_TYPE (ctree), parmse->expr);
       gfc_add_modify (&parmse->pre, ctree, tmp);
     }
   else
     {
+      parmse->ss = ss;
       gfc_conv_expr (parmse, e);
       gfc_add_modify (&parmse->pre, ctree, parmse->expr);
     }

[-- Attachment #3: class_23.f03 --]
[-- Type: application/octet-stream, Size: 470 bytes --]

! { dg-do compile }
!
! PR 42051: [OOP] ICE on array-valued function with CLASS formal argument
!
! Original test case by Damian Rouson <damian@rouson.net>
! Modified by Janus Weil <janus@gcc.gnu.org>

  type grid
  end type 

contains

  function return_x(this) result(this_x)
    class(grid) :: this
    real  ,dimension(1) :: this_x
  end function

  subroutine output()
    type(grid) :: mesh
    real ,dimension(1) :: x
    x = return_x(mesh)
  end subroutine

end

Re: [Patch, Fortran] PR 42051,43896: [OOP] ICE in gfc_conv_variable, at fortran/trans-expr.c:551

[Paul Richard Thomas]

Janus,


> b) it has been lying around for a while

Indeed, I have been trying to get back to completing the incorporation
of arrays....

> c) it fixes two bugs (both of which have been reported by real-world
> users. In fact they are duplicates.)
> d) Paul seems to have little time for gfortran lately :(

...but, as you say, I have little time for gfortran of late.

> e) I simply want to get this baby in finally :)
>
> [Btw, the original test case in PR 42051 still does not work, but this
> is apparently due to some other bug, possibly related or equal to PR
> 44064.]
>
> The patch has been regtested on x86_64-unknown-linux-gnu. Ok for trunk?
>
> Actually, since it's Paul's patch, I guess I could just approve it
> myself (I do indeed think it's fine). Therefore I'll just commit it
> tomorrow on behalf of Paul if no one objects by then.

I think that approach would be appropriate.

Thanks

Paul

Re: [Patch, Fortran] PR 42051,43896: [OOP] ICE in gfc_conv_variable, at fortran/trans-expr.c:551

[Paul Richard Thomas]

[-- Attachment #1: Type: text/plain, Size: 1301 bytes --]

Dear Janus,

Please find attached a development of a fix for pr41539, which touches
the same place as the patch that you have forwarded this morning.  I
seem to recall that it fixes PR42051 as well.  I'll check at
lunchtime.

Cheers

Paul

>> b) it has been lying around for a while
>
> Indeed, I have been trying to get back to completing the incorporation
> of arrays....
>
>> c) it fixes two bugs (both of which have been reported by real-world
>> users. In fact they are duplicates.)
>> d) Paul seems to have little time for gfortran lately :(
>
> ...but, as you say, I have little time for gfortran of late.
>
>> e) I simply want to get this baby in finally :)
>>
>> [Btw, the original test case in PR 42051 still does not work, but this
>> is apparently due to some other bug, possibly related or equal to PR
>> 44064.]
>>
>> The patch has been regtested on x86_64-unknown-linux-gnu. Ok for trunk?
>>
>> Actually, since it's Paul's patch, I guess I could just approve it
>> myself (I do indeed think it's fine). Therefore I'll just commit it
>> tomorrow on behalf of Paul if no one objects by then.
>
> I think that approach would be appropriate.
>
> Thanks
>
> Paul
>



-- 
The knack of flying is learning how to throw yourself at the ground and miss.
       --Hitchhikers Guide to the Galaxy

[-- Attachment #2: submit.diff --]
[-- Type: text/x-patch, Size: 5831 bytes --]

Index: gcc/fortran/interface.c
===================================================================
--- gcc/fortran/interface.c	(revision 157307)
+++ gcc/fortran/interface.c	(working copy)
@@ -1447,6 +1447,23 @@
   if (symbol_rank (formal) == actual->rank)
     return 1;
 
+  if (formal->ts.type == BT_CLASS)
+    {
+      int formal_rank;
+      formal_rank = formal->ts.u.derived->components->as
+			? formal->ts.u.derived->components->as->rank : 0;
+      if (formal_rank == actual->rank)
+   	return 1;
+      else
+	{
+	  if (where)
+	    gfc_error ("Rank mismatch in argument '%s' at %L (%d and %d)",
+		       formal->name, &actual->where, formal_rank,
+		       actual->rank);
+	  return 0;
+	}
+    }
+
   rank_check = where != NULL && !is_elemental && formal->as
 	       && (formal->as->type == AS_ASSUMED_SHAPE
 		   || formal->as->type == AS_DEFERRED)
Index: gcc/fortran/match.c
===================================================================
--- gcc/fortran/match.c	(revision 157307)
+++ gcc/fortran/match.c	(working copy)
@@ -2394,7 +2394,7 @@
 gfc_match_allocate (void)
 {
   gfc_alloc *head, *tail;
-  gfc_expr *stat, *errmsg, *tmp, *source;
+  gfc_expr *stat, *errmsg, *tmp, *source, *e;
   gfc_typespec ts;
   gfc_symbol *sym;
   match m;
@@ -2455,6 +2455,18 @@
 	  goto cleanup;
 	}
 
+      /* A class object's array reference changes the expression type to that
+	 of the declared type. Change it back to the class type for allocate
+	 expressions.  */
+      e = tail->expr;
+      if (e->symtree->n.sym->ts.type == BT_CLASS
+	    && e->ts.type == BT_DERIVED
+	    && e->ref && e->ref->type == REF_COMPONENT
+	    && strcmp (e->ref->u.c.component->name, "$data") == 0
+	    && (!e->ref->next
+		   || (e->ref->next->type == REF_ARRAY && !e->ref->next->next)))
+	e->ts = e->symtree->n.sym->ts;
+
       /* The ALLOCATE statement had an optional typespec.  Check the
 	 constraints.  */
       if (ts.type != BT_UNKNOWN)
Index: gcc/fortran/primary.c
===================================================================
--- gcc/fortran/primary.c	(revision 157307)
+++ gcc/fortran/primary.c	(working copy)
@@ -1738,7 +1738,7 @@
 {
   char name[GFC_MAX_SYMBOL_LEN + 1];
   gfc_ref *substring, *tail;
-  gfc_component *component;
+  gfc_component *component = NULL;
   gfc_symbol *sym = primary->symtree->n.sym;
   match m;
   bool unknown;
@@ -1754,14 +1754,30 @@
       || (sym->ts.type == BT_CLASS
 	  && sym->ts.u.derived->components->attr.dimension))
     {
+      if (sym->ts.type == BT_CLASS && gfc_peek_ascii_char () == '(')
+	{
+	  component = gfc_find_component (sym->ts.u.derived, "$data",
+					  true, true);
+	  tail = extend_ref (primary, tail);
+	  tail->type = REF_COMPONENT;
+	  tail->u.c.component = component;
+	  tail->u.c.sym = sym;
+	}
+
       /* In EQUIVALENCE, we don't know yet whether we are seeing
 	 an array, character variable or array of character
 	 variables.  We'll leave the decision till resolve time.  */
       tail = extend_ref (primary, tail);
       tail->type = REF_ARRAY;
 
-      m = gfc_match_array_ref (&tail->u.ar, equiv_flag ? NULL : sym->as,
-			       equiv_flag);
+      if (sym->ts.type == BT_CLASS)
+	m = gfc_match_array_ref (&tail->u.ar, equiv_flag ? NULL :
+				 sym->ts.u.derived->components->as,
+				 equiv_flag);
+      else
+	m = gfc_match_array_ref (&tail->u.ar, equiv_flag ? NULL :
+				 sym->as, equiv_flag);
+
       if (m != MATCH_YES)
 	return m;
 
@@ -1777,7 +1793,10 @@
 	}
     }
 
-  primary->ts = sym->ts;
+  if (sym->ts.type == BT_CLASS && component)
+    primary->ts = component->ts;
+  else
+    primary->ts = sym->ts;
 
   if (equiv_flag)
     return MATCH_YES;
@@ -2757,7 +2776,8 @@
       /* If the symbol has a dimension attribute, the expression is a
 	 variable.  */
 
-      if (sym->attr.dimension)
+      if (sym->attr.dimension
+	    || (sym->ts.type == BT_CLASS && sym->ts.u.derived->components->attr.dimension))
 	{
 	  if (gfc_add_flavor (&sym->attr, FL_VARIABLE,
 			      sym->name, NULL) == FAILURE)
Index: gcc/fortran/trans-expr.c
===================================================================
--- gcc/fortran/trans-expr.c	(revision 157307)
+++ gcc/fortran/trans-expr.c	(working copy)
@@ -2596,14 +2596,28 @@
 
   /* Set the vptr.  */
   cmp = gfc_find_component (declared, "$vptr", true, true);
-  ctree = fold_build3 (COMPONENT_REF, TREE_TYPE (cmp->backend_decl),
-		       var, cmp->backend_decl, NULL_TREE);
 
-  /* Remember the vtab corresponds to the derived type
-    not to the class declared type.  */
-  vtab = gfc_find_derived_vtab (e->ts.u.derived);
-  gcc_assert (vtab);
-  tmp = gfc_build_addr_expr (NULL_TREE, gfc_get_symbol_decl (vtab));
+  /* Remember the vtab corresponds to the derived type not to the
+     class declared type, unless this is an array reference to a
+     class object.  */
+  if (((e->expr_type == EXPR_VARIABLE) || (e->expr_type == EXPR_FUNCTION))
+	 && e->symtree->n.sym->ts.type == BT_CLASS)
+    {
+      tmp = gfc_get_symbol_decl (e->symtree->n.sym);
+      ctree = fold_build3 (COMPONENT_REF, TREE_TYPE (cmp->backend_decl),
+			   tmp, cmp->backend_decl, NULL_TREE);
+      tmp = fold_build3 (COMPONENT_REF, TREE_TYPE (cmp->backend_decl),
+			 tmp, cmp->backend_decl, NULL_TREE);    
+    }
+  else
+    {
+      ctree = fold_build3 (COMPONENT_REF, TREE_TYPE (cmp->backend_decl),
+			   var, cmp->backend_decl, NULL_TREE);
+      vtab = gfc_find_derived_vtab (e->ts.u.derived);
+      gcc_assert (vtab);
+      tmp = gfc_build_addr_expr (NULL_TREE, gfc_get_symbol_decl (vtab));
+    }
+
   gfc_add_modify (&parmse->pre, ctree,
 		  fold_convert (TREE_TYPE (ctree), tmp));
 
@@ -2620,7 +2634,7 @@
     }
   else
     {
-      gfc_conv_expr (parmse, e);
+      gfc_conv_expr_descriptor (parmse, e, ss);
       gfc_add_modify (&parmse->pre, ctree, parmse->expr);
     }
 

Re: [Patch, Fortran] PR 42051,43896: [OOP] ICE in gfc_conv_variable, at fortran/trans-expr.c:551

[Paul Richard Thomas]

[-- Attachment #1: Type: text/plain, Size: 426 bytes --]

Please find attached an updated version of the patch that I sent this
morning.  It applies to revision 159428.

It fixes pr41539, pr42051_1.f90 and pr43896.f90, up to defined
operators for classes.

I was in the midst of tackling pr41951, when I ground to a halt.

I'll take a further look at it over the weekend.

(Janus, feel free to commit the posted patch.  I will modify this one
to suit - thanks, by the way :-) )

Paul

[-- Attachment #2: submit.diff --]
[-- Type: text/x-patch, Size: 5714 bytes --]

Index: gcc/fortran/interface.c
===================================================================
--- gcc/fortran/interface.c	(revision 159428)
+++ gcc/fortran/interface.c	(working copy)
@@ -1507,6 +1507,23 @@
   if (symbol_rank (formal) == actual->rank)
     return 1;
 
+  if (formal->ts.type == BT_CLASS)
+    {
+      int formal_rank;
+      formal_rank = formal->ts.u.derived->components->as
+			? formal->ts.u.derived->components->as->rank : 0;
+      if (formal_rank == actual->rank)
+   	return 1;
+      else
+	{
+	  if (where)
+	    gfc_error ("Rank mismatch in argument '%s' at %L (%d and %d)",
+		       formal->name, &actual->where, formal_rank,
+		       actual->rank);
+	  return 0;
+	}
+    }
+
   rank_check = where != NULL && !is_elemental && formal->as
 	       && (formal->as->type == AS_ASSUMED_SHAPE
 		   || formal->as->type == AS_DEFERRED)
Index: gcc/fortran/trans-expr.c
===================================================================
--- gcc/fortran/trans-expr.c	(revision 159428)
+++ gcc/fortran/trans-expr.c	(working copy)
@@ -2479,15 +2479,28 @@
 
   /* Set the vptr.  */
   cmp = gfc_find_component (declared, "$vptr", true, true);
-  ctree = fold_build3 (COMPONENT_REF, TREE_TYPE (cmp->backend_decl),
-		       var, cmp->backend_decl, NULL_TREE);
+  
+  /* Remember the vtab corresponds to the derived type not to the
+     class declared type, unless this is an array reference to a
+     class object.  */
+  if (((e->expr_type == EXPR_VARIABLE) || (e->expr_type == EXPR_FUNCTION))
+	 && e->symtree->n.sym->ts.type == BT_CLASS)
+    {
+      tmp = gfc_get_symbol_decl (e->symtree->n.sym);
+      ctree = fold_build3 (COMPONENT_REF, TREE_TYPE (cmp->backend_decl),
+			   tmp, cmp->backend_decl, NULL_TREE);
+      tmp = fold_build3 (COMPONENT_REF, TREE_TYPE (cmp->backend_decl),
+			 tmp, cmp->backend_decl, NULL_TREE);    
+    }
+  else
+    {
+      ctree = fold_build3 (COMPONENT_REF, TREE_TYPE (cmp->backend_decl),
+			   var, cmp->backend_decl, NULL_TREE);
+      vtab = gfc_find_derived_vtab (e->ts.u.derived, true);
+      gcc_assert (vtab);
+      tmp = gfc_build_addr_expr (NULL_TREE, gfc_get_symbol_decl (vtab));
+    }
 
-  /* Remember the vtab corresponds to the derived type
-    not to the class declared type.  */
-  vtab = gfc_find_derived_vtab (e->ts.u.derived, true);
-  gcc_assert (vtab);
-  gfc_trans_assign_vtab_procs (&parmse->pre, e->ts.u.derived, vtab);
-  tmp = gfc_build_addr_expr (NULL_TREE, gfc_get_symbol_decl (vtab));
   gfc_add_modify (&parmse->pre, ctree,
 		  fold_convert (TREE_TYPE (ctree), tmp));
 
@@ -2498,13 +2511,15 @@
   ss = gfc_walk_expr (e);
   if (ss == gfc_ss_terminator)
     {
+      parmse->ss = NULL;
       gfc_conv_expr_reference (parmse, e);
       tmp = fold_convert (TREE_TYPE (ctree), parmse->expr);
       gfc_add_modify (&parmse->pre, ctree, tmp);
     }
   else
     {
-      gfc_conv_expr (parmse, e);
+      parmse->ss = ss;
+      gfc_conv_expr_descriptor (parmse, e, ss);
       gfc_add_modify (&parmse->pre, ctree, parmse->expr);
     }
 
Index: gcc/fortran/match.c
===================================================================
--- gcc/fortran/match.c	(revision 159428)
+++ gcc/fortran/match.c	(working copy)
@@ -2679,7 +2679,7 @@
 gfc_match_allocate (void)
 {
   gfc_alloc *head, *tail;
-  gfc_expr *stat, *errmsg, *tmp, *source;
+  gfc_expr *stat, *errmsg, *tmp, *source, *e;
   gfc_typespec ts;
   gfc_symbol *sym;
   match m;
@@ -2740,6 +2740,18 @@
 	  goto cleanup;
 	}
 
+      /* A class object's array reference changes the expression type to that
+	 of the declared type. Change it back to the class type for allocate
+	 expressions.  */
+      e = tail->expr;
+      if (e->symtree->n.sym->ts.type == BT_CLASS
+	    && e->ts.type == BT_DERIVED
+	    && e->ref && e->ref->type == REF_COMPONENT
+	    && strcmp (e->ref->u.c.component->name, "$data") == 0
+	    && (!e->ref->next
+		   || (e->ref->next->type == REF_ARRAY && !e->ref->next->next)))
+	e->ts = e->symtree->n.sym->ts;
+
       /* The ALLOCATE statement had an optional typespec.  Check the
 	 constraints.  */
       if (ts.type != BT_UNKNOWN)
Index: gcc/fortran/primary.c
===================================================================
--- gcc/fortran/primary.c	(revision 159428)
+++ gcc/fortran/primary.c	(working copy)
@@ -1725,7 +1725,7 @@
 {
   char name[GFC_MAX_SYMBOL_LEN + 1];
   gfc_ref *substring, *tail;
-  gfc_component *component;
+  gfc_component *component = NULL;
   gfc_symbol *sym = primary->symtree->n.sym;
   match m;
   bool unknown;
@@ -1759,6 +1759,16 @@
       || (sym->ts.type == BT_CLASS
 	  && sym->ts.u.derived->components->attr.dimension))
     {
+      if (sym->ts.type == BT_CLASS && gfc_peek_ascii_char () == '(')
+	{
+	  component = gfc_find_component (sym->ts.u.derived, "$data",
+					  true, true);
+	  tail = extend_ref (primary, tail);
+	  tail->type = REF_COMPONENT;
+	  tail->u.c.component = component;
+	  tail->u.c.sym = sym;
+	}
+
       /* In EQUIVALENCE, we don't know yet whether we are seeing
 	 an array, character variable or array of character
 	 variables.  We'll leave the decision till resolve time.  */
@@ -1782,7 +1792,10 @@
 	}
     }
 
-  primary->ts = sym->ts;
+  if (sym->ts.type == BT_CLASS && component)
+    primary->ts = component->ts;
+  else
+    primary->ts = sym->ts;
 
   if (equiv_flag)
     return MATCH_YES;
@@ -2765,7 +2778,8 @@
       /* If the symbol has a dimension attribute, the expression is a
 	 variable.  */
 
-      if (sym->attr.dimension)
+      if (sym->attr.dimension
+	    || (sym->ts.type == BT_CLASS && sym->ts.u.derived->components->attr.dimension))
 	{
 	  if (gfc_add_flavor (&sym->attr, FL_VARIABLE,
 			      sym->name, NULL) == FAILURE)

Re: [Patch, Fortran] PR 42051,43896: [OOP] ICE in gfc_conv_variable, at fortran/trans-expr.c:551

[Janus Weil]

> (Janus, feel free to commit the posted patch.  I will modify this one
> to suit - thanks, by the way :-) )


Did that (r160622).

Thanks,
Janus

Re: [Patch, Fortran] PR 42051,43896: [OOP] ICE in gfc_conv_variable, at fortran/trans-expr.c:551

[Dominique Dhumieres]

Paul,

Your patch in http://gcc.gnu.org/ml/fortran/2010-06/msg00112.html
breaks the tests for pr44465 while fixing various problems:
several ICEs for tests in pr41951 disappear, but are replaced with
questionable errors, others give 'Internal Error at (1)', e.g.
a test for pr42888, now the first test in pr42051 compiles and
links along with some others (no more errors "Undefined symbols:
"_vtab$...").

Dominique

Re: [Patch, Fortran] PR 42051,43896: [OOP] ICE in gfc_conv_variable, at fortran/trans-expr.c:551

[Dominique Dhumieres]

With the patch in http://gcc.gnu.org/ml/fortran/2010-06/msg00112.html
I have the following failures in the test suite:

[macbook] f90/bug% gfc /opt/gcc/work/gcc/testsuite/gfortran.dg/allocate_derived_1.f90
/opt/gcc/work/gcc/testsuite/gfortran.dg/allocate_derived_1.f90:35.12:
...
end
   1
Internal Error at (1):
find_array_spec(): Component not found

and

[macbook] f90/bug% gfc -O /opt/gcc/work/gcc/testsuite/gfortran.dg/dynamic_dispatch_4.f03
/opt/gcc/work/gcc/testsuite/gfortran.dg/dynamic_dispatch_4.f03: In function 'getit':
/opt/gcc/work/gcc/testsuite/gfortran.dg/dynamic_dispatch_4.f03:79:0: error: statement makes a memory store, but has no VDEFS
a_3.$vptr = D.1778_2;
/opt/gcc/work/gcc/testsuite/gfortran.dg/dynamic_dispatch_4.f03:79:0: internal compiler error: verify_ssa failed

Dominique