[Bug fortran/41139] New: [4.5 Regression] a procedure pointer call as actual argument

[janus at gcc dot gnu dot org]

PROGRAM test

 PROCEDURE(add), POINTER :: f

 ! Passing the function works
 print *,greater(4.,add(1.,2.))

 ! Passing the procedure pointer fails
 f => add
 print *,greater(4.,f(1.,2.))

CONTAINS

 REAL FUNCTION add(x,y)
   REAL, INTENT(in) :: x,y
   print *,"add:",x,y
   add = x+y
 END FUNCTION add

 LOGICAL FUNCTION greater(x,y)
   REAL, INTENT(in) :: x, y
   greater = (x > y)
 END FUNCTION greater

END PROGRAM test


This code produces the following output:

 add:   1.0000000       2.0000000    
 T
 add:   1.0000000       2.0000000    
Segmentation fault


Looking at the ouput of -fdump-tree-original, one can see a difference between
the two calls to 'greater':

      D.1563 = add (&C.1561, &C.1562);
      D.1564 = greater (&C.1560, &D.1563);

      D.1570 = greater (&C.1567, f (&C.1568, &C.1569));

I.e. in the first case a temporary variable is inserted for the result of
'add', while in the second case this is not the case (which is the reason for
the segfault).

gfortran 4.4 behaves correctly.

Reported by Barron Bichon.


-- 
           Summary: [4.5 Regression] a procedure pointer call as actual
                    argument
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: janus at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41139