Original articles by Mark Leair, PGI Compiler Engineer
This is Part 1 of a series of articles:
Polymorphism is a term used in software development to describe a variety of techniques employed by programmers to create flexible and reusable software components. The term is Greek and it loosely translates to "many forms".
In programming languages, a polymorphic object is an entity, such as a variable or a procedure, that can hold or operate on values of differing types during the program's execution. Because a polymorphic object can operate on a variety of values and types, it can also be used in a variety of programs, sometimes with little or no change by the programmer. The idea of write once, run many, also known as code reusability, is an important characteristic to the programming paradigm known as Object-Oriented Programming (OOP).
OOP describes an approach to programming where a program is viewed as a collection of interacting, but mostly independent software components. These software components are known as objects in OOP and they are typically implemented in a programming language as an entity that encapsulates both data and procedures.
A Fortran 90/95 module can be viewed as an object because it can encapsulate both data and procedures. Fortran 2003 (F2003) added the ability for a derived type to encapsulate procedures in addition to data. So, by definition, a derived type can now be viewed as an object as well in F2003.
F2003 also introduced type extension to its derived types. This feature allows F2003 programmers to take advantage of one of the more powerful OOP features known as inheritance. Inheritance allows code reusability through an implied inheritance link in which leaf objects, known as children, reuse components from their parent and ancestor objects. For example,
type shape
integer :: color
logical :: filled
integer :: x
integer :: y
end type shape
type, extends(shape) :: rectangle
integer :: length
integer :: width
end type rectangle
type, extends(rectangle) :: square
end type square
In the example above, we have a square
type that inherits components from rectangle
which inherits components from shape
. The programmer indicates the inheritance relationship with the EXTENDS keyword followed by the name of the parent type in parentheses. A type that EXTENDS another type is known as a type extension (e.g., rectangle
is a type extension of shape
, square
is a type extension of rectangle
and shape
). A type without any EXTENDS keyword is known as a base type (e.g., shape
is a base type).
A type extension inherits all of the components of its parent (and ancestor) types. A type extension can also define additional components as well. For example, rectangle
has a length
and width
component in addition to the color
, filled
, x
, and y
components that were inherited from shape
. The square
type, on the other hand, inherits all of the components from rectangle
and shape
, but does not define any components specific to square
objects. Below is an example on how we may access the color component of square
:
type(square) :: sq ! declare sq as a square object
sq%color ! access color component for sq
sq%rectangle%color ! access color component for sq
sq%reactangle%shape%color ! access color component for sq
Note the three different ways for accessing the color component for sq
. A type extension includes an implicit component with the same name and type as its parent type. This can come in handy when the programmer wants to operate on components specific to a parent type. It also helps illustrate an important relationship between the child and parent types.
We often say the child and parent types have a "is a" relationship. Using our shape
example above, we can say "a square
is a rectangle
", "a rectangle
is a shape
", "a square
is a shape
", and "a shape
is a base type". We can also apply this relationship to the type itself (e.g., "a shape
is a shape
", "a rectangle
is a rectangle
", and "a squar
e is a square
").
Note that the "is a" relationship does not imply the converse. A rectangle
is a shape
, but a shape
is not a rectangle
since there are components found in rectangle
that are not found in shape
. Furthermore, a rectangle
is not a square
because square
has a component not found in rectangle
; the implicit rectangle
parent component.
The "is a" relationship also helps us visualize how polymorphic variables interact with type extensions. The CLASS keyword allows F2003 programmers to create polymorphic variables. A polymorphic variable is a variable whose data type is dynamic at runtime. It must be a pointer variable, allocatable variable, or a dummy argument. Below is an example:
class(shape), pointer :: sh
In the example above, the sh
object can be a pointer to a shape
or any of its type extensions. So, it can be a pointer to a shape
, a rectangle
, a square
, or any future type extension of shape
. As long as the type of the pointer target "is a" shape
, sh
can point to it.
There are two basic types of polymorphism: procedure polymorphism and data polymorphism. Procedure polymorphism deals with procedures that can operate on a variety of data types and values. Data polymorphism, a topic for part 2 of this article, deals with program variables that can store and operate on a variety of data types and values.
Procedure polymorphism occurs when a procedure, such as a function or a subroutine, can take a variety of data types as arguments. This is accomplished in F2003 when a procedure has one or more dummy arguments declared with the CLASS keyword. For example,
subroutine setColor(sh, color)
class(shape) :: sh
integer :: color
sh%color = color
end subroutine setColor
The setColor
subroutine takes two arguments, sh
and color
. The sh
dummy argument is polymorphic, based on the usage of class(shape)
. The subroutine can operate on objects that satisfy the "is a" shape relationship. So, setColor
can be called with a shape
, rectangle
, square
, or any future type extension of shape. However, by default, only those components found in the declared type of an object are accessible. For example, shape
is the declared type of sh
. Therefore, you can only access the shape
components, by default, for sh
in setColor
(i.e., sh%color
, sh%filled
, sh%x
, sh%y
). If the programmer needs to access the components of the dynamic type of an object (e.g., sh%length
when sh
is a rectangle
), then they can use the F2003 SELECT TYPE construct. The following example illustrates how a SELECT TYPE construct can access the components of a dynamic type of an object:
subroutine initialize(sh, color, filled, x, y, length, width)
! initialize shape objects
class(shape) :: sh
integer :: color
logical :: filled
integer :: x
integer :: y
integer, optional :: length
integer, optional :: width
sh%color = color
sh%filled = filled
sh%x = x
sh%y = y
select type (sh)
type is (shape)
! no further initialization required
class is (rectangle)
! rectangle or square specific initializations
if (present(length)) then
sh%length = length
else
sh%length = 0
endif
if (present(width)) then
sh%width = width
else
sh%width = 0
endif
class default
! give error for unexpected/unsupported type
stop 'initialize: unexpected type for sh object!'
end select
end subroutine initialize
The above example illustrates an initialization procedure for our shape
example. It takes one shape argument, sh
, and a set of initial values for the components of sh
. Two optional arguments, length
and width
, are specified when we want to initialize a rectangle or a square object. The SELECT TYPE construct allows us to perform a type check on an object. There are two styles of type checks that we can perform. The first type check is called type is. This type test is satisfied if the dynamic type of the object is the same as the type specified in parentheses following the type is keyword. The second type check is called class is. This type test is satisfied if the dynamic type of the object is the same or an extension of the specified type in parentheses following the class is keyword.
Returning to our example, we will initialize the length and width fields if the type of sh
is rectangle
or square
. If the dynamic type of sh
is not a shape
, rectangle
, or square
, then we will execute the class default branch. This branch may get executed if we extended the shape
type without updating the initialize
subroutine. Because we added a class default branch, we also added the type is (shape) branch, even though it does not perform any additional assignments. Otherwise, we would incorrectly print our error message when sh
is of type shape
.
Section 2, "Objects in Fortran 90/95/2003", mentioned that derived types in F2003 are considered objects because they now can encapsulate data as well as procedures. Procedures encapsulated in a derived type are called type-bound procedures. Below illustrates how we may add a type-bound procedure to shape:
type shape
integer :: color
logical :: filled
integer :: x
integer :: y
contains
procedure :: initialize
end type shape
F2003 added a contains keyword to its derived types to separate a type's data definitions from its procedures. Anything that appears after the contains keyword in a derived type must be a type-bound procedure declaration. Below is the syntax of the type-bound procedure declaration:
PROCEDURE [(interface-name)] [[,binding-attr-list ]::] binding-name[=> procedure-name]
Anything in brackets is considered optional in the type-bound procedure syntax above. At the minimum, a type-bound procedure is declared with the PROCEDURE keyword followed with a binding-name
. The binding-name
is the name of the type-bound procedure.
The first option is called interface-name
. This option is a topic of discussion in part 2 of this article.
The binding-attr-list
option is a list of binding-attributes
. The binding-attributes
that we will discuss in this article include PASS, NOPASS, NON_OVERRIDABLE, PUBLIC, and PRIVATE. There is one other binding-attribute
, called DEFERRED, that is a topic of discussion in part 2 of this article.
The procedure-name option is the name of the underlying procedure that implements the type-bound procedure. This option is required if the name of the underlying procedure differs from the binding-name. The procedure-name can be either a module procedure or an external procedure with an explicit interface.
In our example above, we have a binding-name
called initialize
. Because procedure-name
was not specified, an implicit procedure-name, called initialize
is also declared. Another way to write our example above is procedure :: initialize => initialize
.
Below is an example of a type-bound procedure that uses a module procedure:
module shape_mod
type shape
integer :: color
logical :: filled
integer :: x
integer :: y
contains
procedure :: initialize
end type shape
type, extends(shape) :: rectangle
integer :: length
integer :: width
end type rectangle
type, extends(rectangle) :: square
end type square
contains
subroutine initialize(sh, color, filled, x, y, length, width)
! initialize shape objects
class(shape) :: sh
integer :: color
logical :: filled
integer :: x
integer :: y
integer, optional :: length
integer, optional :: width
sh%color = color
sh%filled = filled
sh%x = x
sh%y = y
select type (sh)
type is (shape)
! no further initialization required
class is (rectangle)
! rectangle or square specific initializations
if (present(length)) then
sh%length = length
else
sh%length = 0
endif
if (present(width)) then
sh%width = width
else
sh%width = 0
endif
class default
! give error for unexpected/unsupported type
stop 'initialize: unexpected type for sh object!'
end select
end subroutine initialize
end module
Below is an example of a type-bound procedure that uses an external procedure with an explicit interface:
module shape_mod
type shape
integer :: color
logical :: filled
integer :: x
integer :: y
contains
procedure :: initialize
end type shape
type, extends(shape) :: rectangle
integer :: length
integer :: width
end type rectangle
type, extends(rectangle) :: square
end type square
interface
subroutine initialize(sh, color, filled, x, y, length, width)
import shape
class(shape) :: sh
integer :: color
logical :: filled
integer :: x
integer :: y
integer, optional :: length
integer, optional :: width
end subroutine
end interface
end module
Using the examples above, we can invoke the type-bound procedure in the following manner:
use shape_mod
type(shape) :: shp ! declare an instance of shape
call shp%initialize(1, .true., 10, 20) ! initialize shape
The syntax for invoking a type-bound procedure is very similar to accessing a data component in a derived type. The name of the component is preceded by the variable name separated by a percent (%) sign. In this case, the name of the component is initialize and the name of the variable is shp
. So, we type shp%initialize
to access the initialize type-bound procedure. The above example calls the initialize
subroutine and passes in 1
for color, .true.
for filled, 10
for x, and 20
for y.
You may notice that we have not yet mentioned anything about the first dummy argument, sh
, in initialize
. This dummy argument is known as the passed-object dummy argument. By default, the passed-object dummy is the first dummy argument in the type-bound procedure. It receives the object that invoked the type-bound procedure. In our example, sh
is the passed-object dummy and the invoking object is shp
. Therefore, the shp
object gets assigned to sh
when initialize is invoked.
The passed-object dummy argument must be declared CLASS and of the same type as the derived type that defined the type-bound procedure. For example, a type bound procedure declared in shape
must have a passed-object dummy argument declared class(shape)
.
We can also specify a different passed-object dummy argument using the PASS binding-attribute. For example, let's say that the sh
dummy in our initialize subroutine did not appear as the first argument. Then we would need to specify a PASS attribute like in the following code:
type shape
integer :: color
logical :: filled
integer :: x
integer :: y
contains
procedure, pass(sh) :: initialize
end type shape
Sometimes we do not want to specify a passed-object dummy argument. We can choose to not specify one using the NOPASS binding-attribute:
type shape
integer :: color
logical :: filled
integer :: x
integer :: y
contains
procedure, nopass :: initialize
end type shape
If we specify NOPASS in our example, then we still invoke the type-bound procedure the same way. The only difference is that the invoking object is not automatically assigned to a passed-object dummy in the type-bound procedure. Therefore, if we were to specify NOPASS in our initialize type-bound procedure, we would invoke initialize in the following manner:
type(shape) :: shp ! declare an instance of shape
call shp%initialize(shp, 1, .true., 10, 20) ! initialize shape
Note that we explicitly specify shp for the first argument of initialize because it was declared NOPASS.
Recall from section 2, "Objects in Fortran 90/95/2003", that a child type inherits or reuses components from their parent or ancestor types. This applies to both data and procedures when dealing with F2003 derived types. In the code below, rectangle and square will both inherit the initialize type-bound procedure from shape.
type shape
integer :: color
logical :: filled
integer :: x
integer :: y
contains
procedure :: initialize
end type shape
type, extends(shape) :: rectangle
integer :: length
integer :: width
end type rectangle
type, extends(rectangle) :: square
end type square
Using the example above, we can invoke initialize with a shape, rectangle, or square object:
type(shape) :: shp ! declare an instance of shape
type(rectangle) :: rect ! declare an instance of rectangle
type(square) :: sq ! declare an instance of square
call shp%initialize(1, .true., 10, 20) ! initialize shape
call rect%initialize(2, .false., 100, 200, 50, 25) ! initialize rectangle
call sq%initialize(3, .false., 400, 500, 30, 20) ! initialize rectangle
Most OOP languages allow a child object to override a procedure inherited from its parent object. This is known as procedure overriding. In F2003, we can specify a type-bound procedure in a child type that has the same binding-name as a type-bound procedure in the parent type. When the child overrides a particular type-bound procedure, the version defined in its derived type will get invoked instead of the version defined in the parent. Below is an example where rectangl
e defines an initialize
type-bound procedure that overrides shape
's initialize type-bound procedure:
module shape_mod
type shape
integer :: color
logical :: filled
integer :: x
integer :: y
contains
procedure :: initialize => initShape
end type shape
type, extends(shape) :: rectangle
integer :: length
integer :: width
contains
procedure :: initialize => initRectangle
end type rectangle
type, extends(rectangle) :: square
end type square
contains
subroutine initShape(this, color, filled, x, y, length, width)
! initialize shape objects
class(shape) :: this
integer :: color
logical :: filled
integer :: x
integer :: y
integer, optional :: length ! ingnored for shape
integer, optional :: width ! ignored for shape
this%color = color
this%filled = filled
this%x = x
this%y = y
end subroutine initShape
subroutine initRectangle(this, color, filled, x, y, length, width)
! initialize rectangle objects
class(rectangle) :: this
integer :: color
logical :: filled
integer :: x
integer :: y
integer, optional :: length
integer, optional :: width
this%color = color
this%filled = filled
this%x = x
this%y = y
if (present(length)) then
this%length = length
else
this%length = 0
endif
if (present(width)) then
this%width = width
else
this%width = 0
endif
end subroutine initRectangle
end module
In the sample code above, we defined a type-bound procedure called initialize
for both shape
and rectangle
. The only difference is that shape
's version of initialize will invoke a procedure called initShape
and rectangle
's version will invoke a procedure called initRectangle
. Note that the passed-object dummy in initShape
is declared class(shape)
and the passed-object dummy in initRectangle
is declared class(rectangle)
. Recall that a type-bound procedure's passed-object dummy must match the type of the derived type that defined it. Other than differing passed-object dummy arguments, the interface for the child's overriding type-bound procedure is identical with the interface for the parent's type-bound procedure. That is because both type-bound procedures are invoked in the same manner:
type(shape) :: shp ! declare an instance of shape
type(rectangle) :: rect ! declare an instance of rectangle
type(square) :: sq ! declare an instance of square
call shp%initialize(1, .true., 10, 20) ! calls initShape
call rect%initialize(2, .false., 100, 200, 11, 22) ! calls initRectangle
call sq%initialize(3, .false., 400, 500) ! calls initRectangle
Note that
sq
is declaredsquare
but its initialize type-bound procedure invokesinitRectangle
becausesq
inherits therectangle
version ofinitialize
.
Although a type may override a type-bound procedure, it is still possible to invoke the version defined by a parent type. Recall in section 2, "Objects in Fortran 90/95/2003", that each type extension contains an implicit parent object of the same name and type as the parent. We can use this implicit parent object to access components specific to a parent, say, a parent's version of a type-bound procedure:
call rect%shape%initialize(2, .false., 100, 200) ! calls initShape
call sq%rectangle%shape%initialize(3, .false., 400, 500) ! calls initShape
Sometimes we may not want a child to override a parent's type-bound procedure. We can use the NON_OVERRIDABLE binding-attribute to prevent any type extensions from overriding a particular type-bound procedure:
type shape
integer :: color
logical :: filled
integer :: x
integer :: y
contains
procedure, non_overridable :: initialize
end type shape
Up to this point, subroutines have been used to implement type-bound procedures. We can also implement type-bound procedures with functions as well. Below is an example with a function that queries the status of the filled component in shape
.
module shape_mod
type shape
integer :: color
logical :: filled
integer :: x
integer :: y
contains
procedure :: isFilled
end type shape
contains
logical function isFilled(this)
class(shape) :: this
isFilled = this%filled
end function isFilled
end module
We can invoke the above function in the following manner:
use shape_mod
type(shape) :: shp ! declare an instance of shape
logical filled
call shp%initialize(1, .true., 10, 20)
filled = shp%isFilled()
In section 7, "Procedure Overriding", we showed how a child type can override a parent's type-bound procedure. This allows a user of our type to invoke, say, the initialize
type-bound procedure, without any knowledge of the implementation details of initialize. This is an example of information hiding, another important feature of OOP.
Information hiding allows the programmer to view an object and its procedures as a "black box". That is, the programmer can use (or reuse) an object without any knowledge of the implementation details of the object.
Inquiry functions, like the isFilled
function in section 8, "Functions as Type-Bound Procedures", are common with information hiding. The motivation for inquiry functions, rather than direct access to the underlying data, is that the object's implementer can change the underlying data without affecting the programs that use the object. Otherwise, each program that uses the object would need to be updated whenever the underlying data of the object changes.
To enable information hiding, F2003 provides a PRIVATE keyword (and binding-attribute). F2003 also provides a PUBLIC keyword (and binding-attribute) to disable information hiding. By default, all derived type components are declared PUBLIC. The PRIVATE keyword can be placed on derived type data and type-bound procedure components (and on module data and procedures). We illustrate PUBLIC and PRIVATE in the sample code below:
module shape_mod
private ! hide the type-bound procedure implementation procedures
public :: shape, constructor ! allow access to shape & constructor procedure
type shape
private ! hide the underlying details
integer :: color
logical :: filled
integer :: x
integer :: y
contains
private ! hide the type bound procedures by default
procedure :: initShape ! private type-bound procedure
procedure, public :: isFilled ! allow access to isFilled type-bound procedure
procedure, public :: print ! allow access to print type-bound procedure
end type shape
contains
logical function isFilled(this)
class(shape) :: this
isFilled = this%filled
end function isFilled
function constructor(color, filled, x, y)
type(shape) :: constructor
integer :: color
logical :: filled
integer :: x
integer :: y
call constructor%initShape(color, filled, x, y)
end function constructor
subroutine initShape(this, color, filled, x, y)
! initialize shape objects
class(shape) :: this
integer :: color
logical :: filled
integer :: x
integer :: y
this%color = color
this%filled = filled
this%x = x
this%y = y
end subroutine initShape
subroutine print(this)
class(shape) :: this
print *, this%color, this%filled, this%x, this%y
end subroutine print
end module
The example above uses information hiding in the host module as well as in the shape type. The private statement, located at the top of the module, enables information hiding on all module data and procedures. The isFilled
module procedure (not to be confused with the isFilled
type-bound procedure) is hidden as a result of the private statement at the top of the module. We added public :: constructor
to allow the user to invoke the constructor module procedure. We added a private statement on the data components of shape
. Now, the only way a user can query the filled component is through the isFilled
type-bound procedure, which is declared public.
Note the private statement after the contains in type shape
. The private that appears after type shape
only affects the data components of shape
. If you want your type-bound procedures to also be private, then a private statement must also be added after the contains keyword. Otherwise, type-bound procedures are public by default.
In our example, the initShape
type-bound procedure is declared private. Therefore, only procedures local to the host module can invoke a private type-bound procedure. In our example above, the constructor
module procedure invokes the initShape
type-bound procedure. Below is how we may invoke our example from above:
program shape_prg
use shape_mod
type(shape) :: sh
logical filled
sh = constructor(5, .true., 100, 200)
call sh%print()
end program shape_prg
Below is a sample compile and sample run of the above program (we assume that the shape_mod module is saved in a file called shape.f03 and that the main program is called main.f03):
% pgfortran -V ; pgfortran shape.f03 main.f03 -o shapeTest
pgfortran 11.2-1 64-bit target on x86-64 Linux -tp penryn
Copyright 1989-2000, The Portland Group, Inc. All Rights Reserved.
Copyright 2000-2011, STMicroelectronics, Inc. All Rights Reserved.
shape.f03:
main.f03:
% shapeTest
5 T 100 200
In our previous example, we created an instance of shape
by invoking a function called constructor
. This allows us to hide the details for constructing a shape
object, including the underlying type-bound procedure that performs the initialization. However, you may have noticed that the word constructor
could very well be defined somewhere else in the host program. If that is the case, the program cannot use our module without renaming one of the constructor functions. But since OOP encourages information hiding and code reusability, it would make more sense to come up with a name that probably is not being defined in the host program. That name is the type name of the object we are constructing.
F2003 allows the programmer to overload a name of a derived type with a generic interface. The generic interface acts as a wrapper for our constructor function. The idea is that the user would then construct a shape in the following manner:
program shape_prg
use shape_mod
type(shape) :: sh
logical filled
sh = shape(5, .true., 100, 200) ! invoke constructor through shape generic interface
call sh%print()
end program shape_prg
Below is the modified version of our example from section 9, "Information Hiding", that uses type overloading:
module shape_mod
private ! hide the type-bound procedure implementation procedures
public :: shape ! allow access to shape
type shape
private ! hide the underlying details
integer :: color
logical :: filled
integer :: x
integer :: y
contains
private ! hide the type bound procedures by default
procedure :: initShape ! private type-bound procedure
procedure, public :: isFilled ! allow access to isFilled type-bound procedure
end type shape
interface shape
procedure constructor ! add constructor to shape generic interface
end interface
contains
:
:
end module shape_mod
Our constructor function is now declared private and it is invoked through the shape
public generic interface.
Code reusability, an important feature of Object-Oriented Programming (OOP), is enabled through inheritance, polymorphism, and information hiding. With inheritance, an object can be extended and code from the parent object can be reused or overloaded in the child object. Code reusability is also enabled through polymorphism. There are two types of polymorphism: procedure polymorphism and data polymorphism. Procedure polymorphism enables code reusability because a procedure can operate on a variety of data types and values. The programmer does not have to reinvent the wheel for every data type a program may encounter with a polymorphic procedure. Part 2 of this article will cover data polymorphism. Finally, we examined information hiding which allows programmers to use an object without having to understand its underlying implementation details. Fortran 2003 (F2003) supports inheritance, polymorphism, and information hiding through type extension, the CLASS keyword, and the PUBLIC/PRIVATE keywords/binding-attributes respectively.
In the next installment of this article we will continue our discussion of OOP with F2003 by examining the other form of polymorphism, data polymorphism, and see how it can be used to create flexible and reusable software components. We will also examine the following OOP features offered by F2003: unlimited polymorphic objects, typed allocation, sourced allocation, generic type-bound procedures, deferred bindings, and abstract types.
really useful