Object Oriented Programming in Matlab: More advanced topics

Contents

Precedence relations

When multiple objects are passed to a method, Matlab determines which class's method to call based on the superior-inferior relation. The most superior class's method is invoked and if all of the classes have equal superiority, the left most object takes precedence.

We specify these relationships when creating classes, right in the classdef statements as follows:

classdef(InferiorClasses = {?class1, ?class2})

The ? marks are used to construct metaclass instances but this detail is not particularly important; class1 and class2 are instances of the inferior classes. We discuss meta classes in a later section.

Arrays of objects can be passed to methods as well but the class of an array of objects is the same as the class of the objects stored inside, (which must all be the same), and this class is used to determine precedence. See the section on Object Arrays for more information.

Set and Get

Matlab supports special kinds of setter and getter functions for assigning and accessing properties that are executed whenever an attempt to set or get the corresponding property is made. Use of these is optional; they are only called if they exist. By taking this approach, we can make properties public so that clients can use the convenient dot notation, while still maintaining a level of indirection by effectively intercepting the call (although this is much slower than the regular dot notation).

We will add get and set methods for the public day property of the date class as an example. We write get followed by a dot and the property name, similarly for set .


 function day = get.day(obj)
     day = obj.day;              % We could execute other code as well.
 end
 function obj = set.day(obj,newday)
     obj.day = newday;
 end


We then assign and query the value as we did before using the dot notation, but the call is intercepted by these functions. We must take care as before to return the object in the setter methods, (as the objects are by default passed by value).

d1 = mydate(1,4,22,3,2008);   % create another mydate
day = d1.day;
d1.day = 5;

Unfortunately, as of Matlab version 2008a, it was not possible to override get and set methods in subclasses, severely limiting the use value of this approach in complex projects. Furthermore, these methods are called even when properties are accessed or set from within the class, making their use for input checking a frustrating exercise. Keep this in mind when initializing variables in the constructor; these functions, if present, are invoked during construction and so must be able to deal with cases in which the variables are not yet set.

Operator Overloading

Every use of a Matlab operator, such as

+ - .* * ./ .\ / \ .^ ^ < > <= >=
== ~ ~= & | && || : ' .' [] [;] () .

is actually short hand or syntactic sugar for a call to a named function like plus(), minus(), times(), power(), lt(), eq(), not(), etc. Section 7.32 of the Matlab OO documentation lists all of these operators with their corresponding function names.

We can define custom behavior for any of these operators by witting class methods by the same name. Since class methods are dynamically dispatched, our own versions of these functions will execute when we use the corresponding operators with our objects. We could write our own plus() method in the date class, for example, to add dates together and then call the function with d1 + d2. Or, we could write our own lt() function, (for less than) to compare dates, calling it with d1 < d2. Such calls get converted automatically to plus(d1,d2) and lt(d1,d2), and our own implementations of these functions are then invoked.

Operators retain their natural precedence so that * takes precedence over + in order of operations, even if one or both have been overloaded.

Another useful method to overload is display() - the function that automatically displays objects when we do not suppress the output with a semicolon. Writing our own display function allows us to display objects in any way we like.

Sometimes it is useful to simply reuse concise informative names that belong to built in functions like plot(). While plot is never automatically invoked, nor does it correspond to an operator, it is used so frequently in Matlab that reusing this name with our own objects can serve to self document their behavior extremely well, (assuming our plot function does something reasonable).

If you overload an operator or function but want to use the original implementation for some reason, use the builtin() function, which takes the string name of the function as the first input, followed by that function's inputs.

subsref and subsasgn

There are two very important functions that are frequently overloaded and deserve specific mention: subsref, and subsasgn. When the dot operator, parentheses, or curly braces, . () {}, are used in indexing operations, subsasgn is called, and when they are used in assignment operations, subsasgn is called. By overloading these functions, we can create customized behavior for our classes.

Suppose we write our own data structure class, for instance, and want calls like obj(3) to retrieve the third element in our structure, we could achieve this by overloading subsref.

Here are the definitions of these functions:

  function obj = subsasgn(obj, S, value)
  function value = subsref(obj, S)
 Type is one of '()' '{}' or '.' depending on the call.
 Subs is a cell array or a string of the actual subscripts used.

In complicated calls involving multiple operators like obj(5,9).prop(1:19)=value, a single call to subsasgn is made and all of the information in the call is passed to the function. In this case, S is an array of structs with the following values.

S(1).type='()'	S(2).type='.'	    S(3).type='()'
S(1).subs={5,9}	S(2).subs='prop'	S(3).subs={1:19}

Note that as of 2008a, overloaded operators do not work within the class methods, only from outside of the class, although this could, and hopefully will change in future versions. This includes any function included in the @ directory (if any) or any subdirectories. You can still call the functions by name, (i.e. plus() instead of +).

Handle Superclass

We previously mentioned that objects in Matlab are, (by default) passed by value, meaning that full copies are passed back and forth in method calls. Matlab graphics objects, however, are passed by reference, (via handles). If we subclass the built in handle class as in

classdef myclass < handle

then objects of our class will be passed by reference too, not value. Doing so has a number of benefits and consequences, which we will now discuss.

When we construct a handle object as in h = myclass(), h stores a pointer or handle to the object not the object itself. If we then execute h2 = h, we simply create another pointer to the same underlying object. For example, we could call h.prop = 3 , and then p = h2.prop and p would equal 3.

In handle method calls, there is no need to return the object because assignments occur in place, (although returning a handle to the object does no harm).

If our objects will be very large, it can be much more space efficient to use handle objects because we no longer need to copy the entire object in every method call. (Note, however, that Matlab does a lot of optimization under the surface and only actually copies objects or variables when it absolutely has to).

Only handle classes support events; we will discuss events shortly.

The major advantage, however, is that it is much easier to write data structures, (particularly recursive structures) such link lists or binary trees. We give a very simple implementation of a binary tree class now and illustrate how we can easily recurse over all of the nodes by simply following handles.


classdef bnode < handle               % subclass handle
    properties
       left;        % left  child
       right;       % right child
       data;        % data stored at the node
    end
    methods
        function obj = bnode(data)
            obj;
            if(nargin > 0)
                obj.data = data;
            end
        end
    end
end
function labelNodes(node,depth)
% recursively label the depth of the nodes
    if(isempty(node)),return,end
    node.data = depth;
    labelNodes(node.left,depth+1);
    labelNodes(node.right,depth+1);
end


It is much more complicated to create an identical copy of a handle object as we cannot simply go h2 = h1. We can use the following code, however, to create a shallow copy of any object we like. It needs full access to all of the properties and so should be added as a class method. Another approach is to use the struct() function to convert an object to a struct and then write the constructor to optionally take a struct, building a new object from its fields.


 function copy = copyobj(obj)
 % Create a shallow copy of the calling object.
     copy = eval(class(obj));
     meta = eval(['?',class(obj)]);
     for p = 1: size(meta.Properties,1)
         pname = meta.Properties{p}.Name;
         try
             eval(['copy.',pname,' = obj.',pname,';']);
         catch
             fprintf(['\nCould not copy ',pname,'.\n']);
         end
     end
 end


When there are no more handles to an object left on the stack, the object is declared invalid and the Matlab garbage collector will free the memory when it gets a chance. We can test if a handle to an object is valid with the isvalid(h) method and delete the object, causing all of its handles to become invalid with delete(h).

dynamicprops & hgsetget

Handle has two subclasses, which you can subclass instead yielding addtional functionality. By subclassing dynamicprops you get all of the benefits of subclassing handle plus the ability to dynamically attach temporary data to objects without modifying the class definition. You simply call the inherited addprop() function, as in P = obj.addprop('newProperty') and you can then make calls like obj.newProperty = 3 or val = obj.newProperty. The return value of addprop() ,P, can be used to set attributes of the property, (i.e. make it hidden, etc.), or to delete the property via delete(P).

The hgsetget class, (also a subclass of handle), lets you use Matlab graphics style set and get methods as in set(h,'property',value). See section 4.19 in the Matlab OO documentation for more details).

Events

Matlab now has quite good support for event based programming in which objects trigger events in response to a change in state, notifying one or more other objects that have registered as listeners. This can be particularly useful when the appropriate flow of control depends upon things external to the program such as a user's interaction with a graphical interface or environmental sensors. It can be a useful paradigm in its own right, however, particularly for simulations. Chapter 8 of the Mathworks OO documentation covers events.

To begin, all classes involved must inherit from the handle class, (or one of its subclasses). The triggering class must declare an events block in its class definition. Event blocks have attributes just like method and property blocks, defining event access control.

The ListenAccess attribute determines where you can create event listeners and NotifyAccess determines where events can be triggered. In the below example, we set ListenAccess to public so that we can register an object as a listener anywhere we like, and NotifyAccess to protected so that only methods of the date class, (or any subclasses of date) can trigger the events.

Within the block, we define the events by simply specifying a name. Here we continue with the date example and will trigger events when the date is equal to either Jan 1, 2000 or the Vancouver Olympics start date of February 12, 2010.


events(ListenAccess = 'public', NotifyAccess = 'protected')
     y2k;                            % define a couple of events
     olympicsStart;
end


Now that we have defined two events, we have to decide when to trigger them. Lets add a line to the set.day method we discussed earlier, (which is called whenever the day property is set). We will have it call a new method we will write called checkDate() , which will fire the events if the current date matches one we are looking for. We use the notify() method, (inherited from handle) to fire the event and simply pass it the name of the event we want to trigger.


   function checkDate(obj)
       if(isempty(obj.year) || isempty(obj.month) || isempty(obj.day))
           return;        % this function may be called before all fields initialized.
       end
       if(obj.year == 2000 && obj.month == 1 && obj.day == 1)
           obj.notify('y2k');
       end
       if(obj.year == 2010 && obj.month == 2 && obj.day == 12)
           obj.notify('olympicsStart');
       end
   end


Notify will send an event notification to every object that is 'listening'. By default the event object will have the name, (as specified in the events block, e.g. 'y2k' or 'olympicsStart') and a handle to the source object itself - the object that triggered the event.

You can create customized event objects with whatever information you like by subclassing event.EventData and passing an instance along with the event name to notify(). For more information see section 8.9 of the Mathworks OO documentation.

Now that we have objects of our date class sending events, we need to add listeners - objects that will be informed when events are triggered. Note that objects of our date class will not necessarily know who is listening, unless we go out of our way to tell them, (which we will not).

There are two ways to register an object as a listener but we will only discuss one here: using the addlistener() method inherited from handle.

(The other approach involves creating an object of type event.listener - see the Matlab OO documentation for more details).

Every class that inherits from handle has an addlistener() method that takes three arguments: a handle to an object that will generate events, the name of the event to listen for, (e.g. 'y2k'), and a handle to a function that should execute when the event is 'heard'. This function, called a callback function, must take two arguments: src - the object that generated the event, and evnt - the event object. Here is a possible class definition for listening objects. Note, we can call addlistener() at any point, not just in the constructor as we do here.


 classdef snoopingClass < handle
   properties
      snoopOn;
   end
    methods
        function obj snoopingClass(dateObj)
        % class constructor
           obj.snoopOn = dateObj;
           y2kListener = addlistener(dateObj,'y2k',@(src,evnt)fixY2Kbugs(obj,src,evnt));
           olympicsListener = addlistener(dateObj,'olympicsStart',@(src,evnt)gossip(obj,src,evnt));
        end
        function fixY2Kbugs(obj,src,evnt)
        % This will be executed when a y2k event is fired by the date object.
           display(evnt.EventName);
        end
        function gossip(obj,src,evnt)
        % This will be executed when a olympicsStart event is fired by the date object
           display(evnt.EventName);
        end
    end


The execution of a callback can be temporarily deactivated by setting the Enabled property of the listener object to false.

y2kListener.Enabled = false

Four types of events are automatically fired in response to the access or assignment of observable properties: PreSet, PostSet, PreGet, & PostGet. The 'pre' events are fired just before a value is changed or serviced, and the 'post' events are fired just after. Observable properties are those defined in a properties block with the setObservable or getObservable attributes set to true as in

properties(SetObservable = true) .

These events are not listed in the event block.

To add a listener for the PostSet event of the day property, for example, use the following syntax.

lh = addlistener(obj,'propertyName','PostSet',@(src,evnt)callbackFunction(obj,src,evnt));

See section 8.14 of the Matlab OO documentation for more details on listening for property changes. Finally, note that subclasses inherit the events of their superclasses.

Meta Classes

Matlab has quite a novel feature, meta classes, which allow you to dynamically inspect the properties of a particular class. Each class, defined using the classdef syntax, has a corresponding metaclass which you can invoke using the ? operator. The resulting object stores information about the class methods, properties, events, superclasses, etc, as well as their attributes. Metaclasses can be used to write highly generic code. For example, viewClassTree makes extensive use of metaclasses.

The following example function, finds all of the superclasses of a particular class, including the superclasses of its superclasses. The metaclass() function operates just like the ? operator but can be used with string names, whereas ? requires an instance of the object.


   function list = ancestors(class)
   % input is the string name of the base class
   % output is a cell array of ancestor class names
       list = {};
       meta = metaclass(class);
       parents = meta.SuperClasses;
       for p=1:numel(parents)
           list = [parents{p}.Name,ancestors(parents{p}.Name)];
       end
   end


Here is a look at the kind of data available.

metadata = ?mydate
metadata = 
  meta.class handle
  Package: meta

  Properties:
                   Name: 'mydate'
            Description: ''
    DetailedDescription: ''
                 Hidden: 0
                 Sealed: 0
        ConstructOnLoad: 0
        InferiorClasses: {0x1 cell}
             Properties: {11x1 cell}
                Methods: {22x1 cell}
                 Events: {3x1 cell}
       EnumeratedValues: {0x1 cell}
           SuperClasses: {[1x1 meta.class]}
      ContainingPackage: {}