Learning Smalltalk by Examples
Smalltalk - coming of age and offering an alternative to C and C++
By R. L. Peskin and S. S. Walther, Landgrove Associates, Flemington, New Jersey
Note: Source code files accompanying article are located on MacTech CD-ROM or source code disks.
This is the first in a series of articles aimed at those who want to learn something about programming in Smalltalk. Our overall objective is to convey the ease and flexibility of Smalltalk programming to those who may be used to programming in more traditional languages and environments. The approach will be practical; learning by example. We will be using the SmalltalkAgents system as the vehicle to illustrate our examples.
The Wire Project
The Wire example is a simple tool to determine the shortest length of wire running between an arbitrarily selected set of points. (If you arent into wire routing, the same sort of analysis applies to routing a salesperson between cities.) Kent Beck and Ward Cunningham first introduced us to this example at the Tektronix, Inc. Smalltalk classes in 1986. The Wire Project is a good illustration of tool building in Smalltalk; that is, integrating the behavior of a Model and the user interface to interact with that Model. The Model refers to your data, perhaps including the computations that generate them. The View is the interface to that data. Commonly this means a visual interface (list, graph, visualization), but it may well include sound, touch, smell, etc. in future platforms. The Controller refers to the set of messages and/or events that facilitate communication and control between the Model and View(s). Our Model is a list of Wire coordinates (list of points), together with the methods (behaviors) that do calculations relevant to those Wire coordinates. Our View is a dynamically changing graphical representation of the Wire points and the lines connecting them. Our Controller is not a formal object, but rather a group of behaviors that effect control; some of these are behaviors of the View and some of the Model.
The Wire tool accepts an arbitrary point from the users click in a window. The length of the Wire is computed and displayed as points are added. When the user clicks a Shorten button, the Wire rearranges its points to minimize the total length and the new arrangement is shown in the window. Here is a diagram of the Wire tool. Well focus on the implementation of the Model (WireList).
Wire Tool Environment
Smalltalk consist of an image which holds objects and their behaviors and other data in the form of interpretable code, and a virtual machine which acts as an interpreter of code. (This is an much oversimplified description, but sufficient for present purposes.) Unlike interpreters of old, these virtual machines can run full-out.
The Smalltalk environment is a basic set of tools. It makes Smalltalk a much more productive language to work in than traditional static languages. These tools allow immediate feedback for testing and debugging, as well as incremental development and alteration of code. Changes to individual algorithms (methods) as well as whole classes are easily accomplished interactively. This is in marked contrast to static object-oriented languages such as C++. The Smalltalk environments tools include classes that are available for the your immediate use. In addition, browsers give you access to the source code for these classes. Debugging tools and text editing fields in browsers and editors let you do code entry, compilation, and execution almost anywhere. Getting from the development process to a standalone application can be done directly within Smalltalks environment.
The main idea in Smalltalk programming is to program from examples. The primary tool is the Browser, an interactive tool which gives you access to the Smalltalk system code. It also lets you modify code and add your own classes and methods. Heres a Browser view.
A Smalltalk Browser
In the upper left hand corner is the class being browsed (Rectangle). The main text edit section shows the method being browsed (intersects:). Comments are in quotes. A typical Smalltalk programming statement is: new := Rectangle basicNew. The object (in this case the class Rectangle) is sent a message (basicNew), and the resulting object assigned to the variable, new. Smalltalk syntax refers to the receiver (Rectangle in the above example) and the message selector (basicNew). Also note the line: «SectRect(self:Ptr; rect:Ptr; new:Ptr):Boolean», this is how a Mac ToolBox call (or other external function) is made in SmalltalkAgents (STA). Smalltalk programming is done by editing new or existing methods in the Browsers edit view and saving the edit; saving automatically invokes the compiler. The result is a new or changed method in your image with associated byte code (or other binary) representation.
The Category View shown at the bottom has four list views; Categories, Classes, Protocols, and Methods. Categories and Protocols are for reference and classification purposes only, they have no programmatic aspect; Categories group classes together and Protocols group methods. Classes list the classes in the Smalltalk class library (including any you have added) and Methods list the behaviors (methods) for the selected class.
How do we decide what are the objects? How do we decide which classes to subclass? These are common agonies in object-oriented programming. Frankly it may not make that much difference as long as the decisions are reasonable. (For example, a class for our Wires should probably not be a subclass of Rectangle; they have no common behavior.) Some useful rules are 1) If in doubt make your class a subclass of Object (the top level generic class); you can always relocate it later very easily. 2) Choose to subclass under a class (other than Object itself) only if there is a reason to do so. For example, a good reason is so that you can make use of existing methods (behaviors). 3) Dont create a class that doesnt have significant computational responsibilities; classes should have behaviors.
A class has a data structure that contains information (data) about its state. These data are objects called instance variables. A class also contains references to behaviors (methods) that let you change or simply read its instance variables. Typically an object is a concrete instantiation, that is, member of a class. An object is a specific instance of a class, and has its own set of instance variables. Sometimes classes have behavior and data sufficient to make them do concrete things on their own without instances. Normally a class functions as a template; an object belonging to that class has real values for the class data.
The Wire Model
Our wire is a collection of (geometric) points. One choice is to make the class associated with the Wire a subclass of one of the collection classes. Well choose to make it a subclass of List. An equally valid (and perhaps better) choice is to make Wire a subclass of Object, and then provide it with an instance variable which itself is a List. In the first case, the Wire will inherit all the behavior of a List; in the second case, the Wires list instance variable would be accessed when List behavior is needed. For simplicity, well chose the former.
We create a subclass of List called WireList. This operation is usually done by filling in a template in the Browser. Below is the class browser view for WireList. The programmer supplies the name of the subclass as a symbol (#WireList); the superclass (#List), and a list of instance variable names #(first second). (Well ignore other information such as the paths (e.g. Environment@#List) for now.
The category view shows three Protocols, accessing (adding point, relocating points, returning specific instance variables), calculating (computing wire length, shortening the wire), and drawing (supporting calculations for drawing the wire). (Note: You might decide that this last protocol is not appropriate for a List. You might want to create a new subclass for Wire directly under class Object, and have the list as an instance variable. The beauty of Smalltalk is that you can make these relocations easily as the program develops, a situation somewhat different than that in C++.)
Now lets fill in the behavior in the first two protocols, starting with accessing. First we want to add points to our list. Well use the following notation: WireList>I>add: to express WireList class, Instance method, add: aPoint. (An instance method is a behavior for specific objects which are instances of a class; well come back to class methods later.)
WireList > I> add:
The meaning of add: is that the message add with a parameter (aPoint) is sent to an object which is an instance WireList. The result is that a new point is added to the specific WireList object. The add: method shown here is sent to super; this starts the search for the implementation in WireLists superclass, List, which has the add: method we need. WireList>I>add: is actually not needed, we include it for illustrative purposes. If WireList didnt have a method, add:, the superclass would be tried automatically. If the method is not found there, the search continues until some class in the hierarchy either knew how to do add:, or we run out of classes to search and a failure message gets returned. It is instructive to use the Browser tools to see all the different implementations of add: found in the system. (List is an STA class; equivalents in other systems are classes like OrderedCollection.)
First and second are examples of accessors - methods to get instance variables. For example,
WireList > I> second
This translates to send the message position: 1 (go to the first position in the list) to self. Self is a reference to the object itself, in this case a specific WireList; self is equivalent to this in C++. The semicolon is shorthand for cascading messages. The message next to find the next element in the list is sent to the result of the position: 1 message. The result returned, indicated by ^, is the second element in the list. Note that neither first nor second are actually needed in the Wire tool; they are there for testing purposes. It is good coding form to use accessor methods to get and set instance variables, rather than directly access data structures. The final method in this group does a simple exchange of two list elements.
WireList > I> exchange:and:
In this method, at:put: is a method of class Object. At: is an accessor method of class Object. Temporary variables (whose scope is limited to the method) are denoted by vertical bars, e.g. |temp1 temp2 temp3|. A period is used to denote the end of a statement, and := is the assignment operator. In creating the Wire tool, we figured out that we needed exchange:and: during development of one of the algorithms under the calculating protocol. Protocol groupings were done as a final step (more on this later).
At this point we can do some computation. We can create a new WireList, put some values in it, access the second element and exchange any two elements. Smalltalks analog to the terminal is a Console or Transcript window. You can also open a window called a workspace and do line by line computing. In fact, you can do this in any appropriate code edit field of a Smalltalk window, including the code edit field of the Browser.
A Console Window
Well show you how some of this interactive stuff works. In the first line of the Console window, we created a new instance (member) of class WireList. new is a constructor method and is an example of a class method; all classes know how to respond to new to create new instances. In the second Console line, we added some points, and the next line shows the list of points returned. Then we asked Q for its second member, and 5@6 was returned. Finally we exchanged the 2nd and 3rd elements, and the resulting new list is shown. (Code in a Console can be executed line-by-line or grouped; once a selection is made, you can issue some sort of execute or doIt command.)
We are now ready to develop the calculating protocols, that is, the algorithms to shorten the Wire by seeking the minimum length connecting its points. Well do this by exchanging point positions in the list until we get a minimum length. Right away, we know well need a method to compute distance between points. (This sort of functionality is supplied as one of the methods in class Point in some Smalltalk versions.)
WireList > I> distance: indx: to: p
This method differs from the usual class Point distance method; here we compute distance between a Wires point referenced via an index and some other point, p. (For simplicity, error checking is not included.) Next well look at the method to compute the length of the Wire.
WireList > I> length
There is a lot in this method. After initializing the length to 0, the List>I>reset method is used to set the WireList pointer to the beginning of the list. Smalltalk ordered collections start their indexing at 1, so we set a temporary variable index, previous, to the first position. Next we encounter an iteration (do:) over a block. Block contexts in Smalltalk are denoted by code contained inside brackets, [ ]. Blocks are a very important part of Smalltalk programming, and have no simple counterpart in C; the context of a Block is preserved even if evaluation is delayed. In the above code, the do: operation iterates over all elements in the list. The syntax [:p | ....code...] denotes that p is a temporary object that will take on different values, the do: assigns p to the points in the list starting at the beginning. Inside the block, we use the method distance to compute the total length, which is returned at the end of the iteration.
The real action algorithmically in the Wire tool is contained in the method, shorten. This is not the appropriate place to discuss the algorithmics employed. It is sufficient that a simplified form of simulated annealing is used. We randomly pick integers from the WireList indices and interchange them and check to see if the interchange shortens the Wire.
WireList > I> shorten
This method has some new features. 100 timesRepeat: [...] does the code in the outer block 100 times. self size is the size of the Wire list. Float random is how random numbers are generated in STA; other systems may use different methods for this. Note the use of length and exchange:and: that were previously developed. self length < minLength returns a Boolean. ifTrue: ifFalse: is a Smalltalk conditional test; if the Boolean is true, one code block is performed, else another code block is performed.
Now we can test out these new methods in a Console window.
In the first line, a new WireList is created. (Smalltalk code is shown in bold; results are in plain.) Next its distance from the first element to the last point is computed; then the length is calculated. The Wire is told to shorten itself, and a new list results; in this simple case only the last two elements were interchanged. Finally the new length of the shortened wire is obtained. At this point in the development process, we have a fairly complete model. The next task is to create the tool per se so that the user can perform these operations on the model via mouse clicks and graphical viewing.
The Wire View and Control
Well have to leave a complete description of the Wire tool window (WireListWindow) and the control class (WireTool) to the next article, but heres a quick overview. The steps are generally as follows:
First, we have to make a Window for the tool; in STA we create a subclass of a generic user interface window (UIWindow) which we call WireList Window. Usually this sort of thing is done with a GUI builder facility built into the Smalltalk system. Following is part of the window creation method, namely some code associated with the addition of a <shorten> button. The line on: #buttonRelease send: #shorten to: self associates the event action (release of the Mouse button) with a method (shorten) that is sent to the WireListWindow when the Mouse is released.
Similar code sets up handling of Mouse action to add points to the Wire. A portion of the WireListWindow is shown below after the user has added some points:
The window looks as follows after pressing the shorten button two times (two iterations of the shorten algorithm):
Notice that the wire is shorter and the length has been reduced. How does the user action of clicking at points in the window result in a) drawing the wire, and b) adding said points to the Wire list? And how do we connect the pressing of the shorten button in the window to invocation of the WireList > I> shorten method?
The answer to these questions rests on the fact that, in Smalltalk, views like the above WireListWindow maintain an instance variable (called module in STA) object which is assigned the model (in this case WireList) object instance. The point to be added is supplied by something like p := Mouse localPosition, which returns local window coordinates from the Mouse click. Code like module add: p adds a point to the list of the selected points. Pressing the shorten button works the same way - the windows shorten method simply asks the WireList to shorten; this invokes the wire shorten method we described previously.
How the actual drawing of the Wire is accomplished is an interesting point. The details are reasonably complex, and can be Smalltalk system dependent, but the programming is straightforward. When Smalltalk receives an update event, that update request is passed through to the WireListWindow, which handles it by invoking a rendering operation that includes the necessary call to draw the latest Wire based on the current contents of the WireList.
In this article we have tried to present an introduction to Smalltalk programming from a practical point of view. All of the code presented for the WireList can be used (with minor changes) in any Smalltalk version, including public domain ones. If a version of Smalltalk is available, we suggest typing in the WireList code and experimenting. The only way to get a feel for the simplicity and productivity of Smalltalk is to experiment with it. In the next article we will cover the programming support needed for the view and control process, as well as consider some additional tool features such as interactive point relocation and inspection tools.
For SmalltalkAgents users, the WireProject code can be found at ftp://qks.com/pub/sta/tutorials/WireList/. A text version of the WireProject suitable for implementation in other Smalltalk versions is available on request (email only) from R. Peskin. Internet: email@example.com, Applelink: D6615, CompuServe: 70372,616. You can also get the project from the usual MacTech Magazine sites or source disk (see p. 2 for details).