Perl and the Tk Extension: Events And Other Things

Steve Lidie

event (i-vent'): something that happens: a noteworthy occurrence or happening: something worthy of remark: an unusual or significant development. (Paraphrased from Webster's Third New International Unabridged Dictionary.)

Events are what drive Perl/Tk programs. In the past I've described these events superficially, sweeping lots of detail under the MainLoop() rug, all for the sake of simplicity. MainLoop() is our friend, since it's all that is needed for nearly every Perl/Tk program. But there are few times when it's not up to the task at hand.

Today's featured program is a simple Pong-like game sporting a new widget derived from the Canvas class, which we'll compare to last issue's Odometer, a composite widget. Our Pong game doesn't use MainLoop(), but handles events itself with DoOneEvent().

Before discussing pong we'll examine some other programs, including a simple animation called neko, which demonstrates the Photo widget and some other Tk commands. Tk defines four broad event categories: X, timer, input/output, and idle. X events are generated in response to mouse motion, button and keyboard actions and window changes. You already know that many of these events have builtin Tk bindings, and that you can create your own bindings, so all you need to do is define the event handler, or callback. (There are lots of other X events which we'll examine in detail in future issues.) Timer events are used for periodic happenings, from blinking items to animating images. Input/output events help prevent your application from freezing when reading and writing to terminals, pipes, or sockets. Finally, idle events are low priority callbacks invoked only when all events from the previous three event queues have been processed. Tk uses the idle events queue to redraw widgets, since it's generally a bad idea to redisplay a widget after every change of state. By deferring redraws until there is nothing left to do, widgets presumably reach their "steady state." The result is improved performance and a flicker-free screen.

Timer Events

Last time we saw a useful idiom for scheduling asynchronous tasks, shown at the top of the next column:


sub modo { # Do stuff, then reschedule myself.
	$MW->after($MILLISECOND_DELAY, \&modo); 

Before modo() returns, it uses after() to schedule a timer event and define the handler (callback). This idiom is so common that Perl/Tk provides repeat() as a shortcut, so the above code can be condensed like so:

$MW->repeat($MILLISECOND_DELAY, \&modo);

A working example named rpt is available on the TPJ web site.

Tk uses timer events to flash the insertion cursor for entry widgets. After the widget gets the keyboard focus, it displays the cursor and queues a timer callback. Then the callback erases the cursor and the cycle repeats, several times per second. This technique is often used to flash alert messages or special buttons. You can use repeat(), but this is the idiom you'll almost always see:

my $b = $MW->Button(-text => 'Hello World!', 
                 -command => \&exit)->pack; 
flash_widget($b, -background, qw(blue yellow), 500);

sub flash_widget {  
    # Flash a widget attribute periodically.  
    my ($w, $opt, $val1, $val2, $interval) = @ARG;  
    $w->configure($opt => $val1);     
    $MW->after($interval, [\&flash_widget, $w, $opt,
			   $val2, $val1, $interval]);  

As you see, the code is quite simple. On the first call to flash_widget(), the button's background is configured blue. A timer event is then scheduled, reversing the background colors, so next time the widget is configured yellow. The periodic change in background color every 500 milliseconds yields the desired flashing effect. A working example, named flash, is on the TPJ web site.

You can also perform crude animations with nothing more than standard Tk timer events. To demonstrate, I created a basic neko program, using frames borrowed from Masayuki Koba's well known xneko. In case you're unfamiliar with xneko, a cat chases the cursor around an X window. When you stop moving the cursor, the cat gives a mighty yawn and settles down to take a nap. When the cursor moves again, neko wakes up and resumes the chase. My rendition of neko doesn't follow the cursor and moves solely in one dimension. (Readers with free time on their hands should note that I've laid the groundwork for implementing a full-fledged tkneko. Use ideas from last issue's Mouse Odometer program, throw in a little trigonometry and mix well with my neko code, and let me see what you can do!)

In the US, television creates the illusion of motion by flashing 30 full images per second. Movies show 24 images per second, but flash each image three times to lessen the flicker. Psychophysicists have determined that 10 images per second is, on average, the minimum number needed to perceive motion, so that's what we'll use for neko. I don't actually have ten images to show, just two: one of neko with his feet together, and one with his feet apart.

When you run neko, the following screen, depicting the six frames used by the application, is momentarily displayed:

Figure 1: Presenting Neko

Figure 1: Presenting Neko

To make use of these frames we need to create Tk images. An image is just another Tk object with special methods for image manipulations. Once created, images are then imported into other widgets, such as a button, canvas or label. For example, this code creates a button with neko's icon on it instead of text:

my $i = $MW->Photo(-file => 'images/icon.ppm'); 
my $b = $MW->Button(  -image => $i, 
                    -command => sub {print "Meow\n"})->pack; 

Images come in two flavors, bitmaps, which have only two colors, and photos, which are photographic quality objects having millions of colors. The six neko frames were originally plain X bitmaps, but have since been converted to colorized PPM files, a format (like GIF) suitable for input to the Photo() command.

The canvas widget provides an ideal backdrop for the animation, since images can be drawn on it and moved using standard canvas methods. Here's the code that created much of the above figure:

# Create the six Photo images from the color PPM files
# and display them in a row. The canvas image IDs
# are stored in the global array @IDS for use by the
# rest of the neko code.  For instance, to perform a
# canvas operation on the neko icon, simply fetch its
# item ID from $IDS[5]. Sorry for using hardcoded 
# values, but this is just "proof of concept" code!  

my $x = 125; 

foreach ( qw(left1 left2 sleep1 sleep2 awake icon) ) {
    push @IDS, $C->create('image', $x, $SCAMPER_Y, 			  
         -image => $MW->Photo(-file => "images/$ARG.ppm"));
    $x += 50; 

# Wait for the main window to appear before hiding the neko
# frames. (Otherwise you might never get to see them.)

$MW->after(2000, \&hide_nekos); 

An immediate problem arises: the animation demands that only one frame be visible at any point in time, so we need to hide arbitrary frames (including the six frames currently on the canvas). One way might be to create and delete the images continually, but that's messy. Instead, neko uses a trick based on the canvas' display list.

Tk uses the display list to control the order in which canvas items are displayed, so that items created later are displayed after items created earlier. If two items are positioned at the same (or overlapping) coordinates, the item earliest in the display list is obscured because the other item is displayed on top of it. Thus, the rightmost item in Figure 1, the neko icon, is on top of the display list. We'll move the icon off to the side, hide all inactive images under it, and no one will be the wiser!

my($i, $done, $rptid, $cb) = ($#IDS, 0, 0, 0);  

$cb = sub { 
            my($ir) = @ARG; 
            hide_frame $IDS[$$ir--];
            $done++ if $$ir< 0
my $rptid = $MW->repeat(1000 => [$cb, \$i]);  


There's more to these five statements than meets the eye, so let's examine them one by one. We want to move the icon image first, so set $i to its index in the @IDS array. Even though the icon is the first image moved it will nevertheless obscure the remaining images because it's at the end of the display list.

The second statement defines a timer callback, $cb, whose sole purpose is to hide one neko frame, decrement the index $i and set the $done flag after the last image has been moved. Here's where it gets tricky: the parameter passed to the anonymous subroutine is not the value of $i itself, but $$i, a reference to $i. Passing $i directly would only postdecrement the copy local to the subroutine, $ir, and not the "real" $i. Thus, only the icon frame would be moved, and the callback would never set the $done flag.

The repeat() queues a timer event which, until cancelled, repeats once a second, forever. However, the callback has been designed to modify the $done variable after the last image has been hidden. Notice that repeat(), like all asynchronous timer event scheduling methods, returns a timer ID, used to subsequently remove the event from the timer queue.

The waitVariable() waits until the value of $done changes. Although the application's flow is logically suspended, it still responds to external events, and so is not frozen.

The afterCancel() cancels the repeat() event. The end result is that the images shown in Figure 1 are hidden, one at a time, once a second, from right to left. You can see below what the window looks like after all the neko images have been moved off to the side. Note the neko icon, sitting in the upper left corner, hiding most of the other images. The snoozing neko has subsequently been unhidden and animated for your viewing pleasure. So, how do we make neko scamper across the canvas? This code snippet does just that:

# Move neko right to left by exposing successive
# frames for 0.1 second.  

my $cb = sub {$done++}; 
my($i, $k) = (0, -1); 
$delay = 100; 

for ($i = 460; $i >= 40; $i -= $DELTA_X) {
    $id = $IDS[++$k % 2];     
    move_frame($id, $i, $SCAMPER_Y);     
    if ($BLOCK) { 	$MW->after($delay) } 
    else { 	
        $MW->after($delay => $cb); 	
    hide_frame $id; 


Take one last look at Figure 1 and note the two leftmost images. Essentially, all we need to do is periodically display those images, one after another, at slightly different positions on the canvas. The scamper code shown above does just that: move one image from underneath the neko icon, wait for 0.1 second, hide it, unhide the second image and display it slightly to the left of the previous, wait for 0.1 second, and repeat until neko reaches the left edge of the canvas. The exhausted neko then takes a well deserved nap.

It's possible to animate neko using a blocking or non-blocking technique, depending on the state of the Block? checkbutton. Try each alternative and note how the buttons respond as you pass the cursor over them. $DELTA_X controls how "fast" neko runs, and is tied to the slender scale widget to the right of the window. Varying its value by moving the slider makes neko either moonwalk or travel at relativistic speeds!

Figure 1: Blocking or not Blocking Neko

Figure 2: Blocking or not Blocking Neko Before we move on, here is how neko images are actually translated (moved) across the canvas (or "hidden" and "unhidden"):

sub move_frame {  

    # Move a neko frame to an absolute canvas 
    # position.  

    my($id, $absx, $absy) = @ARG;     
    my ($x, $y) = $C->coords($id);     
    $C->move($id, $absx-$x, $absy-$y);     

The canvas move() method moves an item to a new position on the canvas relative to its current position. Here we don't even know the absolute coordinates, so we use coords() to get the neko's current position and perform a subtraction to determine the X and Y differences needed. When a neko image is hidden it's simply moved to the "hide" coordinates occupied by the neko icon. The idletasks() statement flushes the idle events queue, ensuring that the display is updated immediately.

I/O Events

If you think about it, a Tk application is somewhat analogous to a multi-tasking operating system: event callbacks must be mutually cooperative and only execute for a reasonable amount of time before relinquishing control to other handlers; otherwise, the application might freeze. This is an important consideration if your Tk application performs terminal, pipe, or socket I/O, since these operations might very well block, taking control away from the user.

Suppose you want to write a small program where you can interactively enter Perl/Tk commands, perhaps to prototype small code snippets of a larger application. The code might look like this:

use Tk; 
while (<>) { 
    eval $ARG; 

When prompted for input you could then enter commands like this:

my $MW = MainWindow->new; 
my $b = $MW->Button(-text => 'Hello world!')->pack;  

What happens now? Actually, nothing! Is the button displayed? No! Why not? Easy: no MainLoop() statement has been executed, so no events are processed, and the display isn't updated!

Realizing what's happening, you then enter a MainLoop() statement, and lo and behold, something appears! But now you're stuck, because MainLoop() never returns until the main window is destroyed,* so once again you're blocked and prevented from entering new Tk commands!

One solution is to rewrite your Perl/Tk shell using fileevent(), the I/O event handler:

$MW->fileevent('STDIN', 'readable' => \&user_input); MainLoop;  

sub user_input {  

    # Called when input is available on STDIN.  

    $ARG =<>;     
    eval $ARG; 

The key difference is that the read from STDIN is now a non-blocking event, which is invoked by MainLoop() whenever input data is available. The fileevent() command expects three arguments: a file handle, an I/O operation ('readable' or 'writable'), and a callback to be invoked when the designated file handle is ready for input or output.

Although not necessary here, it's good practice to delete all file event handlers, in the same spirit as closing files and canceling timer events:

$MW->fileevent('STDIN', 'readable' => ''); 

The entire ptksh1 program is on the TPJ web site. Another program, tktail, demonstrating a pipe I/O event handler, is available from Peter Prymmer's Perl/Tk FAQ at:

Idle Events

The idle event queue isn't restricted to redisplaying. You can use it for low priority callbacks of your own. This silly example uses afterIdle() to ring the bell after 5 seconds:

#!/usr/bin/perl -w 
# Demonstrate use of afterIdle() to queue a 
# low priority callback.  

require 5.002; 
use English; 
use Tk; 
use Tk::DoOneEvent; 
use strict;  

my $MW = MainWindow->new; 
$MW->Button(-text => 'afterIdle', 
         -command => \&queue_afterIdle)->pack; 

sub queue_afterIdle {     
    $MW->afterIdle(sub {$MW->bell});     
    print "afterIdle event queued, block for 5 seconds...\n";
    print "It's 5 seconds later, call idletasks() ",
          "to activate the handler.\n";     
    print "The bell should have sounded ...\n";     


To recap, we are responsible for three event-related activities: registering events, creating event handlers and dispatching events. Until now MainLoop() has dispatched events for us, running in an endless loop, handing off events to handlers as they arise, and putting the application to sleep if no events are pending. When the application's last main window is destroyed, MainLoop() returns and the program terminates.

Perl/Tk allows low-level access to Tk events via DoOneEvent(). This event dispatcher is passed a single argument: a bit pattern describing which events to process. As you might guess, the event categories are those we've just explored. Direct access to DoOneEvent() from the Tk level is normally unavailable, but the C interface exports it specifically for Perl/Tk. The C interface has a header file that defines event symbols, but these symbols haven't reached the Tk level yet. To avoid the dubious use of numeric values, here is a small package that you can use() to import the appropriate bit patterns (shown below):

package Tk::DoOneEvent;  

require Exporter; 
@Tk::DoOneEvent::ISA = qw(Exporter); 
             $TK_IDLE_EVENTS     $TK_ALL_EVENTS);  

# Bit patterns describing what you want DoOneEvent() to do,
# swiped directly from the Tk distribution.  If you know where
# tk.h resides, I suggest using "h2ph" to generate the symbol
# definitions in case the header file changes.  

$TK_WAIT         = 0x00; 
$TK_DONT_WAIT    = 0x01; 
$TK_X_EVENTS     = 0x02; 
$TK_FILE_EVENTS  = 0x04; 
$TK_IDLE_EVENTS  = 0x10; 
                 $TK_TIMER_EVENTS | $TK_IDLE_EVENTS;  


As it turns out, MainLoop() is implemented using DoOneEvent(), similar to this meta-code:

MainLoop { 
    while (NumMainWindows > 0) {         

When passed $TK_WAIT or $TK_ALL_EVENTS, DoOneEvent() processes events as they arise, and puts the application to sleep when no further events are outstanding. DoOneEvent() first looks for an X or I/O event and, if found, calls the handler and returns. If there is no X or I/O event, it looks for a single timer event, invokes the callback, and returns. If no X, I/O or timer event is ready, all pending idle callbacks are executed. In all cases DoOneEvent() returns 1.

When passed $TK_DONT_WAIT, DoOneEvent() works as above, except that if there are no events to process, it returns immediately with a value of 0, indicating it didn't find any work to do.

With this new knowledge, here is another implementation of our Perl/Tk shell that doesn't need fileevent():

#!/usr/bin/perl -w 
# ptksh2 - another Perl/Tk shell using DoOneEvent()
# rather than fileevent().  

require 5.002; 
use English; 
use Tk; 
use Tk::DoOneEvent; 
use strict;  

my $MW = MainWindow->new; 

while(1) {     
    while(1) {        
        last unless DoOneEvent($TK_DONT_WAIT);     
    print "ptksh> ";     
    { no strict; eval<>; }
    print $EVAL_ERROR if $EVAL_ERROR; 

The outer while loop accepts terminal input, and the inner while loop cycles as long as Tk events arise as a result of that input.


At the risk of creating mass depression, I must confess that this implementation of pong isn't the real thing. You won't see multiple game levels of ever increasing difficulty, or even a database of high scores. Nope, all you get is a basic paddle, one ball, and the chance to bounce the ball around until you grow bored, which for me took less than a minute. The idea in this game is to keep the ball confined within the playing field; you get a point every time you hit the ball with the paddle, but lose a point every time the ball hits the floor or ceiling. This means that the paddle is tied to your mouse and follows its every motion. If at game's end the score is positive you win, else you lose. pong is derived in large part from bounce, the widget bouncing ball simulation written by Gurusamy Sarathy.

Figure 3: Breakout Simulator

Figure 3: Breakout Simulator

Of course pong isn't meant to be fun, but to showcase Perl/Tk features: events, canvas commands and the Pong derived widget.

pong really wants to be a CPU resource hog in order to keep the ball and paddle lively, but at the same time it needs to allow Tk events safe passage, so it has its own version of MainLoop():

while (1) {  
   exit if $QUIT;     
   $pong->move_balls($SPEED->get / 100.0) if $RUNNING;  
The variable $RUNNING is a boolean indicating whether the game is in progress or has been paused. If the game has been paused ($RUNNING = 0), DoOneEvent() is called with $TK_WAIT, and sleeps until Tk events arise, but the ball and paddle aren't moved. Otherwise, DoOneEvent() is called with $TK_DONT_WAIT, which may process one or more events (but certainly won't block), and then the game's ball and paddle are moved.

If this is the entire pong MainLoop, obviously the $PONG widget must be handling a lot behind the scenes. Indeed, the heart of the game is this single widget, which maintains the entire game state: paddle and ball position and movement, and game score. $PONG is a widget derived from a canvas, meaning that it automatically assumes all methods inherent in a canvas (and may provide more of its own, like move_balls()).

A properly defined derived widget like Pong follows standard Perl/Tk conventions:

$PONG = $drawarea->Pong(
        -relief => 'ridge', 
        -height => 400,
        -width  => 600, 
        -bd     => 2, 
        -balls  => [{-color    => 'yellow', 
                     -size     => 40,
                     -position => [90, 250]}]

This command creates a 400x600 pixel canvas, with one paddle and one ball (but there could have been more). The ball is yellow, 40 pixels in diameter, and is placed at canvas coordinates (90,250). Because the Pong widget ISA canvas, anything you can do with a canvas you can do with a Pong widget. Defining a derived widget class is similar to defining a composite class (like Odometer from last issue).

package Tk::Pong; 
require 5.002; 
use English; 
use Tk::Canvas; 
@Tk::Pong::ISA = qw(Tk::Derived Tk::Canvas); 

sub Populate {  # the Pong constructor
    my($dw, $args) = @ARG;  

    # Create needed canvas items here.     
    $dw->ConfigSpecs( ... );         

    return $dw;  


These statements

pong's Populate() procedure is really quite simple because it relies on existing canvas methods to create the game interface. This code automatically creates the paddle and one or more balls:

my $paddle = $dw->create('rectangle', @paddle_shape,
                          -fill    => 'orange',
                          -outline => 'orange',     
$dw->{paddle_ID} = $paddle;     
$dw->Tk::bind('<Motion>' => \&move_paddle);  

    -balls  => ['METHOD', undef, undef, [{}]], 
    -cursor => ['SELF', undef, undef,
              ($dw->configure(-background))[4], 'orange'] 

The create('rectangle') statement makes an orange paddle, whose shape is defined by the canvas coordinates of diagonally opposed rectangle corners. The paddle's canvas ID is saved in the object as an instance variable so that move_paddle() can move the paddle around the canvas - this private class method is bound to pointer motion events.

Once again, in general, Populate() should not directly configure its widget. That's why there's no code to create the ball. Instead, ConfigSpecs() is used to define the widget's valid configuration options (-balls is one) and how to handle them. When Populate() returns, Perl/Tk then examines the configuration specifications and auto-configures the derived widget.

A call to ConfigSpecs() consists of a series of keyword => value pairs, where the widget's keyword value is a list of 4 items: a string specifying exactly how to configure the keyword, its name in the X resource database, its class name, and its default value.

We've seen the ConfigSpecs() 'METHOD' option before: when Perl/Tk sees a -balls attribute it invokes a method of the same name, minus the dash: balls(). And if you examine the source code on the TPJ web site, you'll note that all subroutine balls() really does is execute a $PONG->create('oval') command.

The -cursor ConfigSpecs() option is moderately interesting. 'SELF' says to apply the option to the derived widget itself. But why do we want to change the canvas' cursor? Well, just waggle your mouse around and watch the cursor closely. Very often it's shaped like an arrow, or perhaps an underscore, rectangle, I-beam, or X. But in a Pong game, when you move the mouse you only want to see the paddle move, not the paddle and a tag-along cursor. So pong defines a cursor consisting of a single orange pixel and associates it with the Pong widget, neatly camouflaging the cursor.

Like neko, the Pong widget uses the canvas move() method to move the paddle around, but is driven by X motion events rather than timer events. An X motion event invokes move_paddle():

sub move_paddle {  
    my($canvas) = @ARG;  
    my $e = $canvas->XEvent;     
    my($x, $y) = ($e->x, $e->y);     

        $x - $canvas->{last_paddle_x}, 
        $y - $canvas->{last_paddle_y});     

    $canvas->{last_paddle_x}, $canvas->{last_paddle_y})
        = ($x, $y);

This subroutine extracts the cursor's current position from the X event structure, executes move() using instance data from the Pong widget, and saves the paddle's position for next time.

That takes care of paddle motion, but ball motion we handle ourselves, via the class method move_balls(), which has its own DoOneEvent() mini MainLoop. Ball movement boils down to yet another call to the move() canvas method, with incidentals like checking for collisions with walls or the paddle. Here's the code:

sub move_balls {  

    # Move all the balls one "tick." We call DoOneEvent()
    # in case there are many balls; with only one it's
    # not strictly necessary.

    my($canvas, $speed) = @ARG;  
    my $ball;     
    foreach $ball (@{$canvas->{balls}}) {         
        $canvas->move_one_ball($ball, $speed); 

        # be kind and process XEvents as they arise     
Although the details of reflecting a ball and detecting collisions are interesting, they're not relevant to our discussion, so feel free to examine move_one_ball() yourself.

Miscellaneous Event Commands

There are three other event commands that merit a little more explanation: update(), waitWindow(), and waitVisibility(). The update() method is useful for CPU-intensive programs in which you still want the application to respond to user interactions. If you occasionally call update() then all pending Tk events will be processed and all windows updated.

The waitWindow() method waits for a widget, supplied as its argument, to be destroyed. For instance, you might use this command to wait for a user to finish interacting with a dialog box before using the result of that interaction. However, doing this requires creating and destroying the dialog each time. If you're concerned about efficiency, try withdrawing the window instead. Then use waitVisibility() to wait for a change in the dialog's visibility state.

And so, until next time, when you manage events, think "mutually cooperative."