[Previous] [Contents] [Next]

24. Scribble, A Simple Example Drawing Program

24.1 Overview

In this section, we will build a simple drawing program. In the process, we will examine how to handle mouse events, how to draw in a window, and how to do drawing better by using a backing pixmap. After creating the simple drawing program, we will extend it by adding support for XInput devices, such as drawing tablets. GTK provides support routines which makes getting extended information, such as pressure and tilt, from such devices quite easy.

24.2 Event Handling

The GTK signals we have already discussed are for high-level actions, such as a menu item being selected. However, sometimes it is useful to learn about lower-level occurrences, such as the mouse being moved, or a key being pressed. There are also GTK signals corresponding to these low-level events. The handlers for these signals have an extra parameter which is a pointer to a structure containing information about the event. For instance, motion event handlers are passed a pointer to a GdkEventMotion record which looks (in part) like:

   PGdkEventMotion = ^TGdkEventMotion;
   TGdkEventMotion = record
                        _type : TGdkEventType;
                        window : PGdkWindow;
                        time : guint32;
                        x : gdouble;
                        y : gdouble;
                        state : guint;

_type will be set to the event type, in this case GDK_MOTION_NOTIFY, window is the window in which the event occurred. x and y give the coordinates of the event. state specifies the modifier state when the event occurred (that is, it specifies which modifier keys and mouse buttons were pressed). It is the bitwise or of some of the following:


As for other signals, to determine what happens when an event occurs we call gtk_signal_connect(). But we also need let GTK know which events we want to be notified about. To do this, we call the procedure:

procedure gtk_widget_set_events (widget : PGtkWidget; events : gint);

The second field specifies the events we are interested in. It is the bitwise or of constants that specify different types of events. For future reference the event types are:


There are a few subtle points that have to be observed when calling gtk_widget_set_events(). First, it must be called before the X window for a GTK widget is created. In practical terms, this means you should call it immediately after creating the widget. Second, the widget must have an associated X window. For efficiency, many widget types do not have their own window, but draw in their parent's window. These widgets are:


To capture events for these widgets, you need to use an EventBox widget. See the section on the EventBox widget for details.

For our drawing program, we want to know when the mouse button is pressed and when the mouse is moved, so we specify GDK_POINTER_MOTION_MASK and GDK_BUTTON_PRESS_MASK. We also want to know when we need to redraw our window, so we specify GDK_EXPOSURE_MASK. Although we want to be notified via a Configure event when our window size changes, we don't have to specify the corresponding GDK_STRUCTURE_MASK flag, because it is automatically specified for all windows.

It turns out, however, that there is a problem with just specifying GDK_POINTER_MOTION_MASK. This will cause the server to add a new motion event to the event queue every time the user moves the mouse. Imagine that it takes us 0.1 seconds to handle a motion event, but the X server queues a new motion event every 0.05 seconds. We will soon get way behind the users drawing. If the user draws for 5 seconds, it will take us another 5 seconds to catch up after they release the mouse button! What we would like is to only get one motion event for each event we process. The way to do this is to specify GDK_POINTER_MOTION_HINT_MASK.

When we specify GDK_POINTER_MOTION_HINT_MASK, the server sends us a motion event the first time the pointer moves after entering our window, or after a button press or release event. Subsequent motion events will be suppressed until we explicitly ask for the position of the pointer using the function:

function gdk_window_get_pointer (window : PGdkWindow;
                                 x, y : pgint;
                                 mask : PGdkModifierType) : PGdkWindow;

(There is another function, gtk_widget_get_pointer() which has a simpler interface, but turns out not to be very useful, since it only retrieves the position of the mouse, not whether the buttons are pressed.)

The code to set the events for our window then looks like:

   g_signal_connect(drawing_area, 'expose_event',
                    G_CALLBACK(@expose_event), nil);
   g_signal_connect(drawing_area, 'configure_event',
                    G_CALLBACK(@configure_event), nil);
   g_signal_connect(drawing_area, 'motion_notify_event',
                    G_CALLBACK(@motion_notify_event), nil);
   g_signal_connect(drawing_area, 'button_press_event',
                    G_CALLBACK(@button_press_event), nil);

                         GDK_EXPOSURE_MASK or
                         GDK_LEAVE_NOTIFY_MASK or
                         GDK_BUTTON_PRESS_MASK or
                         GDK_POINTER_MOTION_MASK or

We'll save the "expose_event" and "configure_event" handlers for later. The "motion_notify_event" and "button_press_event" handlers are pretty simple:

function button_press_event (widget : PGtkWidget;
                             event : PGdkEventButton) : boolean; cdecl;
   if (event^.button = 1) and (pixmap <> nil) then
      draw_brush(widget, event^.x, event^.y);

   button_press_event := true;

function motion_notify_event (widget : PGtkWidget;
                              event : PGdkEventMotion) : boolean; cdecl;
   x, y : integer;
   state : TGdkModifierType;
   if (event^.is_hint <> 0) then
      gdk_window_get_pointer(event^.window, @x, @y, @state)
   else begin
      x := trunc(event^.x);
      y := trunc(event^.y);
      state := event^.state;

   if ((state and GDK_BUTTON1_MASK) <> 0) and (pixmap <> nil) then
      draw_brush(widget, x, y);

   motion_notify_event := true;

24.3 The DrawingArea Widget, and Drawing

We now turn to the process of drawing on the screen. The widget we use for this is the DrawingArea widget. A drawing area widget is essentially an X window and nothing more. It is a blank canvas in which we can draw whatever we like. A drawing area is created using the call:

function gtk_drawing_area_new () : PGtkWidget;

A default size for the widget can be specified by calling:

procedure gtk_widget_set_size_request (widget : PGtkWidget;
                                       width : gint; height : gint);

This default size can be overridden if the user manually resizes the the window containing the drawing area.

It should be noted that when we create a DrawingArea widget, we are completely responsible for drawing the contents. If our window is obscured then uncovered, we get an exposure event and must redraw what was previously hidden.

Having to remember everything that was drawn on the screen so we can properly redraw it can, to say the least, be a nuisance. In addition, it can be visually distracting if portions of the window are cleared, then redrawn step by step. The solution to this problem is to use an offscreen backing pixmap. Instead of drawing directly to the screen, we draw to an image stored in server memory but not displayed, then when the image changes or new portions of the image are displayed, we copy the relevant portions onto the screen.

To create an offscreen pixmap, we call the function:

function gdk_pixmap_new (window : PGdkDrawable;
                         width, height, depth : gint) : PGdkPixmap;

The drawable parameter specifies a GDK drawable (such as a GdkWindow) that this pixmap takes some of its properties from. width and height specify the size of the pixmap. depth specifies the color depth, that is the number of bits per pixel, for the new window. If the depth is specified as -1, it will match the depth of drawable.

We create the pixmap in our "configure_event" handler. This event is generated whenever the window changes size, including when it is originally created.

{ Backing pixmap for drawing area }

   pixmap : PGdkPixmap = nil;

{ Create a new backing pixmap of the appropriate size }

function configure_event (widget : PGtkWidget;
                          event : PGdkEventConfigure): boolean; cdecl;
   if pixmap <> nil then
   pixmap := gdk_pixmap_new(widget^.window,
                      0, 0,
   configure_event := true;

The call to gdk_draw_rectangle() clears the pixmap initially to white. We'll say more about that in a moment.

Our exposure event handler then simply copies the relevant portion of the pixmap onto the screen (we determine the area we need to redraw by using the event^.area field of the exposure event):

{ Redraw the screen from the backing pixmap }

function expose_event (widget : PGtkWidget;
                       event : PGdkEventExpose) : boolean; cdecl;
                     event^.area.x, event^.area.y,
                     event^.area.x, event^.area.y,
                     event^.area.width, event^.area.height);
   expose_event := false;

We've now seen how to keep the screen up to date with our pixmap, but how do we actually draw interesting stuff on our pixmap? There are a large number of calls in GTK's GDK library for drawing on drawables. A drawable is simply something that can be drawn upon. It can be a window, a pixmap, or a bitmap (a black and white image). We've already seen two such calls above, gdk_draw_rectangle() and gdk_draw_drawable(). The complete list is:

gdk_draw_point ()
gdk_draw_line ()
gdk_draw_rectangle ()
gdk_draw_arc ()
gdk_draw_polygon ()
gdk_draw_pixmap ()
gdk_draw_bitmap ()
gdk_draw_image ()
gdk_draw_points ()
gdk_draw_segments ()
gdk_draw_lines ()
gdk_draw_pixbuf ()
gdk_draw_glyphs ()
gdk_draw_layout_line ()
gdk_draw_layout ()
gdk_draw_layout_line_with_colors ()
gdk_draw_layout_with_colors ()
gdk_draw_glyphs_transformed ()
gdk_draw_glyphs_trapezoids ()

See the reference documentation or the Pascal source file gdk/gdkmain.pp for further details on these functions. These functions all share the same first two arguments. The first argument is the drawable to draw upon, the second argument is a graphics context (GC).

A graphics context encapsulates information about things such as foreground and background color and line width. GDK has a full set of functions for creating and modifying graphics contexts, but to keep things simple we'll just use predefined graphics contexts. Each widget has an associated style. (Which can be modified in a gtkrc file, see the section on GTK's rc file.) This, among other things, stores a number of graphics contexts. Some examples of accessing these graphics contexts are:


The fields fg_gc, bg_gc, dark_gc, and light_gc are indexed by a parameter of type TGtkStateType which can take on the values:


For instance, for GTK_STATE_SELECTED the default foreground color is white and the default background color, dark blue.

Our function draw_brush(), which does the actual drawing on the screen, is then:

{ Draw a rectangle on the screen }

procedure draw_brush (widget : PGtkWidget; x, y : gdouble);
   update_rect : TGdkRectangle;
   update_rect.x := trunc(x - 5);
   update_rect.y := trunc(y - 5);
   update_rect.width := 10;
   update_rect.height := 10;
                      update_rect.x, update_rect.y,
                      update_rect.width, update_rect.height);
                              update_rect.x, update_rect.y,
                              update_rect.width, update_rect.height);

After we draw the rectangle representing the brush onto the pixmap, we call the procedure:

procedure gtk_widget_queue_draw_area (widget : PGtkWidget;
                                      x, y : gint;
                                      width, height : gint);

which notifies X that the area given by the x, y, width and height parameters needs to be updated. X will eventually generate an expose event (possibly combining the areas passed in several calls to gtk_widget_queue_draw_area()) which will cause our expose event handler to copy the relevant portions to the screen.

We have now covered the entire drawing program except for a few mundane details like creating the main window.

24.4 Adding XInput support

It is now possible to buy quite inexpensive input devices such as drawing tablets, which allow drawing with a much greater ease of artistic expression than does a mouse. The simplest way to use such devices is simply as a replacement for the mouse, but that misses out many of the advantages of these devices, such as:

For information about the XInput extension, see the XInput-HOWTO.

If we examine the full definition of, for example, the TGdkEventMotion record, we see that it has fields to support extended device information.

   PGdkEventMotion = ^TGdkEventMotion;
   TGdkEventMotion = record
                        _type : TGdkEventType;
                        window : PGdkWindow;
                        time : guint32;
                        x : gdouble;
                        y : gdouble;
                        pressure : gdouble;
                        xtilt : gdouble;
                        ytilt : gdouble;
                        state : guint;
                        is_hint : gint16;
                        source : TGdkInputSource;
                        deviceid : guint32;

pressure gives the pressure as a floating point number between 0 and 1. xtilt and ytilt can take on values between -1 and 1, corresponding to the degree of tilt in each direction. source and deviceid specify the device for which the event occurred in two different ways. source gives some simple information about the type of device. It can take the enumeration values:


deviceid specifies a unique numeric ID for the device. This can be used to find out further information about the device using the gdk_input_list_devices() call (see below).

Enabling extended device information

To let GTK know about our interest in the extended device information, we merely have to add a single line to our program:

   gtk_widget_set_extension_events(drawing_area, GDK_EXTENSION_EVENTS_CURSOR);

By giving the value GDK_EXTENSION_EVENTS_CURSOR we say that we are interested in extension events, but only if we don't have to draw our own cursor. See the section Further Sophistications below for more information about drawing the cursor. We could also give the values GDK_EXTENSION_EVENTS_ALL if we were willing to draw our own cursor, or GDK_EXTENSION_EVENTS_NONE to revert back to the default condition.

This is not completely the end of the story however. By default, no extension devices are enabled. We need a mechanism to allow users to enable and configure their extension devices. GTK provides the InputDialog widget to automate this process. The following procedure manages an InputDialog widget. It creates the dialog if it isn't present, and raises it to the top otherwise.

   inputd : PGtkWidget = nil;

procedure input_dialog_destroy (widget : PGtkWidget;
                                data : PGtkInputDialog); cdecl;
   inputd := nil;

procedure create_input_dialog ();
   if (inputd = nil) then begin
      inputd := gtk_input_dialog_new();

      g_signal_connect(GTK_OBJECT(inputd), 'destroy',
                       G_CALLBACK(@input_dialog_destroy), inputd);

   else begin
      if not GTK_WIDGET_MAPPED(inputd) then

(You might want to take note of the way we handle this dialog. By connecting to the "destroy" signal, we make sure that we don't keep a pointer to the dialog around after it is destroyed - that could lead to a segfault.)

The InputDialog has two buttons "Close" and "Save", which by default have no actions assigned to them. In the above procedure we make "Close" hide the dialog, and we hide the "Save" button, since we don't implement saving of XInput options in this program.

Using Extended Device Information

Once we've enabled the device, we can just use the extended device information in the extra fields of the event records. In fact, it is always safe to use this information since these fields will have reasonable default values even when extended events are not enabled.

One change we do have to make is to call gdk_input_window_get_pointer() instead of gdk_window_get_pointer(). This is necessary because gdk_window_get_pointer() doesn't return the extended device information.

procedure gdk_input_window_get_pointer (window : PGdkWindow;
                                        deviceid : guint32;
                                        x, y : pgdouble;
                                        pressure : pgdouble;
                                        xtilt : pgdouble;
                                        ytilt : pgdouble;
                                        mask : PGdkModifierType);

When calling this procedure, we need to specify the device ID as well as the window. Usually, we'll get the device ID from the deviceid field of an event record. Again, this function will return reasonable values when extension events are not enabled. (In this case, event^.deviceid will have the value GDK_CORE_POINTER).

So the basic structure of our button-press and motion event handlers doesn't change much - we just need to add code to deal with the extended information.

function button_press_event (widget : PGtkWidget;
                             event: PGdkEventButton) : boolean; cdecl;
   pressure : gdouble;
   if (event^.button = 1) and (pixmap <> nil) then then begin
      gdk_event_get_axis(PGdkEvent(event), GDK_AXIS_PRESSURE, @pressure);
      draw_brush(widget, event^.device^.source, event^.x, event^.y, pressure);

   button_press_event := true;

function motion_notify_event (widget : PGtkWidget;
                              event : PGdkEventMotion) : boolean; cdecl;
   x, y : gdouble;
   presssure : gdouble;
   state : TGdkModifierType;
   if (event^.is_hint <> 0) then begin
      gdk_device_get_state(event^.device, event^.window, nil, @state);
      gdk_event_get_axis(PGdkEvent(event), GDK_AXIS_X, @x);
      gdk_event_get_axis(PGdkEvent(event), GDK_AXIS_Y, @y);
      gdk_event_get_axis(PGdkEvent(event), GDK_AXIS_PRESSURE, @pressure);
   else begin
      x := event^.x;
      y := event^.y;
      gdk_event_get_axis(PGdkEvent(event), GDK_AXIS_PRESSURE, @pressure);
      state := event^.state;
   if ((state and GDK_BUTTON1_MASK) <> 0) and (pixmap <> nil) then
      draw_brush(widget, event^.device^.source, x, y, pressure);
   motion_notify_event := true;

We also need to do something with the new information. Our new draw_brush() function draws with a different color for each event^.source and changes the brush size depending on the pressure.

{ Draw a rectangle on the screen, size depending on pressure,
  and color on the type of device }

procedure draw_brush (widget : PGtkWidget; source : TGdkInputSource;
                      x, y : gdouble;
                      pressure : gdouble);
   gc : PGdkGC;
   update_rect : TGdkRectangle;
   case source of
     gc := gtk_widget_get_style(widget)^.dark_gc[GTK_WIDGET_STATE(widget)];
     gc := gtk_widget_get_style(widget)^.black_gc;
     gc := gtk_widget_get_style(widget)^.white_gc;
     gc := gtk_widget_get_style(widget)^.light_gc[GTK_WIDGET_STATE(widget)];
   update_rect.x := trunc(x - 10 * pressure);
   update_rect.y := trunc(y - 10 * pressure);
   update_rect.width := trunc(20 * pressure);
   update_rect.height := trunc(20 * pressure);
   gdk_draw_rectangle(pixmap, gc, gint(true),
                      update_rect.x, update_rect.y,
                      update_rect.width, update_rect.height);
                              update_rect.x, update_rect.y,
                              update_rect.width, update_rect.height);

Finding out more about a device

As an example of how to find out more about a device, our program will print the name of the device that generates each button press. To find out the name of a device, we call the function:

function gdk_input_list_devices () : PGList;

which returns a GList (a linked list type from the GLib library) of GdkDeviceInfo records. The GdkDeviceInfo record is defined as:

   PGdkDeviceInfo = ^TGdkDeviceInfo;
   TGdkDeviceInfo = record
                       deviceid : guint32;
                       name : pgchar;
                       source : TGdkInputSource;
                       mode : TGdkInputMode;
                       has_cursor : gint;
                       num_axes : gint;
                       axes : ^TGdkAxisUse;
                       num_keys : gint;
                       keys : PGdkDeviceKey;

Most of these fields are configuration information that you can ignore unless you are implementing XInput configuration saving. The field we are interested in here is name which is simply the name that X assigns to the device. The other field that isn't configuration information is has_cursor. If has_cursor is false, then we need to draw our own cursor. But since we've specified GDK_EXTENSION_EVENTS_CURSOR, we don't have to worry about this.

Our print_button_press() procedure simply iterates through the returned list until it finds a match, then prints out the name of the device.

procedure print_button_press (deviceid : guint32);
   tmp_list : pGList;
   info : PGdkDeviceInfo;
   { gdk_input_list_devices returns an internal list, so we shouldn't 
     free it afterwards }
   tmp_list := gdk_input_list_devices();
   while tmp_list <> nil do begin
      info := PGdkDeviceInfo(tmp_list^.data);
      if (info^.deviceid = deviceid) then begin
         writeln('Button press on device ', info^.name);

      tmp_list := tmp_list^.next;

That completes the changes to "XInputize" our program.

Further Sophistications

Although our program now supports XInput quite well, it lacks some features we would want in a full-featured application. First, the user probably doesn't want to have to configure their device each time they run the program, so we should allow them to save the device configuration. This is done by iterating through the return of gdk_input_list_devices() and writing out the configuration to a file.

To restore the state next time the program is run, GDK provides functions to change device configuration:


(The list returned from gdk_input_list_devices() should not be modified directly.) A C language example of doing this can be found in the drawing program gsumi. (Available from http://www.msc.cornell.edu/~otaylor/gsumi/) Eventually, it would be nice to have a standard way of doing this for all applications. This probably belongs at a slightly higher level than GTK, perhaps in the GNOME library.

Another major omission that we have mentioned above is the lack of cursor drawing. Platforms other than XFree86 currently do not allow simultaneously using a device as both the core pointer and directly by an application. See the XInput-HOWTO for more information about this. This means that applications that want to support the widest audience need to draw their own cursor.

An application that draws its own cursor needs to do two things: determine if the current device needs a cursor drawn or not, and determine if the current device is in proximity. (If the current device is a drawing tablet, it's a nice touch to make the cursor disappear when the stylus is lifted from the tablet. When the device is touching the stylus, that is called "in proximity".) The first is done by searching the device list, as we did to find out the device name. The second is achieved by selecting "proximity_out" events. An example of drawing one's own cursor is found in the "testinput" program found in the GTK distribution.

[Previous] [Contents] [Next]