Linux::Joystick is an object-oriented, pure Perl API for accessing joystick devices under Linux-based operating systems. It is capable of using either blocking or non-blocking I/O, and represents each axis change or button press as a Linux::Joystick::Event object.


If you want your application to be driven by joystick events, use blocking I/O and an event loop:

        use Linux::Joystick;

        my $js = new Linux::Joystick;
        my $event;

        print "Joystick has " . $js->buttonCount() . " buttons ".
                "and " . $js->axisCount() . " axes.\n";

        # blocking reads:
        while( $event = $js->nextEvent ) {

                print "Event type: " . $event->type . ", ";
                if($event->isButton) {
                        print "Button " . $event->button;
                        if($event->buttonDown) {
                                print " pressed";
                        } else {
                                print " released";
                } elsif($event->isAxis) {
                        print "Axis " . $event->axis . ", value " . $event->axisValue . ", ";
                        print "UP" if $event->stickUp;
                        print "DOWN" if $event->stickDown;
                        print "LEFT" if $event->stickLeft;
                        print "RIGHT" if $event->stickRight;
                } else { # should never happen
                        print "Unknown event " . $event->hexDump;

                print "\n";

        # if the while loop terminates, we got a false (undefined) event:
        die "Error reading joystick: " . $js->errorString;

You can also use non-blocking I/O, in which case nextEvent() returning undef just means there was no event to read:

        my $js = Linux::Joystick->new(nonblocking => 1);
        # use this to open 2nd joystick in nonblocking mode instead:
        # my $js = Linux::Joystick->new(device => 1, nonblocking => 1);

        while(1) {
                my $event = $js->nextEvent;

                if($event) {
                        print "Got a joystick event\n";
                        # process the event here

                # Do other processing here (graphics, sound, I/O, calculation)


It is possible to switch between blocking and non-blocking I/O without reopening the device (see the setNonblocking() method, below).


Create a new Linux::Joystick object using the default joystick device (usually the first on the system):

        my $js = Linux::Joystick->new; # Default device, same as new(0)

Same as above, but for a specific joystick (numbered starting with 0):

        # Default device (first joystick)
        my $js = Linux::Joystick->new(0);
        my $js = Linux::Joystick->new(device => 0); # same thing

        # Second joystick (player 2?)
        my $js = Linux::Joystick->new(1);
        my $js = Linux::Joystick->new(device => 1); # same thing

By default, we search for joystick devices by prepending the string /dev/input/js to the device number, then falling back to /dev/js if that fails. Most modern Linux systems will have /dev/js0 as a symlink to /dev/input/js0 anyway.

If you need to, you can also use a constructor like this:

        my $js = Linux::Joystick->new("/dev/js0");
        my $js = Linux::Joystick->new(device => "/dev/js0"); # same thing

...but this practice isn't recommended: if next year Red Hat decides to call their joystick device /dev/gamecontroller0, I (or someone) will update this Perl module to reflect that fact, and your perl scripts that use the numeric form will continue to work.

Any of these constructors will return undef on failure. The $! variable might or might not contain a useful error message. Possible failure reasons include the usual suspects (no joystick plugged in, no driver loaded, no permission to read from the device node, cat chewed through the USB cable, etc).

Creating multiple Linux::Joystick objects that read from the same device results in undefined behaviour, primarily because I haven't tested it, so I can't define it yet... but it's probably not a good idea (what happens will probably be dependent on what kernel version and kernel joystick driver you happen to be using).

Constructor parameters

The Linux::Joystick contructor uses named parameters, in the same way that IO::Socket and many other Perl modules do. Here is a constructor that sets all possible values to their defaults:

        # The following is exactly equivalent to just using
        # new() with no arguments:
        my $js = Linux::Joystick->new(
                        device => 0,
                        threshold => 1000,
                        nonblocking => 0,
                        fh => undef);

Here are the descriptions of all these parameters:


The device number to open. These are numbered starting from 0. Depending on your system configuration and how many joystick devices you have connected, you may have a large number of these to choose from. The default value is 0 (zero), which is the first joystick. If you specify a non-numeric parameter, it will be treated as the absolute path to a device node (such as /dev/js0). There are NO checks to make sure the path is actually a joystick device (or even a device node at all). Attempting to open a regular file or anything else other than a joystick device leads to unpredictable and generally useless behaviour.


Whether or not to use nonblocking I/O mode in the nextEvent method (1 or any true value for yes, 0 or any false value for no). This is off by default (0). Normally, in regular (blocking) mode, the nextEvent method blocks (waits) until a joystick event is received. With non-blocking I/O, nextEvent will return immediately. Its return value will be undef if there was no event ready (normally it always returns a valid event). Turning this on requires you to restructure your code somewhat (see examples above), but it allows your app to do other things while it's waiting for joystick movement.


How far the joystick must be moved from the center before it's registered as a directional movement. The default is 1000, which is appropriate for most (all?) digital controls, and for the analog thumb sticks on my `Axis Pad', but might be a bit too sensitive for a traditional analog `flight' stick. The bigger the threshold is, the bigger the `dead' zone will be, and the less `jitter' you'll experience. For digital (d-pad) style controls, there's no dead zone or jitter to worry about.


A Perl file handle reference (or glob). This is intended primarily for testing Linux::Joystick itself, but you could use it to e.g. read fake joystick events from a pipe or something. Use of this parameter causes the device parameter to be ignored. As with the device parameter, there is no check done to verify that the filehandle actually represents a joystick device. There is no default for this parameter.

Any parameter may be omitted, which will give that parameter its default value. A constructor with no arguments will cause all parameters to be set to their defaults.

The one-argument constructor is a convenient shorthand for setting the device parameter. The following 2 lines are equivalent:

        Linux::Joystick->new(device => $dev);

If you want to set the device and specify other parameters at the same time, you'll have to use the full constructor with the device argument.


These are the methods for the Linux::Joystick class itself. Event methods are described in the next section (Events). $js is an object of class Linux::Joystick, in the descriptions below.

List of methods



Returns the version of the Linux::Joystick module. This method may be called as either an instance method (as shown above) or as a class method: my $ver = Linux::Joystick->version;


Returns a joystick event, or undef if there is no event.

Joystick events are Linux::Joystick::Event objects (see below).

In blocking mode (the default), nextEvent waits until there is an event to return. This could mean it waits forever, if the user walks away from the joystick. If you don't like this, either use nonblocking mode or wrap in an eval/alarm block.

nextEvent should never return undef in blocking mode, but you should check for it anyway. I don't know what circumstances could cause it to happen (user unplugs the joystick? Not for USB controllers at least), but it'd definitely count as an exceptional condition. die might be the appropriate response, but I defer that decision to you.

In nonblocking mode (constructor with nonblocking => 1), nextEvent will return undef if there's no pending event ready to be read. This isn't an error or exception: most of the time there's no input. Nonblocking mode is what you'll want to use in all but the simplest applications.


Flushes any pending events in the input buffer. This is most useful in blocking mode, when your program does some long calculation or time consuming I/O. The user might get restless and twiddle the joystick while waiting. Since the kernel joystick buffer is 64 events, this means your program would suddenly read up to 64 random joystick events when its time-consuming subroutine returns, which could cause all kinds of havoc.

This method works in either blocking or non-blocking I/O mode, though it's most useful in blocking mode. Beware: calling flushEvents causes a 0.2 second delay in your program's execution.


Returns the number of buttons on the joystick.

Buttons are numbered starting with 0, so the highest-numbered button will be one less than buttonCount's return value.

For USB joysticks, this count is almost always correct. For gameport joysticks, it's possible that a 2-button generic gamepad/stick will appear to have 4 buttons (I've seen this happen before, but it was a long time ago). It's also possible that you're using a device with the generic gameport joystick driver (which only supports 4 buttons), but that device has a more specific driver you could be using that supports all the buttons on the device. I've had this problem with a Gravis Gamepad Pro gameport controller in the past.

It's possible for a joystick to have 0 buttons, but not very likely (who makes a joystick with no buttons?)

It's also possible for a device to report more buttons than it physically has. I have an `Axis Pad' (manufacturer unknown, made in China) that claims to have 20 buttons, though it really has 11. Strangely enough, 10 of the buttons show up as buttons 0 through 9, and the 11th button (actually a `Game/Set' switch) shows up as number 19!

I've got a gameport to USB adaptor that I use to plug my old Gamepad Pro into my new (USB-only) PC. It supports 4 axes and 8 buttons, and always reports all the buttons and axes, regardless of what kind of gameport controller is plugged in (or not plugged in: the PC can't tell). This part was made by Radio Shack, but I bet other gameport/USB adaptors will exhibit the same behaviour.


Returns the number of axes on the device.

Axes are numbered starting with 0, so the highest-numbered axis will be one less than axisCount's return value.

It's theoretically possible to have a joystick device with no axes (buttons only), but I've never seen one.


Returns the number of `sticks' on the device.

Sticks are numbered starting with 0, so the highest-numbered stick will be one less than stickCount's return value.

This is equal to the number of axes divided by two (rounded down). A stick is equivalent to two axes (vert and horiz), although there's no guarantee that a stick actually represents a physical stick (or d-pad, or whatever): if you have a device with one d-pad, a spinner, and a throttle slider, stickCount will report that you have two sticks (the d-pad counts as one, and the two other single-axis devices together count as the other).

If axisCount is 1 greater than stickCount*2, the leftover axis is a single-axis control. Most single-axis controls are analog, not digital (you can use the axisValue for proportional movement).

The native Linux joystick API has no concept of sticks. I invented this for convenience, because I prefer to think of a d-pad or stick as a single stick, rather than two axes. You are free to treat them as sticks, or axes, or mix and match both forms of addressing.

A few words about joystick axes, sticks, and buttons:

There's no way to tell what axes or buttons correspond to which physical controls on a gamepad or joystick. This is not a limitation of Joystick::Linux, it's a limitation of the underlying kernel API.

That said, there are conventions followed by (almost) all devices.

Even-numbered axes (including 0) are horizontal (left/right) axes. Odd-numbered axes are vertical (up/down) axes. A pair of such axes constitures a `stick'. Even though I call it a stick, it might be a d-pad, or a trakball, or whatever. The important point is that a stick (usually) represents a single physical control that can be used to detect movement.

For d-pads, sticks, hats, trakballs, and the like, there will be 2 sequentially-numbered axes per control. Typically axes 0 and 1 (stick 0) are the primary control (d-pad or stick), 2 and 3 (stick 1) are the hat or analog thumbstick, etc.

If you have single-axis controls (throttles or spinners), they will be the highest-numbered axes, and will only have one axis each. A device with 5 axes might use 0/1 for the main stick control, 2/3 for the hat, and 4 for the throttle slider. Generally, any controller with an odd number of axes has a slider, throttle, knob, or whatever. You are of course free to ignore axes you don't care about (most apps, even games, won't need more than 2 axes (1 stick)).

For buttons, usually the ones directly under the user's thumb will be the lowest-numbered ones (typically these are in a diamond-shaped cluster and labelled A, B, X, Y or A, B, C, D). Usually, but not always, the button numbers returned by $event->button will correspond to the alphabetical ordering of the buttons (button 0 is the A button, 1 is the B, etc.)

`Shoulder' buttons (like the L and R on a SNES controller) will be next (left shoulder having a lower number than the right), and then any pause/select/start buttons.

Turbo buttons are usually implemented in hardware, inside the controller. This means that they don't get their own button numbers. Instead, holding down the `turbo A' button will cause the joystick to send a stream of events (pressed A, released A, pressed A, etc) to the PC. It's impossible for the joystick driver to tell the difference, so the buttonCount method won't include turbo buttons in the count.

(Historical note: Turbo buttons were originally implemented this way because early console games typically didn't have a `rapid fire' mode at all (since it would make a lot of the games really easy). Third party manufacturers would sell joysticks with turbo buttons as `cheating' devices, and they had to work with unmodified consoles and games, hence the transparent hardware implementation).

Not all devices follow the rules.

You need to decide how many buttons and axes you need in your application, keeping in mind that all you can *really* count on are 2 axes and 2 buttons (all PC controllers have at least 2 axes and 2 buttons). These days, it's fine to rely on there being 4 buttons: if anyone still owns a 2-button controller, it should be in a museum.

One trick you can do to semi-support extra buttons/axes is to use the modulus operator:

        # we only have 2 possible actions, only care about 2 buttons
        if($event->isButton) {
                if($event->button % 2 == 0) {
                        # all even-numbered buttons do one action...
                } else {
                        # all odd-numbered buttons do the other action...

This way, the user can use whichever two buttons are most comfortable to him. The same applies to axes: if you only care about up/down/left/right, why not let the user use either the d-pad or the analog thumbstick, his choice?


In the unlikely event of an error reading from the joystick device, this method will give you a human-readable error message. If there was no error, errorSting returns undef.

Currently (in version 0.0.1), no error strings are defined.


Returns the path to the device node that was opened, e.g. /dev/js0, or undef if the device couldn't be opened.


Returns the Perl filehandle that Linux::Joystick is reading events from. You could use this to do a select() on the filehandle (and any other filehandles you need to handle).

Attempting to read from this filehandle will (at best) confuse the joystick driver temporarily, or (at worst) cause your read to block forever (particularly if you're trying to use buffered reads). You have been warned! If you want to use select(), here's one way to do it:

        # assume that $input represents some stream such as a keyboard
        # or network socket, and $js is our Linux::Joystick device. Further
        # assume that $input is opened in non-blocking mode (it shouldn't
        # matter for $js, since the kernel *always* returns 8 bytes per event).

        # adapted from `perldoc -f select', which see for details.

        while(1) {
                my $buf; # $input buffer
                my $BUFLEN = 1024; # size of $input buffer

                my $rin = '';
                for($js->fileHandle, $input) {
                        vec($rin, fileno($_), 1) = 1;

                # 4th parameter is timeout. 0 means return immediately,
                # undef means block forever, anything else is number of
                # seconds to wait. $nfound will tell how many fd's had
                # input pending, which really isn't too useful...

                my $nfound = select($rin, undef, undef, undef);

                if( vec($rin, fileno($js->fileHandle), 1) == 1 ) {
                        my $event = $js->nextEvent();
                        process_event($event); # or whatever

                if( vec($rin, fileno($input), 1) == 1 ) {
                        # GOTCHA: do NOT use <$input> here! (see select() perldoc)
                        while( ($bytes = sysread($input, $buf, $BUFLEN) > 0) ) {
                                # process $input data one buffer at a time

Notice that the above routine doesn't read from the filehandle returned by $js->fileHandle. Instead it's just used in the select() call.

You can make that a lot more readable by using the IO::Select module instead of all that mess with vec() and select().

Of course, variations on this theme are possible. You could use the Curses or IO::Stty modules to read one character at a time from STDIN, in which case you'd just process one keystroke per loop iteration...


Sets or clears non-blocking mode. Takes one scalar parameter, which is treated as a boolean: a true value turns on non-blocking I/O, and a false value turns it off. It doesn't hurt anything to attempt to set the same mode that's already in use, and you can switch between the I/O modes as many times as you want.

setNonblocking uses an fcntl() call to change the file descriptor's mode, so it doesn't close and reopen the device. Remember that when you're in non-blocking mode, all calls to nextEvent immediately return. When there's no input, the event returned will be undef. Also remember not to busy-wait on events! If you find yourself using 99% of the CPU according to `top', you need to restructure your code so that it works in blocking mode. If you can't do this (e.g. because you're reading from a network socket as well as a joystick), at least use a call like select(undef, undef, undef, 0.01) to yield the CPU so that other processes can run. An even better idea would be to stay in blocking mode, but use the fileHandle method to get the joystick's file handle, then select() on both the joystick and network filehandles. This way, the kernel will put your process to sleep until there's some input available on one stream or the other.

Since setNonblocking uses fcntl(), it may behave strangely on really old (2.2 or earlier) kernels. I have only tested this module on Linux 2.4 and 2.6 kernels.




Returns true if this event was caused by a button press, or false if not.

The next 3 method calls are only valid for button events (e.g. when isButton returns true). If called on a non-button event, they will return undef.


Returns the number of the button that caused this event, if it was a button event, or undef it it wasn't a button event. Keep in mind that 0 is a valid button (it's the first button), so you don't want to treat this as a boolean (use isButton instead).

NOTE: In the Linux joystick API, each button is reported separately, even if more than one button was pressed or released simultaneously. For example, pressing 2 buttons at once on a gamepad results in 2 events: one for each button. This may sound like a problem, but in practice it works out just fine if you process each event as soon as it comes in.



Returns true if this was a button press event, false if it was a button release event, or undef if it was not a button event at all.

With the optional $b parameter, returns true if this is a button event and if the button $b was pressed.



Returns true if this was a button release event, false if it was a button press event, or undef if it was not a button event at all.

With the optional $b parameter, returns true if this is a button event and if the button $b was pressed.

For button events, buttonDown returns !buttonUp, and buttonUp returns !buttonDown. Use whichever method makes your code most readable. For non-button events, both methods return undef.


Returns true if this event was caused by joystick axis movement, false otherwise.


Returns the axis number that caused this event, if it was an axis event. Otherwise, returns undef. Remember, 0 is a valid axis number, so don't treat this as a boolean value (use isAxis for that).

NOTE: In the Linux joystick API, each axis is reported separately, even if more than one axis changed simultaneously. For example, diagonal movement on a gamepad results in 2 events: one for the vertical axis and one for the horizontal. This may sound like a problem, but in practice it works out just fine if you process each event as soon as it comes in. However, it does mean that you can't test an individual event to determine whether or not a stick is centered.


Returns the stick (or d-pad, or whatever) number that caused this event. This is determined by which axis caused the movement: Axes 0 and 1 are considered to be stick 1, axes 2 and 3 are stick 2, etc.

More formally, ($ev->stick == $ev->axis >> 1) is always true.

This is the same logical stick number that you provide to the stickUp/Down/Left/Right methods, below.

The Linux joystick API does not include the concept of a stick number; I invented this as a convenience for Perl programmers (to give you More Than One Way To Do It(tm)).


Returns the current value of the axis, if this event was an axis event. Returns undef for non-axis events. 0 is a valid value (it's the center position), so don't use this as a boolean value (use isAxis for that).

This is a signed 16-bit value. Negative values indicate movement to the left (for horizontal axes) or up (for vertical axes). Positive values indicate movement to the right or down. Zero is the center position.

The axisValue method works for either analog or digital controls. In the C API, all joystick devices are treated as analog devices. A digital gamepad will typically return only -32768, 0, or 32767 for each axis, while an analog stick will return values anywhere in the range of -32768 to 32767, with 0 being the center.

You should NOT rely on the values being exact, however: sometimes the calibration is off, so the center value is something other than 0, or the maximum range is less than usual. Typically, you'll want to ignore values less than some threshold (possibly configurable by the user of your app). This keeps `jiter' from affecting your app. If you're having calibration issues, the jscal utility will help.

Actually, you should only use axisValue if your app is using the joystick's analog value to control something like a mouse pointer (other examples would be: the paddles in a Pong/Breakout type game, or the control yoke in a flight simulator). Keep in mind that a digital gamepad-style controller will be useless for such applications. Also keep in mind that the joystick API doesn't give us a way to know whether the joystick we're reading is a digital gamepad or an analog stick.

If you're only interested in which direction the stick is pressed, use the stickLeft/Right/Up/Down methods, below.



Returns true if the event was caused by movement to the left, false if otherwise, and undef if the event isn't an axis event.

If no parameter is provided, a true result means any vertical axis was moved left. If the optional $stick parameter is given, it is used to decide which axis-pair to check for movement. If $stick is 0, axes (0,1) are checked. If $stick is 1, axes(2,3) are checked, etc.

If your app doesn't need more than one pair of axes (one stick), it is recommended that you use the no-argument forms of stickUp, stickDown, stickLeft, and stickRight.



Returns true if the event was caused by movement to the right, false if otherwise, and undef if the event isn't an axis event.

see stickLeft for explanation of the optional $stick parameter



Returns true if the event was caused by upwards movement, false if otherwise, and undef if the event isn't an axis event.

see stickLeft for explanation of the optional $stick parameter



Returns true if the event was caused by downwards movement, false if otherwise, and undef if the event isn't an axis event.

see stickLeft for explanation of the optional $stick parameter

The stickLeft/Right/Up/Down methods are provided as a convenience for applications that only care about which direction the stick was moved, not how far it was moved. This includes digital controls like the d-pad on a gamepad (which can only report all-or-nothing).

These methods are a bit special in that they take into account which axis was moved and whether it increased or decreased. Your app code doesn't need to check e.g. whether the horizontal axis (even numbered) was moved before it checks for left/right movement.

Example use:

        # no-argument forms:
        if($ev->isAxis) {
                move_pacman_left() if $ev->stickLeft;
                move_pacman_right() if $ev->stickRight;
                move_pacman_up() if $ev->stickUp;
                move_pacman_down() if $ev->stickDown;

        # use $stick to test two controls on same device:
        if($ev->isAxis) {
                print "Stick 0 (probably the D-Pad) moved left" if $ev->stickLeft(0);
                print "Stick 1 (probably the analog) moved left" if $ev->stickLeft(1);

The $ev->isAxis test is actually superfluous: all four methods will return undef (a false value) for non-axis events. The snippets above would work just as well (though just slightly slower) without the if() around them.

Notice that there is NO stickCenter method. This is due to the fact that each event only reports movement for one axis. Since this module doesn't save state between events, there's no way to tell (by looking at just one event) the state of both axes. Future versions of Linux::Joystick may address this issue. For now, if you need to detect a centered stick, you'll need to remember the axis values of that stick in your application code.


Returns the string BUTTON for button events, AXIS for axis events, or UNKNOWN for unknown events. Primarily intended for debugging. You shouldn't be comparing this string to determine event type (that's what isAxis and isButton are for).

There should never be any unknown events. If you're getting them, it's because this module has a bug in it, or else the Linux kernel developers have invented a new type of joystick event (not likely to happen any time soon).


Returns a string consisting of hex representations of the raw bytes, as read from the joystick device file descriptor. Only meant for debugging. There is no information here that you can't get from one of the other methods in a friendlier way.


Returns the timestamp of this event. This is an integer number of milliseconds. Linux::Joystick does not use or modify this value; it's the js_event.time field from the C API. The kernel documentation doesn't say a lot about this field. Here's what my copy of joystick-api.txt says:

        The time an event was generated is stored in
        ``js_event.time''. It's a time in milliseconds since ... well,
        since sometime in the past.  This eases the task of detecting
        double clicks, figuring out if movement of axis and button
        presses happened at the same time, and similar.

Use it for whatever you want, or ignore it. Future versions of this Perl module might include support for detecting double-clicks, but if so, it'll be something you have to enable (the default behaviour will not change).

Alternatives to Linux::Joystick

You could always open /dev/js0 yourself and read from it (that's all this module does). I've tried to make the code fairly readable, so you can look at site_per/Linux/ in your perl lib directory and see how it's done (well, one way to do it, anyway). Search for _read_event as a starting point.

You could also use the SDL module from CPAN, which provides lots of other nice stuff besides joystick support, and is portable to lots of other platforms (unlike Linux::Joystick, which only works on Linux).

On the minus side, SDL requires you to create an application window (not necessarily an X11 window) before it can access the joysticks. If you're trying to add joystick support to an existing non-SDL app, or writing a textmode interface with joystick support, or using a joystick as an alternate input for disabled people, or anything else that doesn't benefit from SDL's graphics & sound capabilities, you probably want this module instead of SDL.

There also exist two CPAN modules you can use to support joysticks on Windows platforms, if that's your goal. They are Win32API::Joystick and Win32::MultiMedia::Joystick. I don't know anything about these modules (not being a win32 programmer), and Linux::Joystick is not compatible with either one.

If you're using FreeBSD, NetBSD, or OpenBSD, you may be able to use Linux::Joystick with your OS's Linux emulation package. I don't know whether this is actually supported or not (send me your results and I'll put them here in the next version). A cursory glance at the man page for NetBSD's joy driver shows that it's nothing like the Linux joystick driver, so you'd definitely need Linux emulation there, assuming the emulation emulates the Linux joystick API.


The C API is described in $srcdir/Documentation/input/joystick-api.txt in the Linux kernel source. This Perl module only supports the new (1.2.x) joystick API, not the old 0.x backwards compatibility API. This shouldn't be a problem: I'm writing this in 2004, and the `new' API has been around for something like 8 years now... if you're really still running an early 1.2.x Linux kernel, you presumably know what you're doing and don't need my help.

The C API sends synthetic JS_INIT events, one per axis or button, when the device is first opened. You can ignore these in your perl scripts: Linux::Joystick intercepts the synthetic events itself and counts how many axes/buttons your device has.


Well, I haven't gone on an exhaustive bug-hunt yet, so that counts as one bug right there :)

I haven't tested this with a regular analog stick, because I don't own one. The closest thing I have are the analog thumb-sticks on my Axis pad. All that should need changing with a flight stick is the threshold.

I need to test with lots of different devices. I own maybe 15 or 20 different PC-compatible game pads, so this is just a matter of time.

Someone needs to test this with some of the really oddball controllers out there (like the homebrew hack that lets you plug a Sega Genesis controller into your serial port). Given my hardware skills, that probably will be someone other than me :)

The lack of a stickCenter event (and more generally, the lack of state across events) might count as a bug.

Not really a bug, but a minor shortcoming: instantiating a Linux::Joystick object causes a short (0.2 second) delay.

The errorString method never returns an error string. Most of the time, if there's an error, it's during the constructor (where we open the device for reading), so we return undef and leave the error message in $!. The only possible use for errorString would be if we got an error in the nextEvent method (maybe an EOF), but I've never actually seen any such error.

If events are not read often enough, the kernel joystick driver will fill up its event queue. According to joystick-api.txt, the queue has room for 64 events, and if it overflows, the joystick driver resets (starts sending synthetic init events again). This isn't likely with a well-designed app, but it's possible (e.g. if pressing the 0 button causes a long, involved process that takes a minute or two to complete, and the impatient user keeps wiggling the d-pad while he waits). We should be checking for synthetic events always, not just when we first open the device.

I don't know what happens if a joystick is unplugged while it's open for reading, then plugged back in. On my test machine, unplugging the USB joystick doesn't cause an error, but plugging it back in doesn't bring it back to life. However, unplugging it & plugging it back in even while it's NOT open, makes it disappear and never come back (I have to `rmmod joydev; modprobe joydev' to get it to work). This isn't the normal behviour for USB sticks: normally they can be unplugged and plugged back in, and they'll still work (though I don't know whether user code that's reading them will need to reopen the device or not). I've *no* idea what happens if a gameport joystick is unplugged!


B. Watson,

Feel free to contact me with bug reports, suggestions for improvement, or even success stories (hey, somebody besides me has got to find this thing useful, right?)


You may use and redistribute this Perl module under the same terms as Perl itself (GPL or Artistic License, your choice).

If you use this module in a commercial product, I'd appreciate it if you let me know. This isn't a licensing requirement; it's just common courtesy.

Although I have made every effort to produce bug-free code, I am not responsible for any loss or damages caused by the use of Linux::Joystick. If it breaks, you get to keep both pieces :)


Copyright (c) 2004, B. Watson