Pausing

Last time, I talked about how triggers can be used to replace traditional step events in order to solve certain esoteric problems in Game Maker. This time I want to explain how to use triggers to achieve something a lot more universally useful: pausing the game.

Game Maker, sadly, has no good built-in methods for pausing that are quite satisfactory (in fact, the documentation doesn't even mention it, except of course the references to the debugging feature). Over the years, I've experimented with the following solutions, but found each of them lacking:

Solution: Using the keyboard_wait() function

Problems: It freezes everything, preventing any sort of pause menu functionality. It also breaks the flow of games designed for controllers or the mouse, because it requires interaction with the keyboard.

Solution: Going to a dedicated Pause room

Problems: This is much better, because you can put whatever you want in the pause room to create any kind of effect you want (menus, faded image, soft music, etc). But it's kind of a pain to juggle the room_persistent status of the calling room, and if you want to show a frozen image of the game, you have to create the resource yourself using something like background_create_from_screen(). Also, if you want access to any of the objects from the calling room, you're sort of out of luck.

Solution: Using instance_deactivate_all()

Problems: This solution uses an object for controlling all pause behaviour that deactivates everything else and takes over until the game is unpaused. It has most of the same problems as the previous solution, though (no access to paused objects; the need to manually create an image resource from the screen), and introduces yet another: Unpausing requires you to call something like instance_activate_all(); if there are any objects you don't want to be reactivated, you're out of luck. Basically, this solution is incompatible with any game that has to activate/deactivate instances for any other purpose.

After all this, one starts to wonder why there isn't just some kind of built-in functionality for pausing in Game Maker. Almost all games need the feature (besides notable exceptions like Phantasy Star Online) so why not? There could be a variable called pause_object which would point to the only object which would remain active while the game was paused (much like how view_object points to the object which the view should follow). And there'd be a function called game_pause() that would take one argument - true or false - to toggle the state on and off. Would this have been so hard?

Fortunately, with a little bit of work, triggers come to our rescue in this situation, too.

Because trigger events can be set to occur before each step moment (beginning, middle, and end), it's possible to create triggers to completely replace the step events:

(If you can't see the image, the condition code for these triggers should be return global.paused;).

If you move the actions in the step events on all your objects to the corresponding trigger events instead, you can prevent the execution of these actions merely by setting global.paused to true. Any object you don't wish to be paused should retain traditional step events, of course.

Using this method, the draw event will still be executed, preventing the need to generate an image. And all paused objects are still active in case you need to access them. Great!

One quickly notices the drawbacks to this solution, though. Beginning, middle, and end step events aren't the only things that happen for objects during each step in Game Maker. There are other events, like keyboard checking and so on, that are also performed.

This doesn't bother me personally, because I tend to ignore these events and write my own code for checking keys and stuff in the step event of some kind of control object. This isn't for everyone, though, so is there some way to get around this?

Well, yes. You can create more custom triggers that check the same conditions as the original, but also check global.paused. For example, return global.paused and mouse_check_button_pressed(mb_left);. This is a bunch of busy work, but do it at the start of your project and it might save you some trouble down the line.

Another drawback to using triggers to pause is that Game Maker handles certain things about objects automatically. Every step, it adds hspeed to x and vspeed to y, and so on. Again, this doesn't affect me because I use all my own variables, but it can be gotten around anyway: in your code that pauses the game, you could just do this...

...and of course the reverse when you unpause.

In conclusion, I admit that this solution is far from perfect also, and it requires a lot of work to get functioning flawlessly. But depending on what your game is like, and considering the issues with competing solutions, it might be useful to you.

If there are better solutions, I'd love to hear about them.

Trigger Happy

When Game Maker 8.0 introduced triggers, I can't say that I was all that impressed with the feature, and I've hardly used it since. This is because, as a matter of personal preference, I tend to eschew such shortcuts in favour of handling things manually; it's far easier for me to debug code when I'm personally responsible for each line, rather than goofing around with automatic functions whose technical nature is hidden in order to make things more palatable to beginners. Sometimes I forget that Game Maker even has things like room transitions and timelines, since I habitually code custom, flexible alternatives for each of my projects.

Recently, though, I've discovered two great uses for triggers that solve problems that are otherwise either very difficult to overcome or flat-out insuperable. I'll talk about the first one of these in this post, and the second, next time.

Code Execution Order

Games (and pretty much all programs) have what's commonly called a "main program loop" which controls all the behaviour in the game. A basic platformer might have one that looks something like this:

1. Check for press of the pause key to toggle pause on and off
2. If paused, branch directly to (8) drawing the screen
3. Handle player input
4. Move all objects
5. Check for collisions
6. Update the camera
7. Handle animations
8. Draw the screen

Obviously, the above is rather simplified; for instance, there's no mention of handling the game's audio. But you get the idea.

Having started experimenting with programming long before helpful game creation suites with user-friendly GUIs like Game Maker came on the scene, I'm very used to starting a project by writing the main loop. So used to it, in fact, that when I first used Game Maker, with its event system, I found it very counter-intuitive and confusing. Buried in the help file one can find a brief explanation of the event order, but otherwise there's very little explanation given of when and how your code is being run. Without going through the effort to manually test the system and gain a detailed understanding of it, one can easily be led to make games with egregious timing issues; for example, the player juttering uncomfortably while riding moving platforms.

In the above example loop, steps 4, 5, 7, and 8 are actually loops in and of themselves. To move all objects, for instance, one must use some kind of loop (for, while, or whichever of their kin is most suitable) that moves each object in turn (usually by adding the speed variable to the position variable).

Since the code for each object must be performed sequentially in some sort order, it's natural to wonder what that order happens to be. In a classic game like Sonic the Hedghog, it is the order in which the objects are actually stored in RAM. In Game Maker, it is the order in which the objects and instances were created.

Because this, especially the object/instance distinction, can be confusing, I will explain it in greater detail.

"Object" is the natural word one wants to use when discussing "things" in a video game: the player, pickups, enemies, platforms, etc. In fact, I can't think of a single non-clumsy alternative. Game Maker, though - in one of its myriad maddening quirks - has a different, technical meaning for "object" which is much closer to something like "class" or "template". The layman's object is referred to in GM parlance as an "instance". The player never interacts with objects, only ever instances of objects.

But if you're reading this, you're probably already passingly familiar with GM's system. Anyway, every instance has an "id" (in the id variable) as well as an "object index" (in the object_index variable). The id is a unique identifying number by which the instance can be referred to. Every time an instance is made, it is given an id number that is 1 greater than the last instance created. Barring some horrible bug in GM (or gap in my knowledge), it's impossible to have two instances with the same id, or to create an instance with an id that is lesser than an existing instance's id. (Note that objects placed in GM's room editor have fixed ids; if you restart a room their ids remain constant.)

So every instance has a unique id, creating a perfect sequence that mirrors the order in which they were created. (Yes, there will be gaps introduced when objects are destroyed, but this hardly matters.) Is this the order in which their code is executed? In short, no.

If it were, one would expect the object index to be irrevelant, causing the following pseudo-code program:

to output the string "ABAB". However, in reality, it will generate the string "AABB". What this means is that all instances of an object are performed in id sequence, and then all instances of the next object, and so on. The order of objects is determined by their object index, which works much like instance ids: the first object created for a game (using the resource tree in GM itself or by using the object_add() function in a running game) will be 0, the second will be 1, and so on.

Knowing this, we can predict issues that many games will encounter. Imagine the player is object 0 and a moving platform is object 26. The player will do its motion code, during which it will align to the platform instance that it's on (if it happens to be on one) -> Then the platform will do its motion code -> Then the screen will be drawn. What will be seen when playing the game is the player always lagging one step behind the platform, because it can't catch up until the next step.

(Quick readers will notice that you can easily solve this problem by simply having the platform move the player if they're on it, as opposed to letting the player take care of their own motion, but this is not always an optimal solution depending on how you've set up your physics. For instance, if the player does take care of their own motion, it's sometimes possible to optimise by removing collision checks.)

It's certainly things like this that explain why GM has the "Begin Step" and "End Step" events. Every game creation engine I've tried out has some form of this; it gives the user the ability to make absolutely sure that certain code is done before/after other code. It works for 99% of situations, and was probably easier on the creators of the program than giving the user full control over the main loop (which might confuse beginners).

But what about that other 1% of the time? Very advanced users might be out of luck, if it weren't for triggers (you knew I had to get back to the actual subject of the post eventually, right?).

Using Triggers to Solve Order Issues

First I'll need to paint a picture of a specific order issue so that I can explain how triggers can be employed in solving it:

It's fun to be able to put cool visual effects in your game, like motion blur, lighting bloom, or ripple distortion effects when underwater. But these are incredibly difficult, if at all possible, to achieve in GM without using surfaces. By drawing the game view to a surface, and then drawing that surface to the screen, the possibilities for effects are endless. Once the view has been rendered to a surface, it can be handled like any other image resource, manipulated in a myriad of ways - even used as a texture!

(Rendering to a small surface and then drawing the surface upscaled without any texture interpolation is a great way to make low-res classic-style games á la Cave Story without that nasty blur that GM normally inflicts upon you. The only other ways to do this aren't really viable - storing all graphics as 2x not only takes up more memory and CPU power, but it's only the graphics - the physics won't be "low-res" without major tweaking.)

The simplest way to convert a game to use a surface instead of drawing directly to the screen is to a) switch off automatic drawing at game start; and b) in a control object's end step event, set the drawing target to a surface, call screen_redraw(), reset the target to the screen, and draw the surface with whatever effects are desired (make sure to call screen_refresh() afterward, or the frame the player sees will always be one frame out-of-date!)

But wait... that all sounds great, but there's a hidden problem lurking amongst those otherwise reasonable instructions: "In a control object's end step event..." According to what we've learnt above, unless the control object is the last object you ever created in the game (which is almost the exact opposite of what will be the case in the vast majority of situations), some objects won't have executed their code when the screen is rendered to the surface! Normally in GM, the screen is drawn last thing, full stop. But our brilliant surface method has just borked that, pushing the drawing back to the end step event. If any object's appearance or position changes in its end step event, then it will drawn out-of-sync, which is unacceptable (to all but the development team of Sonic Genesis).

Well, what's needed (but doesn't seem to exist) is an "End End (No, really this time!) Step Event" that can be used by the control object for handling the drawing instead. If one were lucky, there might be some kind of event type (like "outside of boundary" or some such) that's handled after the end step. One might, in a hackish way, abuse it for such a purpose. However, this is not the case. This is the event order:

• Begin step events
• Alarm events
• Keyboard, Key press, and Key release events
• Mouse events
• Normal step events
• (now all instances are set to their new positions)
• Collision events
• End step events
• Draw events

There's nothing after end step that can be conveniently repurposed.

Here (finally) is where triggers come to our rescue. Take a look at the trigger window:

(When using Windows 98, one sometimes resorts to festive colour schemes for the sake of entertainment. =P)

The magic we're about to make is all because of that little phrase toward the bottom: "moment of checking". Trigger events can be made to happen at each of GM's step events. And if "at" seems a little vague, let me amend that - it's actually directly before. (If multiple triggers are set to the same moment of checking, they'll be evaluated in the order they were created, but they'll all still happen before the associated step event.)

Well, you might be thinking, that's great - we can add new events before the end step. Good going, genius - we needed one for after.

But here's the thing: it's entirely possible to swap out the end step event of all your objects for a trigger event (let's go ahead and call it "pre-end step") - well, all except for the control object, which will use the end step to draw the surface. A large amount of clicks, surely (depending on the size of your game), but a small price to pay for full control over the order in which things happen. By the way, the pre-end step trigger should have return true; as its condition code, so that it happens no matter what.

And I'm sure a clever Code Ninja like you can imagine ways, through the use of many triggers, to beat GM at its own game, essentially injecting all sorts of custom step events directly into the main program loop. (Imagine a trigger whose condition was a variable called exit_step; you could effectively make a custom step event which could be exited at any point!) With power like that you could - dare I say it? - take over the world! =P

Jump Height Calculator

Hello again, Code Ninjas!

Today is something very simple: a formula that allows you to plug in variables for "jump force" and "gravity" and get the maximum height of a jump as the result.

When designing platformers, tweaking the physics until they're just right is very important. Having a formula such as this will be useful for zeroing in on exactly what values constants like gravity should have.

Another use might be this: say you already have established physics, for instance because you're hacking a Sonic game. But you want to add a new object, such as a new kind of bumper or spring, that bounces Sonic exactly 5 times his height. Determining the force at which Sonic should be impelled by the object in order to achieve that height would be a snap using the following formula.

The simplest way to write the formula is this:

g is your gravity, and f is the jump force, which can be set to anything. In a Genesis Sonic game, for example, gravity is 0.21875pps (pixels per step) and the jump force is 6.5pps.

h is the height value we are trying to find out - the number of pixels the character will travel given the jump force f with gravity g. t is time, which is optional - it will be the number of steps it takes to reach the apex of the jump. Both h and t are initialised at 0.

Then we run a loop while f is greater than 0. In the loop, first f is added to h, then g is subtracted from f. This simulates the jump: the force acts upon the character's position, then the gravity acts upon the force. t is then incremented by 1 in order to count the time.

When the loop is finished, h will be the ultimate height of the jump.

Note that this code assumes your physics to handle similarly to Sonic's. In Sonic, because of the particulars of the original code, the jump force is added once to the character's position before gravity acts upon it. If this is not the case in your game, then the loop should be restructured accordingly; g should be subtracted from f before f is added to h.

Earlier I said "the simplest way" to write the formula. The truth is that this, while simple, is a rather brute force method. Depending on the strength of the jump force and gravity, the loop could run dozens of times. Yes, modern computers can handle this without breaking a sweat, and yes, you probably won't even be using the formula in a running game, anyway. But for the sake of mathematical beauty, we can find a better way that doesn't employ a loop.

So let's build this new formula piece by piece. First, we need to find how many steps it will take for gravity to whittle the jump force away to 0 (or less). This will be time t again.

We find t by dividing f by g and rounding up to the nearest 1. Why the rounding up? Well, if the jump force isn't perfectly divisible by gravity, the remainder would still count as upward velocity and the player would still move up by a little bit during that step. Since it counts toward the total, it must be taken into account.

Now that we know how long the jump will take to reach the apex, we can easily discover the distance the character would travel during that time, without gravity, merely by multiplying f by t.

Of course this seems a little silly, because we're trying to find the height covered with gravity taken into account. But knowing this value is useful; 'cos if we can also determine how much force is deducted by gravity from the jump force over t steps, we can multiply gravity by that number, subtract it from h, and have our correct result.

In the first step (t = 0), the jump force is unaffected. In the next, it is lesser by gravity. In the next, it is lesser by gravity again, i.e. it is equal to the initial jump force value minus gravity times 2. Next step, times 3, then in the next, times 4, and so on, until gravity overcomes the jump force in step t minus 1.

Visually represented, you might think of the amount of force lost to gravity as a triangular stack like this:

----- (t=0)
g---- (t=1)
gg--- (t=2)
ggg-- (t=3)
gggg- (t=4)
ggggg (t=5)
...

Fortunately it is easy to find the area of a triangle such as this by finding the area of a square the size of t times t-1 and then cutting that value in half. (If, as mentioned above, your physics subtract gravity before the character moves once, the square should have a size of t times t+1 instead).

Et voilà, you have the correct height of the jump, identical to the result of the while statement method used above.

Finally, because multiplication is a commutative process, the formula can be recast in a simpler way for our final code:

You can find a small example .gmk here that lets you play with the variables to get different jump heights. Until next time, happy coding!

Partially Erasing Surfaces in Game Maker

Hey there, Code Ninjas!

One of the cool things about Game Maker 8 is the ability to export PNGs with an alpha channel for transparency. There's no separate functions for doing so, though; if you want to be sure the exported image is in PNG format, you have to make sure the extension is explicitly ".png", like so:

It's especially cool to create partially transparent surfaces and export them. After creating a surface, you can use the draw_clear_alpha() function to make it completely transparent:

One thing that's annoying, though, is the way that drawing with a partial alpha to a surface works. Instead of blending with the colour underneath, the colour is completely replaced, alpha and all. Effectively this punches "holes" in the surface image.

You'd think this could exploited to create some kind of eraser tool. Draw pixels with an alpha of 0 to the surface to erase pixels that are already there, leaving fully transparent pixels behind.

This doesn't work, though, for some reason. Very low alpha values such as 0 or 0.01 function exactly as you'd normally expect when drawing to the screen, even though higher values such as 0.7 differ when using surfaces.

So much for the ability to erase pixels from a surface using that method... But there is another way.

Set the blend mode to bm_subtract before drawing to the surface and you'll effectively be able to erase from the image:

This trick may come in handy on occasion, especially when making games where the user is allowed to paint custom textures for things - an erase tool is essential.

Until we meet again, happy coding!

Looking Up and Down

Hello again, Code Ninjas!

In most 2D platformers the player has the ability to shift the camera a short way by holding up or down on the D-pad. In Sonic games, it works great but there is minor flaw that's hard to notice and really doesn't cause any problems, but I thought it would nice to show how to fix it anyway.

The Problem

The problem arises when the player looks up or down near the top or bottom boundaries of the level (or whatever current boundaries the camera is limited to). In order to understand the problem, we need to look at the basic idea behind shifting the camera.

If there were no ability to look up or down, the process would be really simple. The camera would simply follow the player's position directly, with a simple check to make sure it doesn't leave the level boundaries. In order to shift the camera up and down, though, an extra step is needed. An offset is added to the player's Y position before the camera follows it. By increasing or decreasing the offset value when the up or down buttons are pressed, the camera will shift vertically relative to the player's position.

Clearly there need to be maximum and minimum values that the offset can't exceed, otherwise the player could continue to scroll the camera freely until it reached the top or bottom of the level. Normally this behaviour is not desired, so limits are set that will keep the player visible on the screen.

Imagine for a moment, though, that there aren't. Say the player is looking up - the offset decreases and decreases. It will continue to decrease even after the camera stops at the top of the level, because the offset doesn't know when to quit. The camera is limited, sure, and the player never sees beyond the boundary of the level, but the offset value is still decreasing away.

Now what would happen if the player stops pressing up? The offset value will start increasing until it returns to 0 (no shift at all). But depending on how long you've been pressing up, you'll have to wait a few moments before the camera starts to visibly scroll back to the neutral position. After all, the offset has been invisibly counting away for an undetermined amount of time, and that extra time has to be made up for when it tries to return.

This example with no upper/lower limit for the offset illustrates the problem I'm talking about, and that we're going to fix. It's less noticeable when limits are present for the offset, because the value can't continue to increase or decrease indefinitely, but it's still there. If you've got a copy of any Genesis Sonic game, try looking up or down near the top or bottom of a zone and see for yourself. Scandalous, isn't it?

It may not be the worst problem in the world, but it can be fixed, so let's give it a go.

The Solution

There are actually two solutions. One would be to reduce the offset by the appropriate amount when the player lets up from pressing up or down, but that's not the solution I'll describe. Why not? Because that method would require the detection of when up or down is released, which - while certainly possible - is harder to slot right into the way the code already works in the original Sonic.

Let's look at some basic code for handling the offset, and then we'll apply the fix to it.

The problem occurs in case -1 and case 1: stopping at the shiftLimit isn't good enough, because at the top and bottom of the level, we need to stop increasing or decreasing early.

How close to the top or bottom of the level must one be in order for the undesired behaviour to occur? Close enough that the distance between the top/bottom of the camera and the top/bottom of the level is less than the shiftLimit.

This suggests the solution. Instead of using the shiftLimit alone, we should use whichever happens to lesser - the shiftLimit or the difference between the view boundary and the level boundary.

(The reason why shiftOffset has to be subtracted from view_yview is so the difference won't change once the screen starts scrolling and the offset starts to change.)

And that's all that's needed. It may not be much, but persnickety folks might enjoy the game more!

If you enjoyed this and the previous Code of the Ninja, be sure to come back tomorrow for one more before I slip back into the shadows. =P

Checking Multiple Joypad Buttons

Welcome back, Code Ninjas!

I apologise for not posting in such a long time (I know ninjas are supposed to be silent, but not that silent), but I've been working on a couple of projects that I hope will soon suprise and delight.

Right now, though, I want to talk about an improvement to my earlier Joypad code. I've been interacting a lot recently with disassemblies of Sonic the Hedgehog, and it's a great learning experience. Regardless of what might have happened to the Sonic series over the years, Yuji Naka's programming remains an inspiration to me. Studying his code has taught me plenty of little tricks, not least because the Genesis is very limited by today's standards and it took a lot of skill to squeeze great results out of it.

Anyway, last time I had described a system that updates a variable called JoyCurrent each step with the current state of the joypad, with each bit representing one button. There was also a second variable called JoyPrevious in which JoyCurrent is stored right before JoyCurrent gets updated. And finally, a third variable called JoyPast, only there to smooth out problems with cheap joypads that occasionally glitch up.

There were three scripts: joy() for checking if a button is down; joy_pressed() for checking if a button is down now but not one step ago; and joy_released() for checking if a button is up now but not one step ago (or even two steps ago, in the case of the aforementioned glitchy controllers).

But I've since discovered a serious deficiency with the joy_pressed() and joy_released() scripts. They can't check for more than one button at a time without causing problems. Let me explain.

The Problem

Since JoyCurrent and its related variables contain bits that are either on or off to represent the state of buttons on the joypad, checking the state of a button is as easy as testing any given bit with code like this:

where argument0 is a power of two such as 1 (for testing the first bit), 2 (for testing the second bit), 4 (for testing the third bit), and so on. It's best to define these values as constants so that they can be sensibly named after the buttons, like A, B, LEFT, or START.

Anyone paying enough attention can see that you can test for more than one button simultaneously just by passing a value as argument0 that has more than one bit on. For instance, you could pass a value like 65535 to test if any button was down, or you could use binary OR to test any combination such as LEFT | RIGHT.

Now, the way the script was written it will return true if any bit of argument0 matches up with one in JoyCurrent. If you want to be sure all the bits match, then you'd have to write something like this:

This is all very well and good, but it falls apart when we get to joy_pressed(). This is how it was written:

Now suddenly, because of that boolean and, the value being returned is degraded - it's now only useful as true or false and doesn't give us as much information. Worse, the following happens:

Imagine you want to check whether A, B, or C are pressed, like in the old Sonic games where any of the three buttons makes him jump or Spin Dash. You don't care if one of the buttons is already down when another is pressed - you still want to detect the new press. The way my code was written, this is impossible with only one call to joy_pressed() because if any of the bits is on in JoyPrevious, the new press won't be detected. The only solution would be to make multiple calls something like this:

which is just tacky and consumes more processor time. It would be far better to be able to type:

and have it be done with. (Of course, A | B | C could be a constant called JUMPBUTTON or something, too, to make it even nicer.)

Well, then, how can we change the code so that this is possible? I'm glad you asked that.

The Solution

At the end of the script joy_step() (the one that updates JoyCurrent and JoyPrevious), we need to update two new variables, JoyPressed and JoyReleased (not to be confused with the scripts that have similar names!) These should be global variables, declared in joy_init().

These variables are destined to behave just like JoyCurrent, only for pressed and released. Just like how you can test to see if buttons are down by checking JoyCurrent as joy() does:

you'll be able to check which buttons are newly down or up in one simple comparison by rewriting joy_pressed() like so:

and joy_released() like so:

(At this point these scripts are all so simple you might not even want to make them scripts at all, but merely type anywhere you would have typed joy_pressed(BUTTON), but it's up to you.)

This sounds great, and it will solve all of the problems I mentioned above, but I haven't told you yet how to update JoyPressed and JoyReleased at the end of joy_step(). It requires a little bit of explanation, though, so we can understand the underlying principles. Otherwise, it would get confusing and complex if you ever need to expand upon it.

First, let's look at a visual representation of our variables. I'm assuming only 8 buttons for convenience. Here's a state with no buttons down:

JoyPrevious: - - - - - - - -
JoyCurrent:  - - - - - - - -

Let's press the first button (we'll call it A).

JoyPrevious: - - - - - - - -
JoyCurrent:  - - - - - - - A

Now let's, without advancing a step yet, add a third variable to this visual guide, temp. It's contents will be JoyPrevious binary AND-ed with JoyCurrent (i.e. ).

JoyPrevious: - - - - - - - -
JoyCurrent:  - - - - - - - A
temp:        - - - - - - - -

As far as temp is concerned, nothing has happened! But what happens when we do advance one step, without letting go of A?

JoyPrevious: - - - - - - - A
JoyCurrent:  - - - - - - - A
temp:        - - - - - - - A

JoyPrevious becomes JoyCurrent, JoyCurrent remains the same, and temp finally notices what's going on. Clearly, temp is no good for checking buttons that are newly down, because for one temp has only detected the new press one step late, and for two if we continue to hold A temp will not revert to 0. Merely using binary AND isn't enough. We need to do one more calculation - binary XOR. Let's go back to our previous step:

JoyPrevious: - - - - - - - -
JoyCurrent:  - - - - - - - A
temp:        - - - - - - - -

and add a fourth variable, called JoyPressed. It's contents will be temp binary XOR-ed with JoyCurrent (i.e. JoyPressed = temp ^ JoyCurrent;).

JoyPrevious: - - - - - - - -
JoyCurrent:  - - - - - - - A
temp:        - - - - - - - -
JoyPressed:  - - - - - - - A

By binary XOR-ing temp and JoyCurrent, JoyPressed contains only bits that are different between them. In the next step, the magic happens:

JoyPrevious: - - - - - - - A
JoyCurrent:  - - - - - - - A
temp:        - - - - - - - A
JoyPressed:  - - - - - - - -

Now JoyPressed has reverted to 0, meaning it accurately represents buttons pressed - bits will only trigger for one frame when their corresponding button is pressed. The same thing will happen even if A is released instead of held down:

JoyPrevious: - - - - - - - A
JoyCurrent:  - - - - - - - -
temp:        - - - - - - - -
JoyPressed:  - - - - - - - -

And, if a new button is pressed while another is held down, it will still be detected as a new press:

JoyPrevious: - - - - - - - A
JoyCurrent:  - - - - - - B A
temp:        - - - - - - - A
JoyPressed:  - - - - - - B -

Fantastic! Let's add another variable, JoyReleased, that is temp binary XOR-ed with JoyPrevious instead of JoyCurrent (i.e. JoyReleased = temp ^ JoyPrevious;) and advance one step while releasing B (but not A).

JoyPrevious: - - - - - - B A
JoyCurrent:  - - - - - - - A
temp:        - - - - - - - A
JoyPressed:  - - - - - - - -
JoyReleased: - - - - - - B -

The same principle operates as with JoyPressed. We just solved the problem. Hooray! The actual code at the end of joy_step() would look something like this:

Really the only thing to be done now is make sure that cheap joypads don't cause false press and release events simply because the signal is interrupted for a step once in a while. This is easily done by binary OR-ing JoyPrevious and JoyPast together to create a sort of "buffered" previous state when checking for presses, and binary OR-ing JoyCurrent and JoyPrevious together for a buffered current state (and using JoyPast in place of JoyPrevious where it used to appear in the line) when checking for releases. For example:

Conceivably you could also, instead of doing everything in 2 lines, store more information like so:

This way you could check JoyDown or JoyUp to see whether a button is down but not pressed, or up but not released, which might be useful. Hey, you never know.

That takes care of today's subject. I'll be posting again soon. Until then, happy coding!