This article is currently only a proposal

The features or design guidelines described in this article are only a proposal made by one or some persons. It has not been evaluated or accepted by the core development team yet. Feel free to add your personal opinion about them or make counter proposals.

Each animation will be defined by an xml document, here follows an example:

<sprite name="player">
 <imageset name="base" src="player-male-base.png" width="64" height="64" />

 <animation name="stand" direction="down">
  <frame imageset="base" index="0" />
 </animation>
 <animation name="stand" direction="left">
  <frame imageset="base" index="18" />
 </animation>
 <animation name="stand" direction="up">
  <frame imageset="base" index="36" />
 </animation>
 <animation name="stand" direction="right">
  <frame imageset="base" index="54" />
 </animation>

 <animation name="walk" direction="down">
  <frames imageset="base" start="1" end="6" delay="10" />
 </animation>
 ...

 <animation name="sit" direction="down">
  <frame imageset="base" index="7" />
 </animation>
 ...

 <animation name="bow-attack" direction="down">
  <frames imageset="base" start="13" end="17" delays="5,5,5,10,50" />
 </animation>
 ...

 <animation name="dead">
  <random>
   <frame imageset="base" index="8" />
   <frame imageset="base" index="26" />
   <frame imageset="base" index="44" />
   <frame imageset="base" index="62" />
  </random>
 </animation>
</sprite>

So if you want to load the playerset you just load player.xml and it takes care of loading all related images. Of course delays are defined in milliseconds.

Specifications

<sprite>

A sprite is an object which can carry several animations, hence I call the root element the sprite. Also, I think we should seperately define animations in several directions. For now this example is based on the current playerset image, but by using multiple imagesets this could of course be split up.

<animation>

Defines an animation sequence that should be displayed when the sprite object is in a specific state (attribute "name") and in a specific direction (attribute "direction").

possible names:

       attack-bow
            |
      attack-stab  sleeping
            |         |
run   attack-swing   sit
 |          |         |
walk     attack      dead
   \        |        /
          stand
            |
       (nameless)

this table also specifies the substitution preference when a specific name is requested but it is not defined. the next lower name is used instead.

possible directions are:

up
down
left
right

when no direction attribute is provided, this animation is used for all missing directions with the same name.

<frame>

Every <animation> has one or more frame child elements. each of them defines one frame of the animation. the required attributes are imageset and ether index or start and end.

imageset defines the spriteset file from which the frame should be taken and index defines the number of the sprite in this spriteset. when a start and a end tag is provided instead of an index tag the frame tag does not only define one animation phase but multiple phases which all use the same imageset and delay and where the sprite indices in the spriteset are from "start" to "end".

an optional attribute is delay. every frame is displayed for delay millisecounds and then replaced by the next frame in sequence. when the last frame is reached the sequence starts with the first frame again. when no delay is specified the animation doesn't continue when the frame is reached.

Each frame and frames element can optionally have attributes x and y to specify an offset from the default drawing position for that frame. This will allow the animation of for example the hairset (or any equipment) to reuse the same frames with different offsets, and be defined exactly like below.

Suggested additions

multiple delays in a <frame> with start and end attribute

in the bow animation I've chosen a shortcut to specify different delays for different frames. I am undecided yet whether this is a nice feature or whether it'll be better to require multiple frame elements in such cases.

<random> childtag for <animation>

Also, it includes an experimental suggestion on specifying that a random frame should be chosen.

sequences based on default sequences

After reflecting on this idea (and the above suggestion) I think a first issue to solve is to reduce the amount of duplicated definitions. For example, at the moment every monster uses the exact same frame size and animation frames. Hence, it would be interesting if the defined sprite could take the actual imageset it uses as a parameter instead of specifying this on its own. This way we can define a monster like:

<sprite name="monster-walking-default">
 <imageset name="base" />
 ...
</sprite>

<being name="Scorpion">
 <sprite ref="monster-walking-default">
  <with-imageset name="base" src="scorpion-base.png" width="60" height="60" />
 </sprite>
</being>

<being name="Red Scorpion">
 <sprite ref="monster-walking-default">
  <with-imageset name="base" src="scorpion-red-base.png" width="60" height="60" />
 </sprite>
</being>

Defining an empty imageset (without src attribute) will make the engine require this imageset to be passed as a parameter whenever the spriteset is referenced.

I'm not convinced yet that the name attribute of the sprite element is really necessary. I think I would prefer each sprite to be defined in its own file. However it could be attractive to define multiple things in the same file, but in that case I think we will have to read everything when the client is launched, and only graphics and imagesets can be created lazily (only when necessary). Because the other data consists of just numbers and strings, this should not be really a problem.

An implementation idea by Peoro

An idea to manage the animations should be a class based on a timer which choose automatically the current frame. Who needs animations (players, monsters and npcs) has only to start the animanion and to request the frame to draw it. Each instance of Animation class keep its own start-time (or a personal timer), its total time, number of frames, and the time of every frame. The Animation class has to provide at least a function to start the animation and another one to get the current frame; other usefull functions should pause and restart the animation and allow to jump to a random frame.

Here a simple example of a prototype of an Animation class: (I mutilated my original Animation class to simplify and summarize it cutting a lot of other functions and vars and using the most possible of standard items)

class Animation
{
    int nframes;
    Img **frames;
    int *times;
    int totalTime;
    int startTime;
    
public:
    Animation ( int n, Img **i, int *t )
    {
        int c;
    	
        nframes = n;
        frames = new Img* [n];
        times = new int [n];
        totalTime = 0;
        
        for ( c = 0; c < n; c ++ ) {
            frames[c] = i[c];
            times[c] = t[c];
            totalTime += t[c];
        }
    }
    
    inline void start ( int t = clock() )
    {
        startTime = t;
    }
    
    Img *getFrame ( int t = clock() )
    {
        int i;
        int it
        
        if ( ! totalTime ) {
            t = 0;
        } else {
            t = (t-startTime) % totalTime;
        }
            
        // searches the current frame
        for ( i = 0, it = 0; (it + times[i]) < t; i ++, it += times[i] );
        
        return frames[i];
    }
	
};

And here how to use this class:

Img * imgs [] = { img1, img2, img3 };
int times [] = { 20, 20, 30 };

Animation a ( 3, imgs, times );

a.start ( );

while ( 1 ) {
    draw ( a.getFrame() );
}

An advantage of this class is that the animation speed is indipendent to FPS and it's very easy to use.

This is a simple way to manage a lot of animations for who needs more than one:

class AnimationsList
{
    // ltstr is only a const char* comparison function
    map<const char*, Animation*, ltstr> animations;
    int startTime;       // now Animation::startTime is no more usefull
    Animation *currentAnimation;
    
public:
    AnimationsList ( ) : currentAnimation(0) { }
    
    inline void addAnimation ( const char *n, Animation *a )
    {
        animations[n] = a;
    }
    
    int setAnimation ( const char *n, int t = clock() )
    {
	map<const char*, Animation*, ltstr>::iterator ai;
        Animation *a;
        
	ai = animations.find ( n );
	
	if ( ai == animations.end() ) {
                // error... animation not found
		return 0;
	}
	a = ai->second;
        
        if ( currentAnimation != a ) {
            a->start ( );
            currentAnimation = a;
            startTime = t;
	}
        
        return 1;
    }
    
    Img * getFrame ( )
    {
        if ( currentAnimation ) {
            return currentAnimation->getFrame ( );
        }
        return 0;
    }
};

Using a multimap instead the map it's possible to keep more than an animation bind to the same string. This would be a good way to make AnimationsList choose a random animation.

And here how to use AnimationsList class:

AnimationsList al;

al.addAnimation ( "walking" "south", anim1 );
al.addAnimation ( "walking" "north", anim2 );
al.addAnimation ( "anim3", anim3 );

al.setAnimation ( "walking" "south" );

while ( 1 ) {
    draw ( al.getFrame() );
}

I hope not to had make a mistake writing... Anyway these are only theorical bases.

Implementation proposal by Doener

This proposal is incomplete regarding how new sequences are added to the Animation (which is not the right name for this type of class *g*)... Anyway, you should get the idea... The Frame class encapsulates images and are used to create a single linked list that creates an animation. The idea behind the linked list is that you can easily create loops, by having the last element of the animation pointing to the first one. Also, you can easily prepend a "starting animation" before the loop, like in an animation of a running player. He starts out with one animation and then loops the real running... The Animation class is responsible for mapping animation names to their starting frame and for disposing the encapsulating Frame objects, thus a std::map and a std::list is used to keep track of those objects. The next frame is simply calculated from the time that has passed since the last run of the logic method. The case that a whole animation loop can be skipped is not taken into consideration, because a) we cannot do this with frames linked together arbitrarily and b) animations will probably take around 100ms or longer, ie. we'd have less than 10fps for that case to happen and then there's nothing left to gain by that simple optimization anymore anyway ;)

For the linked list features, of course the xml format needs to be adjusted to support that.

/*
 * Frame class, stored information about frames in an animation
 */
struct Frame
{
	Image *mImg;  // Image for this frame
	int mDelay;   // Time in ms for how long this frame is shown
	Frame *mNext; // Next frame in sequence
};

class Animation
{
	public:
		Animation():mCurrentFrame(0);
		~Animation();

		void setSequence(const std::string &);

		void logic(int timePassed);

		Image* getCurrentImage()
			{ return mCurrentFrame ? mCurrentFrame->mImg : 0; }

	private:
		Frame *mCurrentFrame;
		typedef std::map<std::string,Frame*> FrameMap;
		typedef FrameMap::iterator FrameMapIterator;
		FrameMap mStartFrames;
		std::list<Frame*> mFrames;
		int mRemain;
}

Animation::~Animation()
{
	for_each(mFrames.begin(), mFrames.end(), make_dtor(mFrames));
}

void Animation::setSequence(const std::string &name)
{
	FrameMapIterator i = mStartFrames.find(name);
	mCurrentFrame = (i == mStartFrames.end()) ? 0 : i->second;
	mRemain = 0;
}

void Animation::logic(int timePassed)
{
	mRemain += timePassed;
	while (mRemain >= mCurrentFrame->mDelay) {
		mRemain -= mCurrentFrame->mDelay;
		mCurrentFrame = mCurrentFrame->mNext;
	}
};