Does a pattern exist

[Jan Vereecke]

I try to make a design in Java and am looking for a pattern to solve my
problem. I can best describe this with an example application: a pinball
machine.

A pinball machine consists of different kind of objects (a bumper, a
flipper, a tab, a light, ...) I could derive all these from a common
HitableByBall. I also have two kinds of Balls (RoundBall and
EllipsoidBall), and as more than one ball can be on the board at the same
time, so they could hit each other too, so I should derive Ball also from
HitableByBall.

To implement the pinball machine, I would have to describe the interactions
between a ball and a bumper, a ball and a flipper, ....

In the superclasses, I could write if 'this ball hits any HitableByBall',
call the relevant interaction method. I can use inheritance to make a
distinction between the two kinds of balls, but within some method
this.Interact(HitableByBall h), i would have to fall back to a switch
statement for each kind of hittable object. Is there a better way ?

I think there must be some pattern that defines a separate 'interaction
object' for each combination of HitableByBall types, but then I don't know
how to effectively bind and select the correct type of interaction object
when two HitableByBall come into contact with each other. Did anyone
recognize this type of problem ?

[Simon Strandgaard]

On Tue, 07 Oct 2003 16:29:17 +0000, Jan Vereecke wrote:

> I try to make a design in Java and am looking for a pattern to solve my
> problem. I can best describe this with an example application: a pinball
> machine.
[snip]
> I think there must be some pattern that defines a separate 'interaction
> object' for each combination of HitableByBall types, but then I don't know
> how to effectively bind and select the correct type of interaction object
> when two HitableByBall come into contact with each other. Did anyone
> recognize this type of problem ?

Sound very likely to be MultipleDispatch. Maybe a DoubleDispatch is
sufficiently, if so then the easiest implementation are VisitorPattern.

--
Simon Strandgaard

[Malte Persike]

Hi Jan,

interesting question.


>A pinball machine consists of different kind of objects (a bumper, a
>flipper, a tab, a light, ...) I could derive all these from a common
>HitableByBall.

It should be recognized that HitableByBall would probably end up as a
base class for ALL objects within the play region of the flipper
machine because walls, the ground plate and even the covering glass
pane are hittable objects in the end.


>To implement the pinball machine, I would have to describe the interactions
>between a ball and a bumper, a ball and a flipper, ....

You are raising one of the most crucial questions in game design
there: Who is responsible for regulating physical interactions? In
your pinball paradigma you could take at least three standpoints:

1. The ball hits the bumper, so the ball must alter speed and
direction itself

2. The bumper pushes the ball, so the bumper has to change the ball's
direction and velocity

3. Ball and bumper collide, so some external instance has to decide
how this affects both objects

The first two suggestions pose serious problems in certain situations,
especially when two balls are in the game. With two balls colliding it
is not clear which of the two balls is the "active" element that shall
change the others' parameters (suggestion 2) or which change will
occur first (suggestion 1).
Hence, I always advocate the solution of the "omnipotent third" that
evaluates collisions, adheres to laws of nature (e. g. gravity) and
thereupon administers the changes.

Now the question arises how the omnipotent third gains knowledge of
the collision. This is a crucial problem because none of the elements
within the pinball machine have any idea about the position of any
other object that might be in there as well. To accomplish a design
where each object (or at least each ball) knows where every other
object is located you would have to propagate position messages from
every single object to every other, thereby enabling each object to
detect a collision with another one. But then again, with two balls
colliding, both would recognize a collision and take measures to react
to it. Complete mayhem!

I would thus recommend the following solution:

1. Movement Initiation Phase: A VISITOR iterates through all objects
in the game and tells them to Move().

2. Movement Phase: Every object knows its coordinates, velocity and
direction of motion. It updates its position and does a Notify() on
all OBSERVERs attached to it.

3. Movement Propagation Phase: The responsible Observer issues a
MESSAGE (or COMMAND) to the omnipotent third. The command must contain
the game element the observer got notified by and - for convenience -
the position of that element.

4. Collision Detection Phase: The omnipotent third (a SINGLETON by all
means) evaluates position messages and checks for collisions. If a
collision between two objects is detected, it calls a
CreateVisitor(object1, object2) from a FACTORY.

5. Primary Update Phase: The VISITOR which was just created visits
both of the objects, considers laws of nature and updates velocity and
direction to both objects.

6. Secondary Update Phase: The omnipotent third now might additionally
call an OBSERVER himself which would then trigger the scoring
mechanism, take an element out of the game or else.

7. Step 1 again


You might wonder why phase 2 exists because every element could as
well be self responsible for propagating a position message to the
omnipotent third.
But phase 2 is crucial for events which do occur independently of the
omnipotent third. Suppose you would like to add some twinkling to a
rolling ball. Just attach a second "twinkle-observer" to the ball and
let this observer trigger changes in twinkling.


>I can use inheritance to make a
>distinction between the two kinds of balls, but within some method
>this.Interact(HitableByBall h), i would have to fall back to a switch
>statement for each kind of hittable object.

That fairly complex switching hierarchy is in the above design
decentralized across the FACTORY class (phase 4), where there is one
CreateVisitor()-method for each possible collision. These methods
return specialized VISITORs which exactly know what to do to the two
colliding objects.

HTH,
Malte

---

The above e-mail address is not valid.
To contact me, please use my real
e-mail address:

malte AT t DASH online DOT de

Just replace the capitalized words with
the corresponding punctuation marks.

[Jan Vereecke]

Simon Strandgaard <qj5nd7l02@sneakemail.com> wrote in
newsan.2003.10.07.16.49.00.546017@sneakemail.com:

> MultipleDispatch, DoubleDispatch
could you give a reference to something discussing these patterns ?

[Jan Vereecke]

Thanks for your answer. It will take some time to think it through.
It is definitely a good discussion on how to tackle this particular
problem. I think it's also a good basis for the more general problem
where any two objects of a same baseclass interact in a way that is
dependent of the actual subclass of the two baseclasses. (e.g. chemical
elements, color-mixing, geometrical problems, ...).
I am just wondering: my main 'problem' was that I did not see a way to
get out of the "giant switch statement", which any OO book tells you to
avoid. I believe that your omnipotent third will still contain such a
giant switch ?

> 4. Collision Detection Phase: The omnipotent third (a SINGLETON by all
> means) evaluates position messages and checks for collisions. If a
> collision between two objects is detected, it calls a
> CreateVisitor(object1, object2) from a FACTORY.

To avoid it, I was thinking about a scheme where each object could
contain a list of interaction objects: one for each other type of object
it could interact with. In the initialization phase, these VISITORS would
be attached to the object. When two objects collide, the lists are
compared, and the interaction object that is common to both is called.
This would avoid the switch (at the cost of very expensive comparisons)
and create an omnipotent third that is independent of any future kinds of
objects.
Whether or not this is feasible will be highly dependent on how complex
and varied the interactions are, and especially on the costs of always
comparing lists of interaction objects. (this might still be feasible if
the lists are short or if RTTI and call-by-method-signature is used).

Thanks

[Simon Strandgaard]

On Wed, 08 Oct 2003 13:19:04 +0000, Jan Vereecke wrote:

> Simon Strandgaard <qj5nd7l02@sneakemail.com> wrote in
> newsan.2003.10.07.16.49.00.546017@sneakemail.com:
>
>> MultipleDispatch, DoubleDispatch
> could you give a reference to something discussing these patterns ?

Sorry I don't know any Java resources on this topic (I use C++Ruby).

The ModernC++Design book has a good chapter about this topic.
But it is probably not the easiest place to start getting familiar with a
new pattern.
http://www.moderncppdesign.com/
http://sourceforge.net/projects/loki-lib/

A quick google-search on 'multiple dispatch' yields many result,
but they seems to be: C++, dylan, and in .pdf!!
very low signal2noise ratio. Anyone who know some good links ?

--
Simon Strandgaard

[David Postill]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

In article <Xns940E91A5F9FE3jvab1@193.210.197.139>, on 08 Oct 2003 12:19:04
GMT, Jan Vereecke
<jan.N0SPAMvereecke@advalvasN0SPAM.be> wrote:

| Simon Strandgaard <qj5nd7l02@sneakemail.com> wrote in
| newsan.2003.10.07.16.49.00.546017@sneakemail.com:
|
| > MultipleDispatch, DoubleDispatch
| could you give a reference to something discussing these patterns ?

Learn to use google.

<http://www.google.co.uk/search?hl=en&lr=&ie=UTF-8&oe=utf-8&q=%2Bdouble+dispatch+%2Bpattern+%2Bjava&spell=1>

<http://www.google.co.uk/search?hl=en...gle+Search&met
a=>

<davidp />

- --
David Postill

-----BEGIN PGP SIGNATURE-----
Version: PGP 8.0.2 - not licensed for commercial use: www.pgp.com
Comment: Get key from pgpkeys.mit.edu:11370

iQA/AwUBP4SuXXxp7q1nhFwUEQIkGQCg3VTZMXRAFwahCMVYSkMPi6fuZc0AmwXU
PtUTzKmfw4nj6/OgfzmLJA+Q
=KDaR
-----END PGP SIGNATURE-----

[Malte Persike]

>I am just wondering: my main 'problem' was that I did not see a way to
>get out of the "giant switch statement", which any OO book tells you to
>avoid. I believe that your omnipotent third will still contain such a
>giant switch ?

Nope. No switch statement in my code. The compiler does things for
you.

Imagine the following situation: You have three types of objects that
can interact pairwise. These three object types are

ClassA
ClassB
ClassC

So you end up with the three possible interactions A-B, A-C and B-C.
Now instead of implementing a switch statement that might get
inmanageable in very short time, the following FACTORY comes in handy:

class
InteractionFactory {
ABInteractionClass CreateInteraction(ClassA a, ClassB b);
ACInteractionClass CreateInteraction(ClassA a, ClassC c);
BCInteractionClass CreateInteraction(ClassB b, ClassC c);
};

Sorry for the C++ code but I'm sure you get the idea.

The return objects of type XYInteractionClass are classes that know
what to do to the two objects that have collided. You might call them
interaction managers or so.

Now look at how things are going to work during runtime when two
objects interact:

1. A and B collide
2. Somebody recognizes the collision
3. This somebody now calls "InteractionFactory.CreateInteraction(A,B)"
The return object (an interaction manager) is exactly the one you
want since the compiler selects it for you based on the types of the
two objects.
5. The interaction takes place and on it goes

See, no switch statement!


>I was thinking about a scheme where each object could
>contain a list of interaction objects:

Well, this would surely solve your problem in many cases but two
issues can arise with that approach.

First, suppose objects A and B come together for an interaction to
occur. Which of the two is now responsible for starting the
interaction? And who controls the interaction process. Here the
MULTIPLEDISPATCH pattern really has its merits, like Simon
recommended.

Second, what if the interaction was moderated by external conditions?
Suppose, for example, a zero-gravity bonus in the pinball machine or a
happy-hour rebate in a billing system.
You would then have to update every single element in your system with
a new list of interaction objects every time the external condition
changes.
And to complicate things even further, remember the PacMan game with
those tiny pills that could make PacMan invulnerable for encounters
with the evil haunters. In a multiplayer situation you would have
situations where the interactions of one PacMan were totally different
from those of the other players' PacMans.
I would implement this not with a conditional statement in the
interaction object but instead with a new set of interaction objects
for that single PacMan.

My recommendation thus remains to take the interaction objects out of
the elements and install them at another level.

Regards,
Malte

---

The above e-mail address is not valid.
To contact me, please use my real
e-mail address:

malte AT t DASH online DOT de

Just replace the capitalized words with
the corresponding punctuation marks.

[Daniel Biesiada]

> Sorry I don't know any Java resources on this topic (I use C++Ruby).

What has Ruby other modern most used languages dont have.
I mean in relation to the opportunity Ruby "can give the industry"?
I ask because few secs of browsing at its site shown me another yet free
scripting language derriving syntax from other langs offering nothing new
to me after those few secs gaze which didnt inspire me to look more
for details. Any good points to go back and take more serious look at?

greetings,
Daniel Biesiada

[Simon Strandgaard]

"Daniel Biesiada" <spam@no.thank.you> skrev i en meddelelse
news:bmm6n9$r0t$1@atlantis.news.tpi.pl...
> > Sorry I don't know any Java resources on this topic (I use C++Ruby).
>
> What has Ruby other modern most used languages dont have.
> I mean in relation to the opportunity Ruby "can give the industry"?
> I ask because few secs of browsing at its site shown me another yet free
> scripting language derriving syntax from other langs offering nothing new
> to me after those few secs gaze which didnt inspire me to look more
> for details. Any good points to go back and take more serious look at?


Citing Ruby's author Matz; "The purpose of Ruby is to maximize programming
pleasure."

I think also Ruby's purpose is to minimizing the programmers stress level.
These 2 things is what Ruby has, what no other languages has... prove me
wrong :-)

Before I knew Ruby, I wrote pseudocode before implementing things in C++.
That code were actually 98% correct Ruby code. Unbelievable..but true.
Besides that Ruby is pure obejct oriented, meaning that everything is an
object
which you can inspect or extend (Neither C++, Java, Python is true oo).

There is an interesting slideshow presentation here:
http://www.ruby-lang.org/en/jaoo2001/index.html

Great introduction + reference:
http://www.rubycentral.com/book/

There also exists a number of physical books:
http://www.rubygarden.org/ruby?RubyBookList


In my sparetime I work on a programmers editor written in Ruby:
http://aeditor.rubyforge.org/

--
Simon Strandgaard