Jump to content

  • Log In with Google      Sign In   
  • Create Account


Management of game objects


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
6 replies to this topic

#1 chiranjivi   Members   -  Reputation: 181

Like
0Likes
Like

Posted 23 March 2014 - 12:08 PM

Hi all,

 

I'm trying to implement a sensible management system for all the objects in a basic platform game. There are various types of objects - projectiles, the player, enemies, powerups, moving platforms, that type of thing.

 

Initially I thought for the sake of simplicity I'd have all objects derive from a base class Entity, and then stick them all in a vector<Entity*>. Then when it's time for them to act or to be drawn to the screen, I can just make a single pass over the vector and call it->update() or it->display(). This saves a bunch of code duplication from having them all in separate vectors for each type and having to iterate over each vector separately to call display functions and so on.

 

Difficulties arise with this method when I need to do something like collision-test all Projectiles against all Enemies - I have to iterate over the vector, testing if each element is a Projectile, and if it is, iterate over the vector again testing to see if each element is an Enemy, and then testing for a collision if so. I have to do this a bunch of times for, say, enemies vs projectiles, player vs projectiles, player vs powerups, enemies vs player, and it turns into a huge tangled mess of code and nested iterations over the vector, and the vast majority of element comparisons don't need to be made as the entities are of the wrong type.

 

I'm looking for an elegant way to simplify this, so that Type A vs Type B checks are as simple as possible, and also so that mass updating and displaying of objects is equally simple.

 

Is there an accepted best practice for this kind of thing? Thanks in advance for any replies.



Sponsor:

#2 SiCrane   Moderators   -  Reputation: 9388

Like
5Likes
Like

Posted 23 March 2014 - 12:38 PM

If you're using pointers, there isn't anything that stops you from having both a master Entity * container and separate containers for specific types. You can iterate over the master list when you want to batch process everything and over the type containers when you just want projectiles or enemies.



#3 ProtectedMode   Members   -  Reputation: 1079

Like
5Likes
Like

Posted 23 March 2014 - 12:47 PM

Maybe what you're searching for is a collision bitmask. This is usually just an array of bools. You will still iterate from every entity to all others, but before you test for collisions, you check if one of the bools is matching the other one. You can use one bit to indicate "I want to check collisions with projectiles", for example. There are also a lot of other ways to speed up collision checking, like using quad trees (just google them).

 

You also need to be aware that you need to manually delete all entities when you remove them from the vector, else it will cause memory leaks. A better solution would be to use:

// Includes
#include <vector>
#include <memory>

// Somewhere else.
std::vector<std::shared_ptr<Entity>> entities; // The entity will be automatically deleted once the object is not being referenced anymore.


#4 chiranjivi   Members   -  Reputation: 181

Like
0Likes
Like

Posted 23 March 2014 - 01:46 PM

Thanks both for your replies.

 

If you're using pointers, there isn't anything that stops you from having both a master Entity * container and separate containers for specific types. You can iterate over the master list when you want to batch process everything and over the type containers when you just want projectiles or enemies.

 

How would I handle deletion of objects in this scenario? Say we update everything in the master list and determine that an enemy's life has reached zero, so we erase it from the vector. Isn't there now a pointer in the Enemies type container which points to an unused memory location? Apologies if there is a simple answer to this that I am not seeing.



#5 ProtectedMode   Members   -  Reputation: 1079

Like
0Likes
Like

Posted 23 March 2014 - 02:06 PM

Thanks both for your replies.

 

If you're using pointers, there isn't anything that stops you from having both a master Entity * container and separate containers for specific types. You can iterate over the master list when you want to batch process everything and over the type containers when you just want projectiles or enemies.

 

How would I handle deletion of objects in this scenario? Say we update everything in the master list and determine that an enemy's life has reached zero, so we erase it from the vector. Isn't there now a pointer in the Enemies type container which points to an unused memory location? Apologies if there is a simple answer to this that I am not seeing.

That's indeed a bad thing about that implementation. But usually you would mark the object as deleted. Then you loop through all object-specific containers, and if they're marked for deletion, you delete them from that vector and the general vector.



#6 Satharis   Members   -  Reputation: 949

Like
0Likes
Like

Posted 23 March 2014 - 10:33 PM

Best practice isn't really a thing, usually the method you use will just be a combination of what works the most simply in your case vs what costs it has. For instance in a simple platform game you could probably flat out get away with iterating the vector multiple times checking a set of objects for collisions against other appropriate objects by identifying them somehow, what SiCrane suggested is also a possibility, just dumping more pointers into friendly containers can help you do things more quickly.

 

Usually how "elegant" something is really means a greater time investment so I would prioritize doing what you think will work and not cause any major hangups over the most elegant solution.

 

Keep in mind the methods change a lot based on design as well, if you're doing something 3d for example then you might have more tools to split up entities into smaller chunks to check against each other like octrees or different spatial partitioning.



#7 Krohm   Crossbones+   -  Reputation: 2961

Like
0Likes
Like

Posted 23 March 2014 - 11:50 PM

I think it is worth considering what your logic will look like in the future.

In the future, for the specific collision example, you're not going to do collision yourself but rather use a library. This library won't care about your objects as it will have its own.

So we somehow need some more flexibility.

Inheritance itself is arguably enough flexible to enable this kind of flexibility dynamically using some tricks and state management. Main issue with extensive inheritance is the inevitable diamond problem but definition of entities is another one which is often understated. Consider Enemy. The difference with Ally is subtle at best.

You might have heard about this "component model". It basically splits your basic entities and considers them a set of properties - just like a color can be split in RGBA components - the properties build up the behavior. The properties themselves are managed independently but their state changes are coalesced inside each single entity.

So, for the specific case of collision testing, you don't check for Entity collisions, because nobody says an entity is "collidable". Instead, each solid entity request creation of a collision rigidbody, which is then tested as usual. You then look at the collision list and propagate the result of collision detection to the Entity for proper state management.

It seems like looking at the set of properties and figure out a constant method of doing things to be a very widespread way of doing this. Personally I don't see how this improves the situation compared to a "big object" approach. I use callbacks and I find them a considerably better alternative.






Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS