Are there any common guidelines?
Currently i just try to give each class a single task and ask myself wether a class really should know about x or y.
For example, im trying to make an octree class which is currently like:
Octree (handles allocating the nodes and finding their children)
Node (carries the node data + data for the octree to find the children etc.)
OctreeIterator (not sure if its an iterator, but anyways, it references a node and allows you to traverse the tree by calling stuff from octree)
but it could have very well been
Octree (has root node and functions to find some specific node)
Node (has functions to go to the child/parent nodes and traverse the tree)
or something like that o,e
How to design good classes?
A good starting point is the SOLID object oriented principles from "objectmentor":
Author
Thanks, any help with my octree thing?
I would like to have the iterator be a single class so it can be used anywhere (i think it wouldnt work to use it elsewhere as a pointer and use polymorphism, as it needs to be cloned to be used properly i think and i dont think that works well with polymorphism without dynamic allocation of the thing)
So the iterator would have a reference to an OctreeBase and an OctreeNodeBase
Now, im not sure how to get that work, because to get the next node i have 2 options i think:
1.Call virtual OctreeNodeBase.GetChildNode(stuff)
*Doesnt it add overhead to have it virtual?
*The node needs to find the child node, but the place theyre at is maintained by the octree, which it doesnt have because the iterator only has a pointer to an OctreeBase
2.Call virtual OctreeBase.GetChildNode(OctreeNodeBase*,stuff)
*The octree needs data from the node to find the child, which doesnt work with a pointer to the base of the node
3.Do something haxy-ish that requires me to cast random pointers into types. I dont want to do that if theres another way.
So is it possible to get 1 or 2 work, do i need to use 3 or is my overall design flawed?
I would like to have the iterator be a single class so it can be used anywhere (i think it wouldnt work to use it elsewhere as a pointer and use polymorphism, as it needs to be cloned to be used properly i think and i dont think that works well with polymorphism without dynamic allocation of the thing)
So the iterator would have a reference to an OctreeBase and an OctreeNodeBase
Now, im not sure how to get that work, because to get the next node i have 2 options i think:
1.Call virtual OctreeNodeBase.GetChildNode(stuff)
*Doesnt it add overhead to have it virtual?
*The node needs to find the child node, but the place theyre at is maintained by the octree, which it doesnt have because the iterator only has a pointer to an OctreeBase
2.Call virtual OctreeBase.GetChildNode(OctreeNodeBase*,stuff)
*The octree needs data from the node to find the child, which doesnt work with a pointer to the base of the node
3.Do something haxy-ish that requires me to cast random pointers into types. I dont want to do that if theres another way.
So is it possible to get 1 or 2 work, do i need to use 3 or is my overall design flawed?
Author
I think this might work:
OctreeBase
OctreeNode<AllocationDataType>
Iterator<AllocationDataType>
AllocationDataType
Finding node:
Iterator<AllocationDataType> ->
Iterator GetChild(x,y,z)
{
Octree ->
Node<AllocationDataType>& GetChild(Node<AllocationDataType> currentnode,x,y,z)
{
AData=node.GetAllocData()
*code to find child of node*
return childnode
}
}
OctreeBase
OctreeNode<AllocationDataType>
Iterator<AllocationDataType>
AllocationDataType
Finding node:
Iterator<AllocationDataType> ->
Iterator GetChild(x,y,z)
{
Octree ->
Node<AllocationDataType>& GetChild(Node<AllocationDataType> currentnode,x,y,z)
{
AData=node.GetAllocData()
*code to find child of node*
return childnode
}
}
In your original post you suggest that an alternative might be:
Octree (has root node and functions to find some specific node)
Node (has functions to go to the child/parent nodes and traverse the tree)
This seems more sensible to me.
Unless you have a need for an iterator class then I wouldn't create one at this time.
The iterator pattern allows client classes to traverse a data structure without needing to know the type of data structure being traversed.
In your case you're using a custom iterator class with the method getChild(x,y,z) which publicly exposes the fact that it's an octree. So unless you're planning on using multiple octree implementations, and then hiding which implementation is in use from its clients, then adding the new iterator doesn't seem to add anything at this time.
This comes back to the mantra keep it simple; only add the iterator when a need for it arises.
As you've mentioned Iterators I'm guessing you've come across Design Patterns.
Design Patterns are very powerful but it seems (from my limited experience, I'm certainly no expert) that it's more important to remember what the patterns are trying to achieve, e.g. reduce repetition / redundancy, reduce dependencies and coupling, reduce complexity etc and to remember the principles of OO that rip-off identified.
There are many guidelines on when to create classes. The book "Code Complete" by Steve McConnell contains a discussion on this subject as well as many other subjects and is worth a buying.
Hope this helps.
Good luck!
Matt
Octree (has root node and functions to find some specific node)
Node (has functions to go to the child/parent nodes and traverse the tree)
This seems more sensible to me.
Unless you have a need for an iterator class then I wouldn't create one at this time.
The iterator pattern allows client classes to traverse a data structure without needing to know the type of data structure being traversed.
In your case you're using a custom iterator class with the method getChild(x,y,z) which publicly exposes the fact that it's an octree. So unless you're planning on using multiple octree implementations, and then hiding which implementation is in use from its clients, then adding the new iterator doesn't seem to add anything at this time.
This comes back to the mantra keep it simple; only add the iterator when a need for it arises.
As you've mentioned Iterators I'm guessing you've come across Design Patterns.
Design Patterns are very powerful but it seems (from my limited experience, I'm certainly no expert) that it's more important to remember what the patterns are trying to achieve, e.g. reduce repetition / redundancy, reduce dependencies and coupling, reduce complexity etc and to remember the principles of OO that rip-off identified.
There are many guidelines on when to create classes. The book "Code Complete" by Steve McConnell contains a discussion on this subject as well as many other subjects and is worth a buying.
Hope this helps.
Good luck!
Matt
It's not so much about designing "good" classes, as it is about representing functionality and relationships in a cohesive and decoupled fashion. I would recommend that you read up on those two principles as a starting point and try to think about them any time you begin building some software.
Cohesion: http://en.wikipedia.org/wiki/Cohesion_(computer_science)
Coupling: http://en.wikipedia.org/wiki/Coupling_(computer_programming)
Also you may have heard the term "abstraction". A good software engineer can learn to identify abstractions and their relationships and represent them clearly.
There is a reason why designers don't call cup holders in your car holes. That is an example of abstraction.
Cohesion: http://en.wikipedia.org/wiki/Cohesion_(computer_science)
Coupling: http://en.wikipedia.org/wiki/Coupling_(computer_programming)
Also you may have heard the term "abstraction". A good software engineer can learn to identify abstractions and their relationships and represent them clearly.
There is a reason why designers don't call cup holders in your car holes. That is an example of abstraction.
If you are asking which one is better, then certainly use iterators. If possible, use the same iterators that comes with the programming language. If you use C++, properly inherit from std::iterator. If you use Java, implements java.util.Iterator.
Author
The iterator is needed because i want to maintain the current location of the iterator in the octree (depth,x,y,z) so if i want to move to another location and it happens to be the adjacent node, i dont have to traverse the whole octree from the top. I think iterators are good when you can find the next element faster by starting from the previous location which probably is right next to it. Or something like that.
I dont think i can use the language iterators in this case as it isnt traversed in a single dimension :c
I dont think i can use the language iterators in this case as it isnt traversed in a single dimension :c
http://en.wikipedia.org/wiki/Iterator_pattern
Iterators are an abstraction. They are a way of iterating and accessing various elements in a data structure without having to couple your code to that data structure's particular implementation. You are right (to some degree) that you can't use your languages iterators to traverse your octree but you can use your language's common iterator interface to make your octree conform to the API (IEnumerator iand IEnumerable and their generic equivalents in C# are examples of this pattern).
Iterators are an abstraction. They are a way of iterating and accessing various elements in a data structure without having to couple your code to that data structure's particular implementation. You are right (to some degree) that you can't use your languages iterators to traverse your octree but you can use your language's common iterator interface to make your octree conform to the API (IEnumerator iand IEnumerable and their generic equivalents in C# are examples of this pattern).
Code Complete Second Edition.
Huge book; you'll probably know a lot about good code design by the end of it - I haven't finished mine, yet.
Huge book; you'll probably know a lot about good code design by the end of it - I haven't finished mine, yet.
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
