Pure functional tree with parent pointer - clojure

I know that RB tree with left and right child can be implemented in pure functional way without degrading log n performance. Can tree with parent pointer be implemented in logarithm time? Seems like cyclic reference child->parent and parent->child requires all tree to be cloned, thus linear time.

Fully persistent, purely functional data structures are tree shaped, but may utilize pointer sharing to become a directed acyclic graph. However, once you introduce a pointer cycle, you can't "change" part of that subgraph without copying that entire subgraph.
The solution is to add an indirection: You assign "identities" to values, which can be objects (like Clojure's atoms) or simple values used as look up keys (such as numbers or symbols). You can think of a immutable pointer as an implementation of IDeref which always returns the same object. A cyclic graph can be represented as an adjacency graph, where "deref-ing" a node by name is the same as looking it up in a the map of names to nodes.
For more information on representing fully persistent graphs, see Fully Persistent Graphs - Which One to Choose?.

Related

Traverse through a DAG-like structure to produce another DAG-like structure in Clojure

I have a DAG-like structure that is essentially a deeply-nested map. The maps in this structure can have common values, so the overall structure is not a tree but a direct acyclic graph. I'll refer to this structure as a DAG for brevity.
The nodes in this graph are of different but finite number of categories. Each category can have its own structure/keywords/number-of-children. There is one unique node that is the source of this DAG, meaning from this node we can reach all nodes in the DAG.
The task is to traverse through the DAG from the source node, and convert each node to another one or more nodes in a new constructed graph. I'll give an example for illustration.
The graph in the upper half is the input one. The lower half is the one after transformation. For simplicity, the transformation is only done on node A where it is split into node 1 and A1. The children of node A are also reallocated.
What I have tried (or in mind):
Write a function to convert one object for different types. Inside this function, recursively call itself to convert each of its children. This method suffers from the problem that data are immutable. The nodes in the transformed graph cannot be changed randomly to add children. To overcome this, I need to wrap every node in a ref/atom/agent.
Do a topological sort on the original graph. Then convert the nodes in the reversed order, i.e., bottom-up. This method requires a extra traverse of the graph but at least the data need not to be mutable. Regarding the topological sort algorithm, I'm considering DFS-based method as stated in the wiki page, which does not require the knowledge of the full graph nor a node's parents.
My question is:
Is there any other approaches you might consider, possibly more elegant/efficient/idiomatic?
I'm more in favour of the second method, is there any flaws or potential problems?
Thanks!
EDIT: On a second thought, a topological sorting is not necessary. The transformation can be done in the post-order traversal already.
This looks like a perfect application of Zippers. They have all the capabilities you described as needed and can produce the edited 'new' DAG. There are also a number of libraries that ease the search and replace capability using predicate threads.
I've used zippers when working with OWL ontologies defined in nested vector or map trees.
Another option would be to take a look at Walkers although I've found these a bit more tedious to use.

How to do per node caching in a tree visitor

I have an application where a want to calculate different representations (mesh, voxelization, signed distance function, ...) of a tree of primitives (leaf nodes) that are combined via boolean operations (inner nodes).
My first approach to this was to write an abstract base class with a virtual getter function for each of the different representations and cached the intermediate results at the respective nodes as long as there was no change in their subtree (which would flush their cache).
However, I was unsatisfied with the ugly coupling of the tree structure with each of the different representations. To alleviate this I removed the abstract base classes and instead set up a visitor for each of the representations.
This neatly decoupled the tree from the representations but left me with the problem that I now need to cache the intermediate results somewhere else and this is where my problem starts.
TL;DR
How do I cache (arbitrary many differently typed) intermediate values at inner nodes of the tree without making the tree dependent on the value type?
My Approaches
The requirements offer two choices:
store the data in the tree but with type erasure
store the data outside the tree and somehow "connect" it to the node
The first one leaves me puzzled with some efficiency problem: I could easily add a container of boost::any (or something equivalent) in the nodes but then each visitor would have to search the whole container for it's own data.
The separation in the second one introduces the problem of keeping the cache up to date to the current tree. If there are changes in the tree (deletions, alterations of nodes) the cached values must at least be invalidated. My intuition was to use some hash function and an unordered_map but I hit some problems there as well:
I cannot use the treenodes themselves as key, so I need to introduce another class that just references tree nodes and represents them in the tree
referencing the values from the unordered_map's keys requires to erase all entries whose referencees are deleted or we have a dangling reference(/pointer) in the unordered_map which could get triggered on rehash
changes in the tree would require to reconstruct the unordered_map because keys might have changed
Am I missing some obvious solution to this?
Which approach would you favor (and why)?
I once had a similar problem and my solution was as follows:
Let each node have an unique identifier.
Let each node have a version number. Modifications that invalidate calculated values for the node just increase the version number.
Let each visitor have a caching map, where the ID pair is the key, mapped to a version/value pair.
When (re-)walking the tree, look for a node's entry in the map. If the version is correct, use the cached value. If it is outdated, calculate the new value and replace the old version/value pair.
At first, I used the node's address as IDs, but for memory reasons I had to reuse subtrees and picked the path to the node as ID. Such a path has the advantage that it can be calculated by each visitor and need not be stored at the node. In my case, each node could have at most two children, so a path was merely a set of left/right decisions, which can be stored in a simple unsigned int with some bit-shifting (my trees did never reach a depth of 32, so a 32 bit unsigned was more than enough as key).

Is there any MFC / STL class that represents Binary Tree [duplicate]

Why does the C++ STL not provide any "tree" containers, and what's the best thing to use instead?
I want to store a hierarchy of objects as a tree, rather than use a tree as a performance enhancement...
There are two reasons you could want to use a tree:
You want to mirror the problem using a tree-like structure:
For this we have boost graph library
Or you want a container that has tree like access characteristics
For this we have
std::map (and std::multimap)
std::set (and std::multiset)
Basically the characteristics of these two containers is such that they practically have to be implemented using trees (though this is not actually a requirement).
See also this question:
C tree Implementation
Probably for the same reason that there is no tree container in boost. There are many ways to implement such a container, and there is no good way to satisfy everyone who would use it.
Some issues to consider:
Are the number of children for a node fixed or variable?
How much overhead per node? - ie, do you need parent pointers, sibling pointers, etc.
What algorithms to provide? - different iterators, search algorithms, etc.
In the end, the problem ends up being that a tree container that would be useful enough to everyone, would be too heavyweight to satisfy most of the people using it. If you are looking for something powerful, Boost Graph Library is essentially a superset of what a tree library could be used for.
Here are some other generic tree implementations:
Kasper Peeters' tree.hh
Adobe's forest
core::tree
"I want to store a hierarchy of objects as a tree"
C++11 has come and gone and they still didn't see a need to provide a std::tree, although the idea did come up (see here). Maybe the reason they haven't added this is that it is trivially easy to build your own on top of the existing containers. For example...
template< typename T >
struct tree_node
{
T t;
std::vector<tree_node> children;
};
A simple traversal would use recursion...
template< typename T >
void tree_node<T>::walk_depth_first() const
{
cout<<t;
for ( auto & n: children ) n.walk_depth_first();
}
If you want to maintain a hierarchy and you want it to work with STL algorithms, then things may get complicated. You could build your own iterators and achieve some compatibility, however many of the algorithms simply don't make any sense for a hierarchy (anything that changes the order of a range, for example). Even defining a range within a hierarchy could be a messy business.
The STL's philosophy is that you choose a container based on guarantees and not based on how the container is implemented. For example, your choice of container may be based on a need for fast lookups. For all you care, the container may be implemented as a unidirectional list -- as long as searching is very fast you'd be happy. That's because you're not touching the internals anyhow, you're using iterators or member functions for the access. Your code is not bound to how the container is implemented but to how fast it is, or whether it has a fixed and defined ordering, or whether it is efficient on space, and so on.
If you are looking for a RB-tree implementation, then stl_tree.h might be appropriate for you too.
the std::map is based on a red black tree. You can also use other containers to help you implement your own types of trees.
The problem is that there is no one-size-fits-all solution. Moreover, there is not even a one-size-fits-all interface for a tree. That is, it is not even clear which methods such a tree data structure should provide and it is not even clear what a tree is.
This explains why there is no STL support on this: The STL is for data structures that most people need, where basically everyone agrees on what a sensible interface and an efficient implementation is. For trees, such a thing just doesn't exist.
The gory details
If want to understand further what the problem is, read on. Otherwise, the paragraph above already should be sufficent to answer your question.
I said that there is not even a common interface. You might disagree, since you have one application in mind, but if you think further about it, you will see that there are countless possible operations on trees. You can either have a data structure that enables most of them efficiently, but therefore is more complex overall and has overhead for that complexity, or you have more simple data structure that only allows basic operations but these as quick as possible.
If you want the complete story, check out my paper on the topic. There you will find possible interface, asymptotic complexities on different implementations, and a general description of the problem and also related work with more possible implementations.
What is a tree?
It already starts with what you consider to be a tree:
Rooted or unrooted: most programmers want rooted, most mathematicians want unrooted. (If you wonder what unrooted is: A - B - C is a tree where either A, B, or C could be the root. A rooted tree defines which one is. An unrooted doesn't)
Single root/connected or multi root/disconnected (tree or forest)
Is sibling order relevant? If no, then can the tree structure internally reorder children on updates? If so, iteration order among siblings is no longer defined. But for most trees, sibiling order is actually not meaningful, and allowing the data structure to reorder children on update is very beneficial for some updates.
Really just a tree, or also allow DAG edges (sounds weird, but many people who initially want a tree eventually want a DAG)
Labeled or unlabled? Do you need to store any data per node, or is it only the tree structure you're interested in (the latter can be stored very succinctly)
Query operations
After we have figured out what we define to be a tree, we should define query operations: Basic operations might be "navigate to children, navigate to parent", but there are way more possible operations, e.g.:
Navigate to next/prev sibling: Even most people would consider this a pretty basic operation, it is actually almost impossible if you only have a parent pointer or a children array. So this already shows you that you might need a totally different implementation based on what operations you need.
Navigate in pre/post order
Subtree size: the number of (transitive) descendants of the current node (possibly in O(1) or O(log n), i.e., don't just enumerate them all to count)
the height of the tree in the current node. That is, the longest path from this node to any leave node. Again, in less than O(n).
Given two nodes, find the least common ancestor of the node (with O(1) memory consumption)
How many nodes are between node A and node B in a pre-/post-order traversal? (less than O(n) runtime)
I emphasized that the interesting thing here is whether these methods can be performed better than O(n), because just enumerating the whole tree is always an option. Depending on your application, it might be absolutely crucial that some operations are faster than O(n), or you might not care at all. Again, you will need vastely different data structures depending on your needs here.
Update operations
Until now, I only talked about query opertions. But now to updates. Again, there are various ways in which a tree could be updated. Depending on which you need, you need a more or less sophisticated data structure:
leaf updates (easy): Delete or add a leaf node
inner node updates (harder): Move or delete move an inner node, making its children the children
of its parent
subtree updates (harder): Move or delete a subtree rooted in a node
To just give you some intuition: If you store a child array and your sibling order is important, even deleting a leaf can be O(n) as all siblings behind it have to be shifted in the child array of its parent. If you instead only have a parent pointer, leaf deletion is trivially O(1). If you don't care about sibiling order, it is also O(1) for the child array, as you can simply replace the gap with the last sibling in the array. This is just one example where different data structures will give you quite different update capabilities.
Moving a whole subtree is again trivially O(1) in case of a parent pointer, but can be O(n) if you have a data structure storing all nodes e.g., in pre-order.
Then, there are orthogonal considerations like which iterators stay valid if you perform updates. Some data structures need to invalidate all iterators in the whole tree, even if you insert a new leaf. Others only invalidate iterators in the part of the tree that is altered. Others keep all iterators (except the ones for deleted nodes) valid.
Space considerations
Tree structures can be very succinct. Roughly two bits per node are enough if you need to save on space (e.g., DFUDS or LOUDS, see this explanation to get the gist). But of course, naively, even a parent pointer is already 64 bits. Once you opt for a nicely-navigable structure, you might rather require 20 bytes per node.
With a lot of sophisication, one can also build a data structure that only takes some bits per entry, can be updated efficiently, and still enables all query operations asymptotically fast, but this is a beast of a structure that is highly complex. I once gave a practical course where I had grad students implement this paper. Some of them were able to implement it in 6 weeks (!), others failed. And while the structure has great asymptotics, its complexity makes it have quite some overhead for very simple operations.
Again, no one-size-fits-all.
Conclusion
I worked 5 years on finding the best data structure to represent a tree, and even though I came up with some and there is quite some related work, my conclusion was that there is not one. Depending on the use case, a highly sophsticated data struture will be outperformed by a simple parent pointer. Even defining the interface for a tree is hard. I tried defining one in my paper, but I have to acknowledge that there are various use cases where the interface I defined is too narrow or too large. So I doubt that this will ever end up in STL, as there are just too many tuning knobs.
In a way, std::map is a tree (it is required to have the same performance characteristics as a balanced binary tree) but it doesn't expose other tree functionality. The likely reasoning behind not including a real tree data structure was probably just a matter of not including everything in the stl. The stl can be looked as a framework to use in implementing your own algorithms and data structures.
In general, if there's a basic library functionality that you want, that's not in the stl, the fix is to look at BOOST.
Otherwise, there's a bunch of libraries out there, depending on the needs of your tree.
All STL container are externally represented as "sequences" with one iteration mechanism.
Trees don't follow this idiom.
I think there are several reasons why there are no STL trees. Primarily Trees are a form of recursive data structure which, like a container (list, vector, set), has very different fine structure which makes the correct choices tricky. They are also very easy to construct in basic form using the STL.
A finite rooted tree can be thought of as a container which has a value or payload, say an instance of a class A and, a possibly empty collection of rooted (sub) trees; trees with empty collection of subtrees are thought of as leaves.
template<class A>
struct unordered_tree : std::set<unordered_tree>, A
{};
template<class A>
struct b_tree : std::vector<b_tree>, A
{};
template<class A>
struct planar_tree : std::list<planar_tree>, A
{};
One has to think a little about iterator design etc. and which product and co-product operations one allows to define and be efficient between trees - and the original STL has to be well written - so that the empty set, vector or list container is really empty of any payload in the default case.
Trees play an essential role in many mathematical structures (see the classical papers of Butcher, Grossman and Larsen; also the papers of Connes and Kriemer for examples of they can be joined, and how they are used to enumerate). It is not correct to think their role is simply to facilitate certain other operations. Rather they facilitate those tasks because of their fundamental role as a data structure.
However, in addition to trees there are also "co-trees"; the trees above all have the property that if you delete the root you delete everything.
Consider iterators on the tree, probably they would be realised as a simple stack of iterators, to a node, and to its parent, ... up to the root.
template<class TREE>
struct node_iterator : std::stack<TREE::iterator>{
operator*() {return *back();}
...};
However, you can have as many as you like; collectively they form a "tree" but where all the arrows flow in the direction toward the root, this co-tree can be iterated through iterators towards the trivial iterator and root; however it cannot be navigated across or down (the other iterators are not known to it) nor can the ensemble of iterators be deleted except by keeping track of all the instances.
Trees are incredibly useful, they have a lot of structure, this makes it a serious challenge to get the definitively correct approach. In my view this is why they are not implemented in the STL. Moreover, in the past, I have seen people get religious and find the idea of a type of container containing instances of its own type challenging - but they have to face it - that is what a tree type represents - it is a node containing a possibly empty collection of (smaller) trees. The current language permits it without challenge providing the default constructor for container<B> does not allocate space on the heap (or anywhere else) for an B, etc.
I for one would be pleased if this did, in a good form, find its way into the standard.
Because the STL is not an "everything" library. It contains, essentially, the minimum structures needed to build things.
This one looks promising and seems to be what you're looking for:
http://tree.phi-sci.com/
IMO, an omission. But I think there is good reason not to include a Tree structure in the STL. There is a lot of logic in maintaining a tree, which is best written as member functions into the base TreeNode object. When TreeNode is wrapped up in an STL header, it just gets messier.
For example:
template <typename T>
struct TreeNode
{
T* DATA ; // data of type T to be stored at this TreeNode
vector< TreeNode<T>* > children ;
// insertion logic for if an insert is asked of me.
// may append to children, or may pass off to one of the child nodes
void insert( T* newData ) ;
} ;
template <typename T>
struct Tree
{
TreeNode<T>* root;
// TREE LEVEL functions
void clear() { delete root ; root=0; }
void insert( T* data ) { if(root)root->insert(data); }
} ;
Reading through the answers here the common named reasons are that one cannot iterate through the tree or that the tree does not assume the similar interface to other STL containers and one could not use STL algorithms with such tree structure.
Having that in mind I tried to design my own tree data structure which will provide STL-like interface and will be usable with existing STL algorthims as much as possible.
My idea was that the tree must be based on the existing STL containers and that it must not hide the container, so that it will be accessible to use with STL algorithms.
The other important feature the tree must provide is the traversing iterators.
Here is what I was able to come up with: https://github.com/cppfw/utki/blob/master/src/utki/tree.hpp
And here are the tests: https://github.com/cppfw/utki/blob/master/tests/unit/src/tree.cpp
All STL containers can be used with iterators. You can't have an iterator an a tree, because you don't have ''one right'' way do go through the tree.

How can one create cyclic (and immutable) data structures in Clojure without extra indirection?

I need to represent directed graphs in Clojure. I'd like to represent each node in the graph as an object (probably a record) that includes a field called :edges that is a collection of the nodes that are directly reachable from the current node. Hopefully it goes without saying, but I would like these graphs to be immutable.
I can construct directed acyclic graphs with this approach as long as I do a topological sort and build each graph "from the leaves up".
This approach doesn't work for cyclic graphs, however. The one workaround I can think of is to have a separate collection (probably a map or vector) of all of the edges for an entire graph. The :edges field in each node would then have the key (or index) into the graph's collection of edges. Adding this extra level of indirection works because I can create keys (or indexes) before the things they (will) refer to exist, but it feels like a kludge. Not only do I need to do an extra lookup whenever I want to visit a neighboring node, but I also have to pass around the global edges collection, which feels very clumsy.
I've heard that some Lisps have a way of creating cyclic lists without resorting to mutation functions. Is there a way to create immutable cyclic data structures in Clojure?
You can wrap each node in a ref to give it a stable handle to point at (and allow you to modify the reference which can start as nil). It is then possible to possible to build cyclic graphs that way. This does have "extra" indirection of course.
I don't think this is a very good idea though. Your second idea is a more common implementation. We built something like this to hold an RDF graph and it is possible to build it out of the core data structures and layer indices over the top of it without too much effort.
I've been playing with this the last few days.
I first tried making each node hold a set of refs to edges, and each edge hold a set of refs to the nodes. I set them equal to each other in a (dosync... (ref-set...)) type of operation. I didn't like this because changing one node requires a large amount of updates, and printing out the graph was a bit tricky. I had to override the print-method multimethod so the repl wouldn't stack overflow. Also any time I wanted to add an edge to an existing node, I had to extract the actual node from the graph first, then do all sorts of edge updates and that sort of thing to make sure everyone was holding on to the most recent version of the other thing. Also, because things were in a ref, determining whether something was connected to something else was a linear-time operation, which seemed inelegant. I didn't get very far before determining that actually performing any useful algorithms with this method would be difficult.
Then I tried another approach which is a variation of the matrix referred to elsewhere. The graph is a clojure map, where the keys are the nodes (not refs to nodes), and the values are another map in which the keys are the neighboring nodes and single value of each key is the edge to that node, represented either as a numerical value indicating the strength of the edge, or an edge structure which I defined elsewhere.
It looks like this, sort of, for 1->2, 1->3, 2->5, 5->2
(def graph {node-1 {node-2 edge12, node-3 edge13},
node-2 {node-5 edge25},
node-3 nil ;;no edge leaves from node 3
node-5 {node-2 edge52}) ;; nodes 2 and 5 have an undirected edge
To access the neighbors of node-1 you go (keys (graph node-1)) or call the function defined elsewhere (neighbors graph node-1), or you can say ((graph node-1) node-2) to get the edge from 1->2.
Several advantages:
Constant time lookup of a node in the graph and of a neighboring node, or return nil if it doesn't exist.
Simple and flexible edge definition. A directed edge exists implicitly when you add a neighbor to a node entry in the map, and its value (or a structure for more information) is provided explicitly, or nil.
You don't have to look up the existing node to do anything to it. It's immutable, so you can define it once before adding it to the graph and then you don't have to chase it around getting the latest version when things change. If a connection in the graph changes, you change the graph structure, not the nodes/edges themselves.
This combines the best features of a matrix representation (the graph topology is in the graph map itself not encoded in the nodes and edges, constant time lookup, and non-mutating nodes and edges), and the adjacency-list (each node "has" a list of its neighboring nodes, space efficient since you don't have any "blanks" like a canonical sparse matrix).
You can have multiples edges between nodes, and if you accidentally define an edge which already exists exactly, the map structure takes care of making sure you are not duplicating it.
Node and edge identity is kept by clojure. I don't have to come up with any sort of indexing scheme or common reference point. The keys and values of the maps are the things they represent, not a lookup elsewhere or ref. Your node structure can be all nils, and as long as it's unique, it can be represented in the graph.
The only big-ish disadvantage I see is that for any given operation (add, remove, any algorithm), you can't just pass it a starting node. You have to pass the whole graph map and a starting node, which is probably a fair price to pay for the simplicity of the whole thing. Another minor disadvantage (or maybe not) is that for an undirected edge you have to define the edge in each direction. This is actually okay because sometimes an edge has a different value for each direction and this scheme allows you to do that.
The only other thing I see here is that because an edge is implicit in the existence of a key-value pair in the map, you cannot define a hyperedge (ie one which connects more than 2 nodes). I don't think this is a big deal necessarily since most graph algorithms I've come across (all?) only deal with an edge that connects 2 nodes.
I ran into this challenge before and concluded that it isn't possible using truly immutable data structures in Clojure at present.
However you may find one or more of the following options acceptable:
Use deftype with ":unsynchronized-mutable" to create a mutable :edges field in each node that you change only once during construction. You can treat it as read-only from then on, with no extra indirection overhead. This approach will probably have the best performance but is a bit of a hack.
Use an atom to implement :edges. There is a bit of extra indirection, but I've personally found reading atoms to be extremely efficient.

Why does the C++ STL not provide any "tree" containers?

Why does the C++ STL not provide any "tree" containers, and what's the best thing to use instead?
I want to store a hierarchy of objects as a tree, rather than use a tree as a performance enhancement...
There are two reasons you could want to use a tree:
You want to mirror the problem using a tree-like structure:
For this we have boost graph library
Or you want a container that has tree like access characteristics
For this we have
std::map (and std::multimap)
std::set (and std::multiset)
Basically the characteristics of these two containers is such that they practically have to be implemented using trees (though this is not actually a requirement).
See also this question:
C tree Implementation
Probably for the same reason that there is no tree container in boost. There are many ways to implement such a container, and there is no good way to satisfy everyone who would use it.
Some issues to consider:
Are the number of children for a node fixed or variable?
How much overhead per node? - ie, do you need parent pointers, sibling pointers, etc.
What algorithms to provide? - different iterators, search algorithms, etc.
In the end, the problem ends up being that a tree container that would be useful enough to everyone, would be too heavyweight to satisfy most of the people using it. If you are looking for something powerful, Boost Graph Library is essentially a superset of what a tree library could be used for.
Here are some other generic tree implementations:
Kasper Peeters' tree.hh
Adobe's forest
core::tree
"I want to store a hierarchy of objects as a tree"
C++11 has come and gone and they still didn't see a need to provide a std::tree, although the idea did come up (see here). Maybe the reason they haven't added this is that it is trivially easy to build your own on top of the existing containers. For example...
template< typename T >
struct tree_node
{
T t;
std::vector<tree_node> children;
};
A simple traversal would use recursion...
template< typename T >
void tree_node<T>::walk_depth_first() const
{
cout<<t;
for ( auto & n: children ) n.walk_depth_first();
}
If you want to maintain a hierarchy and you want it to work with STL algorithms, then things may get complicated. You could build your own iterators and achieve some compatibility, however many of the algorithms simply don't make any sense for a hierarchy (anything that changes the order of a range, for example). Even defining a range within a hierarchy could be a messy business.
The STL's philosophy is that you choose a container based on guarantees and not based on how the container is implemented. For example, your choice of container may be based on a need for fast lookups. For all you care, the container may be implemented as a unidirectional list -- as long as searching is very fast you'd be happy. That's because you're not touching the internals anyhow, you're using iterators or member functions for the access. Your code is not bound to how the container is implemented but to how fast it is, or whether it has a fixed and defined ordering, or whether it is efficient on space, and so on.
If you are looking for a RB-tree implementation, then stl_tree.h might be appropriate for you too.
the std::map is based on a red black tree. You can also use other containers to help you implement your own types of trees.
The problem is that there is no one-size-fits-all solution. Moreover, there is not even a one-size-fits-all interface for a tree. That is, it is not even clear which methods such a tree data structure should provide and it is not even clear what a tree is.
This explains why there is no STL support on this: The STL is for data structures that most people need, where basically everyone agrees on what a sensible interface and an efficient implementation is. For trees, such a thing just doesn't exist.
The gory details
If want to understand further what the problem is, read on. Otherwise, the paragraph above already should be sufficent to answer your question.
I said that there is not even a common interface. You might disagree, since you have one application in mind, but if you think further about it, you will see that there are countless possible operations on trees. You can either have a data structure that enables most of them efficiently, but therefore is more complex overall and has overhead for that complexity, or you have more simple data structure that only allows basic operations but these as quick as possible.
If you want the complete story, check out my paper on the topic. There you will find possible interface, asymptotic complexities on different implementations, and a general description of the problem and also related work with more possible implementations.
What is a tree?
It already starts with what you consider to be a tree:
Rooted or unrooted: most programmers want rooted, most mathematicians want unrooted. (If you wonder what unrooted is: A - B - C is a tree where either A, B, or C could be the root. A rooted tree defines which one is. An unrooted doesn't)
Single root/connected or multi root/disconnected (tree or forest)
Is sibling order relevant? If no, then can the tree structure internally reorder children on updates? If so, iteration order among siblings is no longer defined. But for most trees, sibiling order is actually not meaningful, and allowing the data structure to reorder children on update is very beneficial for some updates.
Really just a tree, or also allow DAG edges (sounds weird, but many people who initially want a tree eventually want a DAG)
Labeled or unlabled? Do you need to store any data per node, or is it only the tree structure you're interested in (the latter can be stored very succinctly)
Query operations
After we have figured out what we define to be a tree, we should define query operations: Basic operations might be "navigate to children, navigate to parent", but there are way more possible operations, e.g.:
Navigate to next/prev sibling: Even most people would consider this a pretty basic operation, it is actually almost impossible if you only have a parent pointer or a children array. So this already shows you that you might need a totally different implementation based on what operations you need.
Navigate in pre/post order
Subtree size: the number of (transitive) descendants of the current node (possibly in O(1) or O(log n), i.e., don't just enumerate them all to count)
the height of the tree in the current node. That is, the longest path from this node to any leave node. Again, in less than O(n).
Given two nodes, find the least common ancestor of the node (with O(1) memory consumption)
How many nodes are between node A and node B in a pre-/post-order traversal? (less than O(n) runtime)
I emphasized that the interesting thing here is whether these methods can be performed better than O(n), because just enumerating the whole tree is always an option. Depending on your application, it might be absolutely crucial that some operations are faster than O(n), or you might not care at all. Again, you will need vastely different data structures depending on your needs here.
Update operations
Until now, I only talked about query opertions. But now to updates. Again, there are various ways in which a tree could be updated. Depending on which you need, you need a more or less sophisticated data structure:
leaf updates (easy): Delete or add a leaf node
inner node updates (harder): Move or delete move an inner node, making its children the children
of its parent
subtree updates (harder): Move or delete a subtree rooted in a node
To just give you some intuition: If you store a child array and your sibling order is important, even deleting a leaf can be O(n) as all siblings behind it have to be shifted in the child array of its parent. If you instead only have a parent pointer, leaf deletion is trivially O(1). If you don't care about sibiling order, it is also O(1) for the child array, as you can simply replace the gap with the last sibling in the array. This is just one example where different data structures will give you quite different update capabilities.
Moving a whole subtree is again trivially O(1) in case of a parent pointer, but can be O(n) if you have a data structure storing all nodes e.g., in pre-order.
Then, there are orthogonal considerations like which iterators stay valid if you perform updates. Some data structures need to invalidate all iterators in the whole tree, even if you insert a new leaf. Others only invalidate iterators in the part of the tree that is altered. Others keep all iterators (except the ones for deleted nodes) valid.
Space considerations
Tree structures can be very succinct. Roughly two bits per node are enough if you need to save on space (e.g., DFUDS or LOUDS, see this explanation to get the gist). But of course, naively, even a parent pointer is already 64 bits. Once you opt for a nicely-navigable structure, you might rather require 20 bytes per node.
With a lot of sophisication, one can also build a data structure that only takes some bits per entry, can be updated efficiently, and still enables all query operations asymptotically fast, but this is a beast of a structure that is highly complex. I once gave a practical course where I had grad students implement this paper. Some of them were able to implement it in 6 weeks (!), others failed. And while the structure has great asymptotics, its complexity makes it have quite some overhead for very simple operations.
Again, no one-size-fits-all.
Conclusion
I worked 5 years on finding the best data structure to represent a tree, and even though I came up with some and there is quite some related work, my conclusion was that there is not one. Depending on the use case, a highly sophsticated data struture will be outperformed by a simple parent pointer. Even defining the interface for a tree is hard. I tried defining one in my paper, but I have to acknowledge that there are various use cases where the interface I defined is too narrow or too large. So I doubt that this will ever end up in STL, as there are just too many tuning knobs.
In a way, std::map is a tree (it is required to have the same performance characteristics as a balanced binary tree) but it doesn't expose other tree functionality. The likely reasoning behind not including a real tree data structure was probably just a matter of not including everything in the stl. The stl can be looked as a framework to use in implementing your own algorithms and data structures.
In general, if there's a basic library functionality that you want, that's not in the stl, the fix is to look at BOOST.
Otherwise, there's a bunch of libraries out there, depending on the needs of your tree.
All STL container are externally represented as "sequences" with one iteration mechanism.
Trees don't follow this idiom.
I think there are several reasons why there are no STL trees. Primarily Trees are a form of recursive data structure which, like a container (list, vector, set), has very different fine structure which makes the correct choices tricky. They are also very easy to construct in basic form using the STL.
A finite rooted tree can be thought of as a container which has a value or payload, say an instance of a class A and, a possibly empty collection of rooted (sub) trees; trees with empty collection of subtrees are thought of as leaves.
template<class A>
struct unordered_tree : std::set<unordered_tree>, A
{};
template<class A>
struct b_tree : std::vector<b_tree>, A
{};
template<class A>
struct planar_tree : std::list<planar_tree>, A
{};
One has to think a little about iterator design etc. and which product and co-product operations one allows to define and be efficient between trees - and the original STL has to be well written - so that the empty set, vector or list container is really empty of any payload in the default case.
Trees play an essential role in many mathematical structures (see the classical papers of Butcher, Grossman and Larsen; also the papers of Connes and Kriemer for examples of they can be joined, and how they are used to enumerate). It is not correct to think their role is simply to facilitate certain other operations. Rather they facilitate those tasks because of their fundamental role as a data structure.
However, in addition to trees there are also "co-trees"; the trees above all have the property that if you delete the root you delete everything.
Consider iterators on the tree, probably they would be realised as a simple stack of iterators, to a node, and to its parent, ... up to the root.
template<class TREE>
struct node_iterator : std::stack<TREE::iterator>{
operator*() {return *back();}
...};
However, you can have as many as you like; collectively they form a "tree" but where all the arrows flow in the direction toward the root, this co-tree can be iterated through iterators towards the trivial iterator and root; however it cannot be navigated across or down (the other iterators are not known to it) nor can the ensemble of iterators be deleted except by keeping track of all the instances.
Trees are incredibly useful, they have a lot of structure, this makes it a serious challenge to get the definitively correct approach. In my view this is why they are not implemented in the STL. Moreover, in the past, I have seen people get religious and find the idea of a type of container containing instances of its own type challenging - but they have to face it - that is what a tree type represents - it is a node containing a possibly empty collection of (smaller) trees. The current language permits it without challenge providing the default constructor for container<B> does not allocate space on the heap (or anywhere else) for an B, etc.
I for one would be pleased if this did, in a good form, find its way into the standard.
Because the STL is not an "everything" library. It contains, essentially, the minimum structures needed to build things.
This one looks promising and seems to be what you're looking for:
http://tree.phi-sci.com/
IMO, an omission. But I think there is good reason not to include a Tree structure in the STL. There is a lot of logic in maintaining a tree, which is best written as member functions into the base TreeNode object. When TreeNode is wrapped up in an STL header, it just gets messier.
For example:
template <typename T>
struct TreeNode
{
T* DATA ; // data of type T to be stored at this TreeNode
vector< TreeNode<T>* > children ;
// insertion logic for if an insert is asked of me.
// may append to children, or may pass off to one of the child nodes
void insert( T* newData ) ;
} ;
template <typename T>
struct Tree
{
TreeNode<T>* root;
// TREE LEVEL functions
void clear() { delete root ; root=0; }
void insert( T* data ) { if(root)root->insert(data); }
} ;
Reading through the answers here the common named reasons are that one cannot iterate through the tree or that the tree does not assume the similar interface to other STL containers and one could not use STL algorithms with such tree structure.
Having that in mind I tried to design my own tree data structure which will provide STL-like interface and will be usable with existing STL algorthims as much as possible.
My idea was that the tree must be based on the existing STL containers and that it must not hide the container, so that it will be accessible to use with STL algorithms.
The other important feature the tree must provide is the traversing iterators.
Here is what I was able to come up with: https://github.com/cppfw/utki/blob/master/src/utki/tree.hpp
And here are the tests: https://github.com/cppfw/utki/blob/master/tests/unit/src/tree.cpp
All STL containers can be used with iterators. You can't have an iterator an a tree, because you don't have ''one right'' way do go through the tree.