Class TernaryTree
- java.lang.Object
-
- com.aowagie.text.pdf.hyphenation.TernaryTree
-
- All Implemented Interfaces:
java.io.Serializable
,java.lang.Cloneable
- Direct Known Subclasses:
HyphenationTree
class TernaryTree extends java.lang.Object implements java.lang.Cloneable, java.io.Serializable
Ternary Search Tree.
A ternary search tree is a hybrid between a binary tree and a digital search tree (trie). Keys are limited to strings. A data value of type char is stored in each leaf node. It can be used as an index (or pointer) to the data. Branches that only contain one key are compressed to one node by storing a pointer to the trailer substring of the key. This class is intended to serve as base class or helper class to implement Dictionary collections or the like. Ternary trees have some nice properties as the following: the tree can be traversed in sorted order, partial matches (wildcard) can be implemented, retrieval of all keys within a given distance from the target, etc. The storage requirements are higher than a binary tree but a lot less than a trie. Performance is comparable with a hash table, sometimes it outperforms a hash function (most of the time can determine a miss faster than a hash).
The main purpose of this java port is to serve as a base for implementing TeX's hyphenation algorithm (see The TeXBook, appendix H). Each language requires from 5000 to 15000 hyphenation patterns which will be keys in this tree. The strings patterns are usually small (from 2 to 5 characters), but each char in the tree is stored in a node. Thus memory usage is the main concern. We will sacrifice 'elegance' to keep memory requirements to the minimum. Using java's char type as pointer (yes, I know pointer it is a forbidden word in java) we can keep the size of the node to be just 8 bytes (3 pointers and the data char). This gives room for about 65000 nodes. In my tests the English patterns took 7694 nodes and the German patterns 10055 nodes, so I think we are safe.
All said, this is a map with strings as keys and char as value. Pretty limited!. It can be extended to a general map by using the string representation of an object and using the char value as an index to an array that contains the object values.
-
-
Nested Class Summary
Nested Classes Modifier and Type Class Description private class
TernaryTree.Iterator
-
Field Summary
Fields Modifier and Type Field Description private static int
BLOCK_SIZE
protected char[]
eq
Pointer to equal branch and to data when this node is a string terminator.private char
freenode
protected char[]
hi
Pointer to high branch.protected CharVector
kv
This vector holds the trailing of the keys when the branch is compressed.private int
length
protected char[]
lo
Pointer to low branch and to rest of the key when it is stored directly in this node, we don't have unions in java!protected char
root
protected char[]
sc
The character stored in this node: splitchar.private static long
serialVersionUID
We use 4 arrays to represent a node.
-
Constructor Summary
Constructors Constructor Description TernaryTree()
-
Method Summary
All Methods Static Methods Instance Methods Concrete Methods Modifier and Type Method Description private void
balance()
Balance the tree for best search performancejava.lang.Object
clone()
private void
compact(CharVector kx, TernaryTree map, char p)
(package private) int
find(char[] key, int start)
(package private) int
find(java.lang.String key)
private void
init()
(package private) void
insert(char[] key, int start, char val)
private char
insert(char p, char[] key, int start, char val)
The actual insertion function, recursive version.(package private) void
insert(java.lang.String key, char val)
Branches are initially compressed, needing one node per key plus the size of the string key.private void
insertBalanced(java.lang.String[] k, char[] v, int offset, int n)
Recursively insert the median first and then the median of the lower and upper halves, and so on in order to get a balanced tree.void
printStats()
private void
redimNodeArrays(int newsize)
private static int
strcmp(char[] a, int startA, char[] b, int startB)
Compares 2 null terminated char arraysprivate static void
strcpy(char[] dst, int di, char[] src, int si)
private static int
strlen(char[] a)
private static int
strlen(char[] a, int start)
(package private) void
trimToSize()
Each node stores a character (splitchar) which is part of some key(s).
-
-
-
Field Detail
-
serialVersionUID
private static final long serialVersionUID
We use 4 arrays to represent a node. I guess I should have created a proper node class, but somehow Knuth's pascal code made me forget we now have a portable language with virtual memory management and automatic garbage collection! And now is kind of late, furthermore, if it ain't broken, don't fix it.- See Also:
- Constant Field Values
-
lo
protected char[] lo
Pointer to low branch and to rest of the key when it is stored directly in this node, we don't have unions in java!
-
hi
protected char[] hi
Pointer to high branch.
-
eq
protected char[] eq
Pointer to equal branch and to data when this node is a string terminator.
-
sc
protected char[] sc
The character stored in this node: splitchar. Two special values are reserved:
- 0x0000 as string terminator
- 0xFFFF to indicate that the branch starting at this node is compressed
This shouldn't be a problem if we give the usual semantics to strings since 0xFFFF is guaranteed not to be an Unicode character.
-
kv
protected CharVector kv
This vector holds the trailing of the keys when the branch is compressed.
-
root
protected char root
-
freenode
private char freenode
-
length
private int length
-
BLOCK_SIZE
private static final int BLOCK_SIZE
- See Also:
- Constant Field Values
-
-
Method Detail
-
init
private void init()
-
insert
void insert(java.lang.String key, char val)
Branches are initially compressed, needing one node per key plus the size of the string key. They are decompressed as needed when another key with same prefix is inserted. This saves a lot of space, specially for long keys.
-
insert
void insert(char[] key, int start, char val)
-
insert
private char insert(char p, char[] key, int start, char val)
The actual insertion function, recursive version.
-
strcmp
private static int strcmp(char[] a, int startA, char[] b, int startB)
Compares 2 null terminated char arrays
-
strcpy
private static void strcpy(char[] dst, int di, char[] src, int si)
-
strlen
private static int strlen(char[] a, int start)
-
strlen
private static int strlen(char[] a)
-
find
int find(java.lang.String key)
-
find
int find(char[] key, int start)
-
redimNodeArrays
private void redimNodeArrays(int newsize)
-
clone
public java.lang.Object clone()
- Overrides:
clone
in classjava.lang.Object
-
insertBalanced
private void insertBalanced(java.lang.String[] k, char[] v, int offset, int n)
Recursively insert the median first and then the median of the lower and upper halves, and so on in order to get a balanced tree. The array of keys is assumed to be sorted in ascending order.
-
balance
private void balance()
Balance the tree for best search performance
-
trimToSize
void trimToSize()
Each node stores a character (splitchar) which is part of some key(s). In a compressed branch (one that only contain a single string key) the trailer of the key which is not already in nodes is stored externally in the kv array. As items are inserted, key substrings decrease. Some substrings may completely disappear when the whole branch is totally decompressed. The tree is traversed to find the key substrings actually used. In addition, duplicate substrings are removed using a map (implemented with a TernaryTree!).
-
compact
private void compact(CharVector kx, TernaryTree map, char p)
-
printStats
public void printStats()
-
-