Question

I want to modify an recursive function from an ‘ternary search tree’ library (sourceforge & http://code.google.com/p/ternary-search-tree/).
The default behaviour is to search in an ternary search tree for all occurences of strings that match an specified wildcard string.
i.e. having ‘KEY’, ‘KE1’, ‘KE2’ in the tree would find all entrys if I search for ‘KE*’.
But I need the opposite behaviour – search in an ternary search tree (which contains wildcards) all entrys that match an specified string.
i.e. having ‘KE*’, ‘KEY’, ‘K*’ in the tree should find all entrys if I search for ‘KEY’.

A tree/node is defined as following:

typedef struct TstNode {
    TstNode( char c ) : splitChar(c), left(0), right(0), mid(0){
    }
    char splitChar;
    TstTree left, right;
    union {
        TstTree mid;
        int index;
    };
} tstNode;

And the function with the default behaviour:

template <class Object>
void TernarySearchTree<Object>::partialMatchSearch(TstTree tree, const char *key)
{
    if (!tree) return;

    // partial match left
    if (*key == '?' || *key == '*' || *key < tree->splitChar)
    {
        partialMatchSearch( tree->left, key );
    }
    // partial match middle
    if (*key == '?' || *key == '*' || *key == tree->splitChar)
    {
        if ( tree->splitChar && *key )
        {
            if ( *key == '*' )
            {
                partialMatchSearch( tree->mid, key );
            }
            else
            {
                partialMatchSearch( tree->mid, key+1 ); // search next pattern char
            }
        }
    }
    if ( ( *key == 0 ||  *key == '*' ) && tree->splitChar == 0 )
    {
        pmVectorPtr->add( tree->index );
    }

    if (*key == '?' || *key == '*' || *key > tree->splitChar)
    {
        partialMatchSearch( tree->right, key );
    }
}

pmVectorPtr is an Pointer to an Vector of int’s and the function get’s called with the root-element and the searchkey as argument. I already tried to adapt that, but can’t get my head around it yet. My own code:

template <class Object>
void TernarySearchTree<Object>::partialMatchSearchInverted(TstTree tree, const char *key)
{
    if (!tree) return;

    if((tree->splitChar == '*') && ( *key != 0 )){
        partialMatchSearchInverted( tree, key+1 );
    }

    if( *key != 0 ){
        if (*key < tree->splitChar){
            partialMatchSearchInverted( tree->left, key );
        }
        if (*key > tree->splitChar){
            partialMatchSearchInverted( tree->right, key );
        }
    }
    if ((*key == tree->splitChar) || (tree->splitChar == '*')){
        if ( tree->splitChar || *key ){
            partialMatchSearchInverted( tree->mid, key+1 ); // search next pattern char
        }
    }
    if ( ( *key == 0 ) && ( tree->splitChar == 0 ) ){
        pmVectorPtr->add( tree->index );
    }
}

I’ve coded this with extensive use of the debugger and as far as I can tell, it ‘seems’ to work (even if the wildcard is at beginning or mid of a string). BUT if I add to example: ‘K*’ and ‘Ke*’ to the tree, it would find only one solution (in this case Ke*) for ‘Key’. If I remove ‘Ke*’ from the tree, it finds ‘K*’ for an search query of ‘Key’. I still don’t get why.

Any ideas about that?

---

Appendix (my testcase):

#include <iostream>
#include "ternarySearchTree/ternarySearchTree.h"

int main(int argc, char *argv[]){
    TernarySearchTree<string> tst;
    Vector< TstItem<String> > itemVector;
    {
        TstItem<String> item( "Ke*", "Value" );
        itemVector.add( item );
    }
    {
        TstItem<String> item( "K*", "Value" );
        itemVector.add( item );
    }
    {
        TstItem<String> item( "Ka*", "Value" );
        itemVector.add( item );
    }
    tst.buildBalancedTree(itemVector);

    Vector<int> matches = tst.partialMatchSearchInverted("Key");// will only find Ke* (wrong - it should find Ke* and K*), if I remove Ke* above it would find K* (right), if I remove that also it would find nothing (also right)
    for (unsigned j=0;j<matches.count();j++)
    {
        std::cout<<"Matching: "<< tst.getKey( matches[j] ) <<" -  "<< tst.getValue( matches[j] )->c_str()<<std::endl;
    }
    std::cout<<"total matches "<< matches.count()<<std::endl;
    return 0;
}

Answer 1

Ok, I could finally track down my issue: I still hadn’t understand how EXACTLY a ternary tree works at all. Because of that, I didn’t visit the left and right nodes when an wildcard could be somewhere inside these branches (sadly that’s almost everytime). So my final algorithm is MUCH slower than a search in an ternary tree should be at all (I’m afraid the complexity would be probably something like O(n)), but at least I got it to work.

So it seems this data structure isn’t suitable for my needs (searching for wildcard strings) – I would probably get the same speed with an linear list. However, if someone with similiar problems will find this question some day, here’s my code (but I don’t suggest to use it in real applications, cause it’s slow and I guess there are other structures around, that can handle these things much better):

template <class Object>
void TernarySearchTree<Object>::partialMatchSearchInverted(TstTree tree, const char *key)
{
    if (!tree) return;
    if((tree->splitChar == '*') && ( *key != 0 )){
        partialMatchSearchInverted( tree, key+1 ); // wildcard occured, search same node with next key
    }
    if( *key != 0 ){
        if (*key < tree->splitChar){
            partialMatchSearchInverted( tree->left, key );
        }else if('*' < tree->splitChar){ // a wildcard could be in this branch
            partialMatchSearchInverted( tree->left, key );
        }
        if (*key > tree->splitChar){
            partialMatchSearchInverted( tree->right, key );
        }else if('*' > tree->splitChar){ // a wildcard could be here too
            partialMatchSearchInverted( tree->right, key );
        }
    }
    if ((*key == tree->splitChar) || (tree->splitChar == '*')){
        if (*key != 0){
            partialMatchSearchInverted( tree->mid, key+1 ); // search next pattern char
        }
    }
    if ( ( *key == 0 ) && ( tree->splitChar == 0 ) ){
        pmVectorPtr->add( tree->index );
    }
}