matcher/exactphrasepostlist.cc

Go to the documentation of this file.

#include <config.h>

#include "exactphrasepostlist.h"
#include "positionlist.h"
#include "omdebug.h"

#include <algorithm>
#include <vector>

using namespace std;

ExactPhrasePostList::ExactPhrasePostList(PostList *source_,
                                         const vector<PostList*> &terms_)
        : SelectPostList(source_), terms(terms_)
{
    size_t n = terms_.size();
    poslists = new PositionList*[n];
    try {
        order = new unsigned[n];
    } catch (...) {
        delete [] poslists;
        throw;
    }
    for (size_t i = 0; i < n; ++i) order[i] = i;
}

ExactPhrasePostList::~ExactPhrasePostList()
{
    delete [] poslists;
    delete [] order;
}

void
ExactPhrasePostList::start_position_list(unsigned i)
{
    unsigned index = order[i];
    poslists[i] = terms[index]->read_position_list();
    poslists[i]->index = index;
}

class TermCompare {
    vector<PostList *> & terms;

  public:
    TermCompare(vector<PostList *> & terms_) : terms(terms_) { }

    bool operator()(unsigned a, unsigned b) const {
        return terms[a]->get_wdf() < terms[b]->get_wdf();
    }
};

bool
ExactPhrasePostList::test_doc()
{
    DEBUGCALL(MATCH, bool, "ExactPhrasePostList::test_doc", "");

    if (terms.size() <= 1) RETURN(true);

    // We often don't need to read all the position lists, so rather than using
    // the shortest position lists first, we approximate by using the terms
    // with the lowest wdf first.  This will typically give the same or a very
    // similar order.
    sort(order, order + terms.size(), TermCompare(terms));

    // If the first term we check only occurs too close to the start of the
    // document, we only need to read one term's positions.  E.g. search for
    // "ripe mango" when the only occurrence of 'mango' in the current document
    // is at position 0.
    start_position_list(0);
    poslists[0]->skip_to(poslists[0]->index);
    if (poslists[0]->at_end()) RETURN(false);

    // If we get here, we'll need to read the positionlists for at least two
    // terms, so check the true positionlist length for the two terms with the
    // lowest wdf and if necessary swap them so the true shorter one is first.
    start_position_list(1);
    if (poslists[0]->get_size() < poslists[1]->get_size()) {
        poslists[1]->skip_to(poslists[1]->index);
        if (poslists[1]->at_end()) RETURN(false);
        swap(poslists[0], poslists[1]);
    }

    unsigned read_hwm = 1;
    Xapian::termpos idx0 = poslists[0]->index;
    do {
        Xapian::termpos base = poslists[0]->get_position() - idx0;
        unsigned i = 1;
        while (true) {
            if (i > read_hwm) {
                read_hwm = i;
                start_position_list(i);
                // FIXME: consider comparing with poslist[0] and swapping
                // if less common.  Should we allow for the number of positions
                // we've read from poslist[0] already?
            }
            Xapian::termpos required = base + poslists[i]->index;
            poslists[i]->skip_to(required);
            if (poslists[i]->at_end()) RETURN(false);
            if (poslists[i]->get_position() != required) break;
            if (++i == terms.size()) RETURN(true);
        }
        poslists[0]->next();
    } while (!poslists[0]->at_end());
    RETURN(false);
}

Xapian::termcount
ExactPhrasePostList::get_wdf() const
{
    // Calculate the an estimate for the wdf of an exact phrase postlist.
    //
    // We use the minimum wdf of a sub-postlist as our estimate.  See the
    // comment in NearPostList::get_wdf for justification of this estimate.
    //
    // We divide the value calculated for a NearPostList by 3, as a very rough
    // heuristic to represent the fact that the words must occur exactly in
    // order, and phrases are therefore rarer than near matches and (non-exact)
    // phrase matches.

    std::vector<PostList *>::const_iterator i = terms.begin();
    Xapian::termcount wdf = (*i)->get_wdf();
    for (; i != terms.end(); i++) {
        wdf = std::min(wdf, (*i)->get_wdf());
    }
    return wdf;
}

Xapian::doccount
ExactPhrasePostList::get_termfreq_est() const
{
    // It's hard to estimate how many times the exact phrase will occur as
    // it depends a lot on the phrase, but usually the exact phrase will
    // occur significantly less often than the individual terms.
    return source->get_termfreq_est() / 4;
}

string
ExactPhrasePostList::get_description() const
{
    return "(ExactPhrase " + source->get_description() + ")";
}

---