[python3] Simplify generated wrapper post-processing

[xapian.git] / xapian-core / api / editdistance.cc

/** @file editdistance.cc
 * @brief Edit distance calculation algorithm.
 *
 *  Based on that described in:
 *
 *  "An extension of Ukkonen's enhanced dynamic programming ASM algorithm"
 *  by Hal Berghel, University of Arkansas
 *  and David Roach, Acxiom Corporation
 *
 *  http://berghel.net/publications/asm/asm.php
 */
/* Copyright (C) 2003 Richard Boulton
 * Copyright (C) 2007,2008,2009,2017 Olly Betts
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
 */

#include <config.h>

#include "editdistance.h"

#include "omassert.h"

#include <algorithm>
#include <cstdlib>
#include <cstring>

using namespace std;

template<class CHR>
struct edist_seq {
    edist_seq(const CHR * ptr_, int len_) : ptr(ptr_), len(len_) { }
    const CHR * ptr;
    int len;
};

template<class CHR>
class edist_state {
    /// Don't allow assignment.
    void operator=(const edist_state &);

    /// Don't allow copying.
    edist_state(const edist_state &);

    edist_seq<CHR> seq1;
    edist_seq<CHR> seq2;

    /* Array of f(k,p) values, where f(k,p) = the largest index i such that
     * d(i,j) = p and d(i,j) is on diagonal k.
     * ie: f(k,p) = largest i s.t. d(i,k+i) = p
     * Where: d(i,j) = edit distance between substrings of length i and j.
     */
    int * fkp;
    int fkp_cols;

    /* Maximum possible edit distance (this is referred to as ZERO_K in
     * the algorithm description by Berghel and Roach). */
    int maxdist;

    int calc_index(int k, int p) const {
        return (k + maxdist) * fkp_cols + p + 1;
    }

  public:

    edist_state(const CHR * ptr1, int len1, const CHR * ptr2, int len2);

    ~edist_state();

    int get_f_kp(int k, int p) const {
        return fkp[calc_index(k, p)];
    }

    void set_f_kp(int k, int p, int val) {
        fkp[calc_index(k, p)] = val;
    }

    bool is_transposed(int pos1, int pos2) const {
        if (pos1 <= 0 || pos2 <= 0 || pos1 >= seq1.len || pos2 >= seq2.len) return false;
        return (seq1.ptr[pos1 - 1] == seq2.ptr[pos2] &&
                seq1.ptr[pos1] == seq2.ptr[pos2 - 1]);
    }

    void edist_calc_f_kp(int k, int p);
};

template<class CHR>
void edist_state<CHR>::edist_calc_f_kp(int k, int p)
{
    int maxlen = get_f_kp(k, p - 1) + 1; /* dist if do substitute */
    int maxlen2 = get_f_kp(k - 1, p - 1); /* dist if do insert */
    int maxlen3 = get_f_kp(k + 1, p - 1) + 1; /* dist if delete */

    if (is_transposed(maxlen, maxlen + k)) {
        // Transposition.
        ++maxlen;
    }

    if (maxlen >= maxlen2) {
        if (maxlen >= maxlen3) {
            // Transposition or Substitution.
        } else {
            // Deletion.
            maxlen = maxlen3;
        }
    } else {
        if (maxlen2 >= maxlen3) {
            // Insertion.
            maxlen = maxlen2;
        } else {
            // Deletion.
            maxlen = maxlen3;
        }
    }

    /* Check for exact matches, and increase the length until we don't have
     * one. */
    while (maxlen < seq1.len &&
           maxlen + k < seq2.len &&
           seq1.ptr[maxlen] == seq2.ptr[maxlen + k]) {
        ++maxlen;
    }
    set_f_kp(k, p, maxlen);
}

#define INF 1000000
template<class CHR>
edist_state<CHR>::edist_state(const CHR * ptr1, int len1,
                              const CHR * ptr2, int len2)
    : seq1(ptr1, len1), seq2(ptr2, len2), maxdist(len2)
{
    Assert(len2 >= len1);
    /* Each row represents a value of k, from -maxdist to maxdist. */
    int fkp_rows = maxdist * 2 + 1;
    /* Each column represents a value of p, from -1 to maxdist. */
    fkp_cols = maxdist + 2;
    /* fkp is stored as a rectangular array, row by row. */
    fkp = new int[fkp_rows * fkp_cols];

    for (int k = -maxdist; k <= maxdist; ++k) {
        for (int p = -1; p <= maxdist; ++p) {
            if (p == abs(k) - 1) {
                if (k < 0) {
                    set_f_kp(k, p, abs(k) - 1);
                } else {
                    set_f_kp(k, p, -1);
                }
            } else if (p < abs(k)) {
                set_f_kp(k, p, -INF);
            }
        }
    }
}

template<class CHR>
edist_state<CHR>::~edist_state() {
    delete [] fkp;
}

template<class CHR>
static int
seqcmp_editdist(const CHR * ptr1, int len1, const CHR * ptr2, int len2,
                int max_distance)
{
    int lendiff = len2 - len1;
    /* Make sure second sequence is longer (or same length). */
    if (lendiff < 0) {
        lendiff = -lendiff;
        swap(ptr1, ptr2);
        swap(len1, len2);
    }

    /* Special case for if one or both sequences are empty. */
    if (len1 == 0) return len2;

    edist_state<CHR> state(ptr1, len1, ptr2, len2);

    int p = lendiff; /* This is the minimum possible edit distance. */
    while (p <= max_distance) {
        for (int temp_p = 0; temp_p != p; ++temp_p) {
            int inc = p - temp_p;
            if (abs(lendiff - inc) <= temp_p) {
                state.edist_calc_f_kp(lendiff - inc, temp_p);
            }
            if (abs(lendiff + inc) <= temp_p) {
                state.edist_calc_f_kp(lendiff + inc, temp_p);
            }
        }
        state.edist_calc_f_kp(lendiff, p);

        if (state.get_f_kp(lendiff, p) == len1) break;
        ++p;
    }

    return p;
}

int
edit_distance_unsigned(const unsigned * ptr1, int len1,
                       const unsigned * ptr2, int len2,
                       int max_distance)
{
    return seqcmp_editdist<unsigned>(ptr1, len1, ptr2, len2, max_distance);
}

// We sum the character frequency histogram absolute differences to compute a
// lower bound on the edit distance.  Rather than counting each Unicode code
// point uniquely, we use an array with VEC_SIZE elements and tally code points
// modulo VEC_SIZE which can only reduce the bound we calculate.
//
// There will be a trade-off between how good the bound is and how large and
// array is used (a larger array takes more time to clear and sum over).  The
// value 64 is somewhat arbitrary - it works as well as 128 for the testsuite
// but that may not reflect real world performance.  FIXME: profile and tune.

#define VEC_SIZE 64

int
freq_edit_lower_bound(const vector<unsigned> & a, const vector<unsigned> & b)
{
    int vec[VEC_SIZE];
    memset(vec, 0, sizeof(vec));
    vector<unsigned>::const_iterator i;
    for (i = a.begin(); i != a.end(); ++i) {
        ++vec[(*i) % VEC_SIZE];
    }
    for (i = b.begin(); i != b.end(); ++i) {
        --vec[(*i) % VEC_SIZE];
    }
    unsigned int total = 0;
    for (size_t j = 0; j < VEC_SIZE; ++j) {
        total += abs(vec[j]);
    }
    // Each insertion or deletion adds at most 1 to total.  Each transposition
    // doesn't change it at all.  But each substitution can change it by 2 so
    // we need to divide it by 2.  Rounding up is OK, since the odd change must
    // be due to an actual edit.
    return (total + 1) / 2;
}